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.0002149 | Rapidly Boosted Plasma IL-5 Induced by Treatment of Human Schistosomiasis haematobium Is Dependent on Antigen Dose, IgE and Eosinophils | IgE specific to worm antigen (SWA) and pre-treatment eosinophil number, are associated with human immunity to re-infection with schistosomes after chemotherapeutic treatment. Treatment significantly elevates circulating IL-5 24-hr post-treatment of Schistosoma mansoni. Here we investigate if praziquantel treatment of human schistosomiasis haematobium also boosts circulating IL-5, the immunological and parasitological factors that predispose to this, and the relationship between these and subsequent immunity to post-treatment re-infection.
The relationship between pre-treatment SWA-IgE, eosinophil number and infection intensity and the 24-hr post-treatment IL-5 boost was investigated in a Malian cohort (aged 5–40 yrs), exposed to S. haematobium. Eotaxin levels were measured at 24-hr post-treatment as a proxy of eosinophil migration. The relationship between the 24-hr post-treatment IL-5 boost and later eosinophil numbers and SWA-IgE levels (9-wk post-treatment) was examined, then investigated in the context of subsequent levels of re-infection (2-yr post-treatment). Circulating IL-5 levels increased 24-hr post-treatment and were associated with pre-treatment infection intensity, SWA-IgE levels, eosinophil number, as well as 24-hr post-treatment eotaxin levels. 24-hr IL-5 levels were, in turn, significantly associated with eosinophil number and elevated SWA-IgE 9-wk later. These SWA-IgE levels were significantly associated with immunity to re-infection.
Early IL-5 production after treatment-induced exposure to S. haematobium worm antigen is positively associated with antigen dose (infection intensity), IgE availability for arming of effector cells at time of treatment and subsequent eosinophil migration response (as indicated by eotaxin levels). The IL-5 produced is positively associated with increased downstream eosinophil number and increases in specific IgE levels, implicating this cytokine boost and its down-stream consequences in the production and maintenance of IgE, and subsequent re-infection immunity.
| Partial human immunity to infection with trematode worms of the genus Schistosoma is associated with IgE specific to adult worm-derived antigens and eosinophils. Treatment studies of Schistosoma infection allow us to examine the temporal features of the immune response post-antigen exposure, their inter-dependence and their relationship with re-infection levels. Here the boosted levels of the cytokine IL-5, measured at 24-hrs post-treatment of a Malian cohort, aged 5–40 yrs, were found to be significantly associated with pre-treatment levels of IgE to worm-derived antigens and eosinophil number, linking this rapid response to two of the main correlates of human immunity to these parasites. The IL-5 levels at 24-hr were in turn related to increased eosinophil counts and SWA-IgE levels at 9-wks post-treatment. In line with previous studies SWA-IgE was associated with resistance to re-infection. The study therefore identifies temporal relationships between immune mediators prior to and post treatment induced antigen exposure that are associated with resistance to re-infection.
| Schistosoma haematobium infection is the most prevalent form of human schistosomiasis and a major public health problem. Schistosomiasis haematobium is associated with bladder calcification, hydronephrosis, kidney failure, lesions of the genital tract and bladder carcinoma. S. haematobium infection intensities follow a similar age distribution to those of other human schistosome infections – rising throughout childhood, peaking in early adolescence, before declining sharply in late adolescence and early adulthood. Epidemiological studies of S. mansoni infection in fishing communities, where exposure to infection is greatest in adult males, show that this characteristic age-infection intensity curve is still present, and thus maintained by factors other than exposure to infection [1]. This implies that in endemic communities, a partial immunity to re-infection slowly develops, and that this significantly contributes to the decline in infection intensities seen in older members of schistosomiasis endemic communities.
Immuno-epidemiological studies that use a treatment/re-infection study design - in which immune mediators are measured, a cohort treated and their re-infection levels determined 1 to 2-yr later - have shown IgE levels, specific for schistosome adult worm derived antigens (SWA), to be a major correlate of immunity to infection with each of the 3 schistosome species that cause most human schistosomiasis [2], [3], [4]. Pre-treatment circulating eosinophil number is also associated with immunity to re-infection with both S. mansoni [5], [6] and S. haematobium [7]. In schistosomiasis endemic areas, human SWA-specific IgE levels increase with age [8], [9], [10], and in response to praziquantel treatment [3], [11], [12], [13]. Adult schistosome worms live for many years in the human blood stream, but treatment disrupts the integrity of the worm's outer tegument exposing a range of formerly cryptic antigens to the host's immune system [14]. The post-treatment IgE levels induced by otherwise cryptic antigens can be more strongly associated with re-infection immunity than pre-treatment IgE levels [3], [13], and multiple rounds of treatment is found to increase resistance [15]. Eosinophil number also alters dramatically post-treatment, peaking between 2 and 4-wk post-treatment [16], [17].
High levels of IL-5 are produced by PBMC and whole blood cell cultures stimulated in vitro with SWA [18], [19], [20], [21] and pre-treatment in vitro SWA specific IL-5 responses are associated with post-treatment SWA-IgE levels [22], suggesting that IL-5 responsiveness is a key component of human immunity to schistosomiasis. In adult fishermen infected with S. mansoni, plasma IL-5 is elevated 24-hr post treatment [17]. Human IL-5 production is therefore dynamic in response to antigen exposed by chemotherapeutic treatment for this parasite. Circulating eosinophil number declines concurrently with the increased plasma IL-5 levels 24-hour post-treatment, suggesting that these cells migrate to tissues responding to antigen released by treatment, and that they may play a role in the production of the elevated IL-5 [17].
We conducted a re-infection study in a S. haematobium endemic Malian population, to determine whether an equivalent to the boost in IL-5 that occurs after treatment of S. mansoni occurs after treatment for S. haematobium infections, and to model any such IL-5 boost against known key components of immunity to schistosomiasis, SWA-IgE and eosinophils. The interdependence of IL-5, SWA-IgE and eosinophils was then modeled in relation to subsequent immunity to re-infection by the parasite. IL-5 levels were boosted 24-hr post-treatment and found to be dependent on pre-treatment SWA-IgE levels, eosinophil count and infection intensity. IL-5 levels 24-hr post-treatment were also significantly associated with increased SWA-IgE levels 9-wk post-treatment, which in turn were associated with partial immunity to re-infection.
The study received ethical approval from the Ethical Review Committee of the National Institute for Research in Public Health, Mali. Informed oral consent was given by the chief of the villages and adults during village meetings. Informed oral consent was given on an individual basis by adult participants and for children by their parents or guardians during recruitment. Consent was recorded by the recruitment team. Due to cultural reasons and low literacy rates in villages, oral consent is deemed acceptable by the Malian Ministry of Health and was approved by the Ethical Review Committee of the National Institute for Research in Public Health.
The study took place in 3 villages, Segou Region, Mali. Two villages, Kaladangan and Guenidaga, are fishing settlements on the banks of the River Niger. The third village, Kalabougou, is on a tributary of the main river and occupation is more diverse, consisting of fishing, farming and pottery. The cohort of 326 individuals, 186 females and 140 males, aged 5 to 40 yrs, were selected from the village populations by generation of random numbers, with the sole selection criteria being aged between 5 and 40 years of age. A number of school-aged children were excluded from the selection as they had previously been treated by the Schistosome Control Initiative. The cohort were part of a multi-disciplinary study to assess the impact of one v two treatment and were assigned by random selection, to receive either a single dose of praziquantel, or a second dose 2-wk after the first. At 9-wk post-treatment 245 individuals participated and 167 who participated in all previous time-points were followed-up 2-yr later. Three urine samples were collected from each individual pre-treatment and 10 ml of each was filtered for S. haematobium egg counts. Three urine samples were collected 9-wk post-treatment to assess efficacy of treatment, and three samples were collected 2-yr post-treatment to assess re-infection intensities. Pre-treatment stool samples were examined for S. mansoni and gut nematode infections by the Kato Katz method.
Five ml of blood were collected by venipuncture into EDTA, pre-treatment and 9-wk post-treatment. A 200 µl aliquot of whole blood was removed for haematology analysis and preparation of May-Grünswald stained blood smears for eosinophil differential counts. Eosinophil number/ml of blood was calculated from the differentials and total white blood cell counts. After centrifugation of remaining blood samples the plasma was harvested. Finger-prick blood samples, collected into EDTA, were taken pre-treatment and 24-hr post-treatment, centrifuged and the plasma harvested. Plasma samples were stored at −20°C prior to shipment on dry ice. In Cambridge, venous plasma samples were treated with 0.3% tributyl phosphate/1% Tween 80 (both Sigma, Poole, UK) to inactivate encapsulated viruses. Samples were stored at −80°C prior to analysis.
SWA-IgE levels and SWA-IgG4 levels in venous plasma samples were measured in duplicate by ELISA. 384-well plates were coated with 8 µg/ml SWA antigen. For IgE, plasma was diluted 1∶20 and SWA-IgE levels detected using anti-human IgE antibody clone G7-26 (BD Pharmingen, San Diego, CA). For IgG4, plasma was diluted 1∶200 and SWA-IgG4 levels detected using anti-human IgG4 antibody clone G17-4 (BD Pharmingen). Assays were developed using OPD substrate (Sigma). SWA-IgE concentrations were extrapolated from standard curves based on a pool of SWA-IgE positive sera from previous studies that had been quantified by the ImmunoCAP assay (Phadia, Uppsala, Sweden). SWA-IgG4 concentrations were extrapolated from purified human IgG4 myeloma (Sigma) derived standard curves.
IL-4, IL-5, IL-13 and eotaxin levels were measured in finger-prick plasma samples by Luminex bead array. Beads were coupled with capture monoclonal Ab (IL-4, IL-5, and IL-13, BD Pharmingen; Eotaxin, R&D Systems, Minneapolis, MN), incubated with 12.5 µl plasma, diluted 1∶8, overnight at 4°C, and levels detected using monoclonal Ab for IL-4, IL-5 and IL-13 (BD Pharmingen), and poly-clonal goat anti-human eotaxin (R&D Systems).
S. haematobium infection intensities, SWA-IgE, eosinophil numbers and plasma IL-5, IL-13 and eotaxin levels were log-transformed prior to statistical modeling. Age-profiles of some variables modeled are not linear, so age was divided into an ordinal variable; in models of IL-5 at 24-hr and SWA-IgE and eosinophils at 9-wk post-treatment, age-groups were: 5–6 yr (n = 63), 7–9 yr (n = 59), 10–14 yr (n = 53), 15–23 yr (n = 48), 24–34 yr (n = 65), and 35–40 yr (n = 38). The number of young adults who participated 2-yr post-treatment was insufficient to maintain these age-groups for re-infection models, so age-groups were collapsed to pre-treatment ages of 5–6-yr (n = 34), 7–9 yr (n = 32), 10–14 yr (n = 21), 15–30 yr (n = 33) and 31–40 yr (n = 32). Students t-test of log-transformed variables were used to compare means of two groups, and paired t-tests to compare means of log-transformed longitudinally matched data. Prevalence was compared using Chi-squared tests.
Linear regression models were constructed, except for models of re-infection 2-yr post-treatment, for which logistic regression models were constructed. Village of residence, sex and age group were added a priori into initial models. Models were reduced using backwards stepwise regression via the step command which applies Akeike's information criterion (AIC) to test all variables for their requirement in achieving maximum fit of the model. Further removal of variables with a p-value >0.05 was carried out, least significant first; ANOVA was used to compare the final reduced model with that returned by the step function to confirm that removal of these variables did not significantly diminish the fit of the model. The exception to reduction by backward regression was when age-group was removed in models of re-infection to detect any masking of potential immune correlates by this confounder.
The geometric mean age, sex ratios, village of residence and geometric mean pre-treatment S. haematobium infection intensity of the initial cohort, and those successfully followed up at 9-wks and 2-yrs post-infection, are shown in table 1. The fishing villages of Kaladangan and Guenidaga had a greater loss to follow-up than Kalabougou (χ2 = 38.33, p<0.001). All other variables remained similar amongst the initial cohort and those successfully followed up, including the immune parameters that were used in further analysis (data not shown).
Pre-treatment S. haematobium infection intensities were high, even in the youngest age-group, and peaked at 7–9 yrs of age before characteristically declining in older children and adults. There was no difference in infection intensities between the sexes (males: geometric mean = 17.02 eggs/10 ml (95% CI: 12.17, 23.81); females: geometric mean = 12.55 eggs/10 ml (95% CI: 9.42 16.73), t = −1.36, p = 0.174). There was a village effect on infection. The prevalence in Kalabougou was 74.3%, with a geometric mean infection intensity of 8.03eggs/10 ml (95% CI: 6.13, 10.52). In Guenidaga and Kaladangan combined, the prevalence of S. haematobium infection was 91.2%, significantly higher than that in Kalabougou (χ2 = 14.35, p<0.001), and the geometric mean infection intensity was also significantly higher (t = −5.057, p<0.001) at 28.89 eggs/10 ml (95% CI: 20.95, 39.83). Treatment successfully cleared all S. haematobium infections, except for 30, 6 from Kalabougou, and 24 from Kaladangan and Guenidaga. Infection intensities at 9-wk for these 30 individuals were greatly reduced (geometric mean = 3.89 (95% CI: 2.57, 5.88)). There was no significant difference in pre-treatment prevalence (χ2 = 0.6791, p = 0.410), infection intensity (t = 0.4391, p = 0.661) nor in treatment efficacy (χ2 = 0.132, p = 0.716) or infection intensities of those whose S. haematobium infection was not cleared (t = −1.225, p = 0.233) between the two praziquantel regimes. Nor were there significant differences between the two regimes in any of the immune parameters analysed below (data not shown). Only 35 of the 252 individuals who provided stool samples had detectable S. mansoni eggs (P = 13.9%), 33 of whom were from Kaladangan or Guenidaga, and for 3 of whom treatment did not clear infection. No gut nematode eggs were detected.
The geometric mean IL-5 level pre-treatment was 9.54 pg/ml (95% CI: 8.80, 10.36). At 24-hr post-treatment the geometric mean IL-5 level was 38.62 pg/ml (95% CI: 33.04, 45.14). This boost in IL-5 levels was significant (t = −19.37, p<0.001) and was in all age-groups (fig. 1). The profile of the IL-5 boost with age (fig. 1) was similar to that of infection intensity with age, however, the peak in the boost occurred slightly later (10 to 14-yr for IL-5 levels, 7 to 9-yr for infection intensity). There was no significant difference in IL-4 levels pre-treatment and 24-hr post-treatment, and they were very low at both time-points (pre-treatment geometric mean = 3.17 pg/ml (95% CI: 2.98, 3.36); 24-hr post-treatment geometric mean = 3.75 pg/ml (95%CI: 3.48, 4.04), t = −0.95, p = 0.340). The pre-treatment geometric mean IL-13 level was 7.04 pg/ml (95% CI: 5.91, 8.37), and was significantly boosted 24 hr post-treatment (t = −5.01, p<0.001), when the geometric mean levels were 11.87 pg/ml (95%CI: 9.90, 14.24). Regression analysis of 24-hr IL-13 levels, controlling for pre-treatment IL-13 levels, showed that the only significant trend with age was due to the failure of the 15–23 yr olds to boost IL-13 24 hr post-treatment (fig. 1, β = −0.73, S.E. = 0.32, p = 0.016).
Linear regression models of 24-hr post-treatment plasma IL-5 levels were constructed, controlling for pre-treatment IL-5. Infection intensity, and the known correlates of immunity, SWA-IgE levels and eosinophil number, were measured pre-treatment, and the 24-hr levels of the eosinophil attractant, eotaxin, was measured as an indicator of its release in response to treatment. These parameters and the a priori variables of village group, sex and age-group were entered into the model. In the initial model neither age-group nor S. haematobium infection intensity were significant predictors of the boost in IL-5 (table 2). After the removal of age-group from the model during backwards stepwise regression, S. haematobium infection intensities were significantly associated with IL-5 levels at 24-hr post-treatment (table 2; reduced model). Pre-treatment levels of SWA-IgE and eosinophil numbers, and the plasma eotaxin levels at 24-hr post-treatment were also significant predictors of the boost in IL-5 (table 2). The model was not improved by the addition of detectable S. mansoni infection (data not shown).
Pre-treatment circulating eosinophil number was 4.53×105 cells/ml (95% CI: 4.11×105, 4.98×105). Nine-wk post-treatment the geometric mean number of circulating eosinophils had increased to 5.27×105 cells/ml (95% CI: 4.81×105, 5.76×105). This increase in circulating eosinophil number was significant (t = −3.11, p<0.001). Models of eosinophil number at 9-wk post-treatment were constructed, controlling for pre-treatment eosinophil number. The same variables as those modeled for the IL-5 boost: i.e. sex, village-group and age-group, pre-treatment infection intensities, SWA-IgE levels and 24-hr post-treatment eotaxin, were entered, along with IL-5 levels 24-hr post-treatment. IL-5 levels 24-hr post-treatment were significantly associated with increased eosinophil number at 9-wk post-treatment (table 3), as were pre-treatment infection intensities. Pre-treatment SWA-IgE levels, sex, age-group, village-group and eotaxin levels at 24-hr post-treatment, did not contribute to the model and were removed (table 3, reduced model). This did not significantly diminish the fit of the model.
SWA-IgE levels pre-treatment were 25.07 ng/ml (95% CI: 20.87, 30.11) and significantly increased in the cohort as a whole at 9-wk post-treatment (t = −3.9982, p<0.001) when the geometric mean level was 32.12 ng/ml (95% CI: 27.07, 38.12). Stratifying the data by age-group, a negligible increase was observed in the younger age-groups, but the increase was prominent in the oldest age-groups (fig. 2). This appears to be due to a pre-treatment decrease in SWA-IgE in these oldest age-groups compared to the 15–23 yr olds, that was no longer apparent post-treatment, when SWA-IgE reached a plateau from 15–25 yrs of age.
Linear regression models of 9-wk post-treatment SWA-IgE, controlled for pre-treatment levels, were constructed. The full model indicated that age-group and 24-hr post-treatment IL-5 levels were significant predictors of 9-wk post-treatment SWA-IgE. None of the other predictors – sex, village-group, pre-treatment infection intensities, eosinophil number and 24-hr post-treatment eotaxin - significantly contributed to the model (table 4), and their removal did not significantly diminish the fit of the model.
When 24-hr post-treatment IL-13 levels replaced 24-hr post-treatment IL-5 levels in the model, IL-13 was not found to be significantly associated with 9-wk post-treatment SWA-IgE levels. Nor did the addition of IL-13 to the model containing 24-hr post-treatment IL-5 levels, alter the significance of the relationship between IL-5 and SWA-IgE (data not shown). Due to the very low levels measured, and the failure to detect a boost in IL-4 at 24-hr post-treatment, this cytokine was not analysed.
SWA-IgG4 levels increased from a pre-treatment geometric mean of 20.74 µg/ml (95% CI: 16.00, 26.88) to a 9-wk post-treatment geometric mean of 43.23 µg/ml (95% CI: 32.06, 58.31). This increase was significant (t = −8.467, p<0.001). In linear regression models of 9-wk post-treatment SWA-IgG4, controlling for pre-treatment levels, sex, village group, age-group, pre-treatment SWA-IgE, pre-treatment eosinophil number, 24-hr eotaxin and 24-hr IL-5 were not significant (data not shown).
Of the 152 individuals followed-up at 2-yr post-treatment, who also had no detectable eggs at 9-wk post-treatment, 99 (65.13%) had been re-infected. A logistic regression model of yes/no re-infected was constructed. Individuals who were egg positive at 9-wk post-treatment were not included in this model. Similar to pre-treatment infection, there was a village effect on re-infection, and older age-groups had significantly reduced odds of being re-infected than the youngest age-group. Neither IL-5 levels 24-hr post-treatment nor SWA-IgE levels 9-wk post-treatment were significant predictors of re-infection status (table 5, model 1). As age is closely related to SWA-IgE levels, and is a known confounder of the relationship between post-treatment SWA-IgE levels and re-infection, the model was re-fitted with age-group removed. When age-group was removed from the model, SWA-IgE levels 9-wk post-treatment were negatively associated with re-infection, but IL-5 levels 24-hr post-treatment were still not significant (table 5, model 2). Eosinophil count at 9-wk post-treatment was positively associated with re-infection in this second model. SWA-IgG4 was added to these models as SWA-IgG4 has previously been reported to be associated with susceptibility to re-infection. It was not significantly associated with re-infection status in either the model with age-group or the model without age-group (table 5). The model that included age-group had a better fit (AIC = 134.87), than the model that did not include age-group (AIC = 171.17).
High levels of IgE are characteristic of helminth infections and of allergy. For some helminth infections, such as schistosomiasis, high levels of circulating parasite-specific IgE are associated with partial immunity to re-infection after chemotherapeutic cure. Control of IgE production is relatively poorly understood, with little known about early cellular and cytokine responses after antigen/allergen exposure and how they relate to later IgE production. Increased IgE levels to worm derived Ag after treatment, in combination with the re-infection study design used in the current study, allows the examination of the early post-treatment responses and their relationship with the later protective IgE response.
Previously, in a small cohort of S. mansoni infected Ugandan fishermen, we showed that plasma IL-5 levels are boosted 24-hr post-treatment [17]. In the current study, conducted in an area where high transmission of S. haematobium occurs, praziquantel-induced death of S. haematobium adult worms also resulted in boosted levels of plasma IL-5. S. haematobium adult worms live in the bladder plexus, unlike S. mansoni worms, which reside in the intestinal mesenteric veins. The treatment induced circulating IL-5 boost is therefore neither schistosome species-specific, nor dependent on site of worm death. A similar boost in circulating IL-5 has been observed in seasonal allergic rhinitis patients 24-hr after nasal allergen challenge. In that study the increase in plasma IL-5 significantly correlated with concurrent increases in eosinophil cationic protein in sputum [23]. In S. mansoni infected fishermen a substantial fall in the numbers of circulating eosinophil, suggesting a rapid migration from the circulation into the tissues, coincided with the 24-hr post-treatment increase in circulating IL-5 [17]. In the current study, data on 24-hr post-treatment eosinophil numbers was not available. However, when the boost in plasma IL-5 at 24-hr post-treatment was modeled, pre-treatment eosinophil number and 24-hr post-treatment levels of plasma eotaxin, a chemokine involved in attraction and migration of eosinophils into tissue [24], were significant. The 24-hr eotaxin levels are likely to indicate the immediate release of this chemokine in response to treatment.
The significance of pre-treatment eosinophil count in the production of IL-5 does not directly implicate them as the source of the IL-5, although they do pre-package IL-5 in their granules [25]. Eosinophils can skew and maintain immune responses via selective release of cytokines [26], [27] but they can also up-regulate co-stimulatory molecules and present antigen [28], [29]. The eosinophils could therefore be responding to early signals to migrate to sites of inflammation, where they drive IL-5 production by T cells. 24-hr post-treatment whole blood cultures release less IL-5 in response to SWA-stimulation than pre-treatment cultures, a responsiveness that is recovered 3-weeks post treatment [17]. A similar inverse relationship between decreased PBMC IL-5 responsiveness and increased plasma IL-5 is also seen early after treatment for lymphatic filariasis [30], suggesting that the lymphocytes capable of producing IL-5 have left the circulation.
As measured in the current study, 24-hr post-treatment plasma IL-5 is part of the in vivo response to challenge as the adult worms' integrity is disrupted in the blood stream, exposing normally cryptic antigen [14]. The association between S. haematobium infection intensities, after the removal of the confounder age, and the boost in plasma IL-5 may indicate that the magnitude of the boost in plasma IL-5 levels is dependent on the dose of antigens that the individual is exposed to upon treatment. The plasma IL-5 boost was also SWA-IgE dependent. The dependence on SWA-IgE, a response that increases with age in populations living in schistosomiasis endemic areas [8], [9], [10], may explain why the peak boost in IL-5 occurred in a slightly older age-group than the peak in infection intensities.
Higher circulating SWA-IgE is likely a proxy of increased SWA-specific arming of IgE-effector cells, such as mast cells, which are present in both the gut mucosa and the bladder wall. Mast cells initiate eosinophil migration to the tissue after treatment for Onchocerca volvulus, with elevated levels of plasma tryptase, indicative of mast cell degranulation, preceding a decrease in circulating and an increase in skin eosinophil numbers [31]. We have previously observed a small elevation in mast-cell tryptase levels in the plasma at 24-hr post-treatment for S. mansoni [32], indicating that a similar mechanism may occur after treatment for schistosomiasis.
IL-5 induces eosinophil maturation and release from the bone marrow [33] and IL-5 levels at 24-hr post-treatment were associated with an increase in eosinophil number at 9-wk post-treatment. A similar burst in IL-5 24–48-hr after treatment, followed by an increase in circulating eosinophil number at 3-days to 2-wk post-treatment, has been seen for the micro-filarial nematodes O. volvulus and Wucheria bancrofti. For both infections, the time to peak IL-5 levels was significantly associated with the time to peak eosinophil number [30], [34], [35]. As the treatment of schistosomiasis has been shown to induce an increase in eosinophil number, which peaks between 2 and 4-wk post-treatment [16], [17], those individuals who still have elevated circulating eosinophils may have a delayed response. The positive relationship between eosinophil number at 9-wk and re-infection 2-yr later, one that appears surprising in the context of the literature showing pre-treatment eosinophil numbers to be associated with resistance [5], [6], could therefore be due to the study time-point.
Pre-treatment in vitro IL-5 responses specific to SWA are associated with protective SWA-IgE responses in S. mansoni infection [22]. Here, plasma IL-5 levels 24-hr post-treatment was the only variable, other than age, that was associated with the increase in SWA-IgE levels at 9-wk post-treatment, so the in vivo, as well as in vitro, IL-5 response is associated with the production of protective SWA-IgE. Class switching of B cells to IgE production is a process predominantly under the control of other Th2 cytokines, IL-4 and IL-13 [36], [37], [38]. As very low levels of IL-4 were measured and no boost was detected, and IL-13 levels at 24-hr post-treatment were not significantly associated with SWA-IgE, the role of these cytokines is likely to be downstream of the IL-5 boost. The observed relationship between 24-hr post-treatment plasma IL-5 levels and 9-wk SWA-IgE could be an indirect relationship, with the IL-5 boost being a proxy of increased Th2 reactivity. An increase in Th2 responsiveness does occur with treatment of schistosomiasis with greater IL-4 and IL-13 in vitro responses to SWA being measured 7-wk, and beyond, post-treatment [20], [39].
SWA-IgE at 9-wk post-treatment was negatively associated with the re-infection that took place over the next 2-yr, but only after the removal of age. SWA-IgE levels at 9-wk post-treatment increased to a plateau with age (but not the pre-treatment levels which declined in the oldest age-groups); due to the close relationship of post-treatment SWA-IgE levels with age, it was not possible to distinguish their relative contribution to re-infection. The improved model fit when age was included indicates that SWA-IgE levels, although likely to be one of the key mediators, are not the only age-related correlate of immunity. However, neither the 24-hr post-treatment IL-5 levels nor the SWA-IgG4 levels at 9-wk post-treatment, an antibody isotype reported to be associated with susceptibility to re-infection [2], [40], were significantly associated with re-infection in this study. Other age-related immune correlates not measured in the present study, could include an increase in CD23 expression by B cells [41]. The expression of CD23, the low affinity IgE-receptor, has been hypothesized to increase specific antigen capture and subsequent presentation to B cells, augmenting IgE production in schistosomiasis [42].
Here we have shown a boost in plasma IL-5 levels 24-hr after human S. haematobium infections are treated with praziquantel, and that this is dependent on both the intensity of the pre-treatment infection (or antigen dose the individual is exposed to by treatment) and pre-treatment levels of SWA-IgE. Eosinophils are implicated in this plasma IL-5 boost, as their pre-treatment levels, and the 24-hr post-treatment levels of the eosinophil chemo-attractant eotaxin, are also significantly associated with the IL-5 boost. The IL-5 in turn, is related to 9-wk post-treatment eosinophil number and elevated SWA-IgE levels. SWA-IgE was, in line with previous studies, shown to be associated with re-infection immunity. The observation that an IL-5 boost also occurs in allergic rhinitis patients 24-hr after allergen exposure, and after treatment for micro-filarial infection, indicates that the results presented here may also have relevance to the increased IgE levels observed in patients suffering from seasonal allergy and in immunity to other helminths.
|
10.1371/journal.pntd.0005943 | Metabolomics analyses identify platelet activating factors and heme breakdown products as Lassa fever biomarkers | Lassa fever afflicts tens of thousands of people in West Africa annually. The rapid progression of patients from febrile illness to fulminant syndrome and death provides incentive for development of clinical prognostic markers that can guide case management. The small molecule profile of serum from febrile patients triaged to the Viral Hemorrhagic Fever Ward at Kenema Government Hospital in Sierra Leone was assessed using untargeted Ultra High Performance Liquid Chromatography Mass Spectrometry. Physiological dysregulation resulting from Lassa virus (LASV) infection occurs at the small molecule level. Effects of LASV infection on pathways mediating blood coagulation, and lipid, amino acid, nucleic acid metabolism are manifest in changes in the levels of numerous metabolites in the circulation. Several compounds, including platelet activating factor (PAF), PAF-like molecules and products of heme breakdown emerged as candidates that may prove useful in diagnostic assays to inform better care of Lassa fever patients.
| Lassa fever afflicts tens of thousands of people in West Africa each year. The disease progresses rapidly, but there are no tests available to determine which patients are at high risk for dying. We measured the levels of small molecules in the blood of febrile patients with and without infection by LASV that presented to Kenema Government Hospital in Sierra Leone using Ultra High Performance Liquid Chromatography Mass Spectrometry (LCMS), which identifies compounds based on their precise mass. Computational analyses were used to identify compounds that differed in patients with an acute LASV infection, patients with evidence of prior exposure to LASV and patients with fever, but who did not have evidence of exposure to LASV. Several serum metabolites, including factors that are involved in blood clotting and breakdown products of heme, were identified that may prove useful in diagnostic assays that will inform better care of Lassa fever patients or development of therapeutic interventions.
| Lassa virus (LASV), an Old World arenavirus, is the etiological agent of Lassa fever [1]. Lassa fever is endemic to West Africa, with tens of thousands of cases or more estimated to occur annually [2]. The case fatality rate (CFR) in acutely ill Lassa fever patients presenting while viremic was 29–31% in Nigeria [3,4] and 69% in Sierra Leone [5]. During a recent surge of Lassa fever cases in Nigeria the CFRs were >50% [6]. Women who are pregnant develop severe disease with increased frequency and have a Lassa fever CFR as high as 90%, with fetal death, miscarriage or spontaneous abortion occurring in nearly all cases [7,8]. Recent cases in Togo, Benin and areas of Nigeria that rarely have Lassa fever [6,9,10], coupled with serological studies in Mali [11–14], suggest that efforts to improve Lassa fever surveillance should continue [15]. Lassa fever ranks among the most common of the viral hemorrhagic fevers that are imported from Africa [16–20]. Recently, the first case of Lassa fever contracted outside Africa was reported in Germany [21]. There is no approved Lassa fever vaccine, and the only available treatment, ribavirin, is effective only during early infection [5,22,23]. Management of Lassa fever principally involves supportive therapy such as fluid replenishment for dehydration [5].
Lassa fever presents in its early stages as a febrile illness indiscernible from multiple diseases of infectious etiology that are common in West Africa, such as malaria, typhoid, leptospirosis, influenza and various arbovirus-induced illnesses [24]. After the brief nondescript prodrome, progression to fulminant disease occurs rapidly [25]. Lassa fever diagnostics include enzyme linked immunosorbant assays (ELISA) detecting LASV antigen (Ag), human anti-LASV IgG and IgM, a rapid lateral flow immunoassay [5], and polymerase chain reaction based assays detecting viral genomic RNA [26]. Diagnostics can provide the serostatus of patients presenting with a febrile illness and offer a quantitative view of antibody responses. However, the presence of anti-LASV IgM lacks utility as a reliable marker of recent LASV infection [27], and IgM persistence can be confounding as a strategy to monitor disease progression [28]. High viremia is an indicator for poor outcome in Lassa fever [28]. At present, few other markers exist that accurately inform clinical management of Lassa fever patients or patients that survive the acute illness. Increases in serum markers of hepatic damage, particularly aspartate aminotransferase (AST), increase during the acute stage of Lassa fever and levels correlate with fatal outcome in Lassa fever [28–30]. Likewise, the serum levels of cytokines interleukin 6, 8, and 10 (IL-6, IL-8, IL-10), macrophage inflammatory protein 1 alpha & beta (MIP-1α/β), and interferons-alpha/gamma (INF- α/γ) are altered during the course of the disease [28,31–35]. However, subjects that survive also display very high serum levels of liver enzymes or various cytokines, which limits the prognostic value of these markers.
Lassa virus exhibits tropism for circulating leukocytes, and there is evidence to suggest LASV infection results in endothelial dyregulation [36–38] that results in loss of intravascular volume [39]. These circulatory pathologies suggest that the serum of a Lassa fever patient may be an informative, clinically relevant medium to monitor perturbations to homeostasis and may have utility in tracking the trajectory of disease. Furthermore, analysis of Lassa fever at the small molecule level may reveal intermediates of cellular pathways disrupted by LASV replication, suggesting pathogenic mechanisms that the virus utilizes and identifying markers for prognostic diagnostics and potential targets for intervention strategies. Herein, we report findings from a Liquid Chromatography Mass Spectrometry (LCMS) serum metabolomics investigation of a heterogeneous clinical population presenting with febrile illness and triaged to the Lassa Fever Ward at Kenema Government Hospital (KGH) in Sierra Leone.
The Tulane University Institutional Review Board and the Sierra Leone Ethics and Scientific Research Committee approved this project. Patients were referred to the KGH Viral Hemorrhagic Fever Ward from the hospital’s general ward or from regional health centers on the basis of suspicion of Lassa fever. Patients who met the case definition of Lassa fever as defined by Khan et al. [40] were admitted and cared for by the ward’s trained staff. All adult subjects provided written informed consent for publication of their case details. A parent or guardian of child participants provided written informed consent on their behalf.
Small blood volumes (approximately five ml) for serum separation were collected from patients presenting to KGH with febrile illnesses that met preclinical criteria of suspected Lassa fever for diagnostic purposes. Patient samples received a coded designation and were collected in serum vacutainer tubes. Blood samples were allowed to coagulate for 20 minutes at room temperature. Serum was separated from coagulated blood by centrifugation (200 x g, 20 minutes at room temperature). For consented subjects for which there was excess serum not needed for clinical evaluations, aliquots of the serum fraction were stored in cryovials at -20°C prior to processing for metabolite analysis.
A lateral flow immunoassay requiring only a drop of blood obtained with a safety lancet and capable of detecting LASV antigenemia within 15 minutes was utilized to triage cases for possible LASV infection [28,35]. Serum from subjects was subsequently tested using recombinant antigen-based Lassa fever antigen-, IgM- and IgG-capture enzyme-linked immunosorbent assays (ELISA) [24]. Limits of detection and quantitation of the ELISA were based on the upper 95th percentile obtained with a panel of sera from U.S. and Sierra Leonean donors lacking detectable LASV antigens or immunoglobulin M or G (IgM, and IgG) antibodies to LASV recombinant proteins.
All sera collected at the Lassa Fever Ward, KGH is treated as if it contains replication-competent LASV. Serum samples were prepared via a validated metabolomics preparation method utilizing ice-cold methanol for extraction [41,42]. Separated serum samples were depleted of protein by addition to one part sera (100 μL) of 4 parts ice-cold methanol (400 μL), the mixture was vortexed vigorously for 10 seconds, and incubated 1 hour at -20°C followed by centrifugation at 14,000 x g, 15 minutes, 4°C. The supernatant was collected and transferred to a new, sterile vial and dried under vacuum. The resultant small-molecule containing pellets were stored in desiccated, sealed containers and shipped to Tulane University where they were gamma-irradiated. Small molecule containing pellets were dissolved in a solution of 95:5 water:acetonitrile transferred to autosampler vials, and held at -20°C or 4°C immediately prior to analysis [43]. All reagents utilized were HPLC grade.
LCMS methods was optimized based upon a meta-sample consisting of an equal-volume mix of all 50 samples. Detection of metabolites was performed via HPLC separation with ESI-MS (electrospray mass spectrometry) detection. HLPC was performed with an aqueous normal-phase, hydrophilic interaction chromatography (ANP/HILIC) HPLC column: a Cogent Diamond Hydride Type-C column with 4 μm particles and dimensions of 150 mm length and 2.1 mm diameter was used with an Agilent 1290 HPLC system (Agilent Technologies, Santa Clara, CA). Two identical Diamond Hydride columns were connected in series to obtain better separations. The column were maintained at 60°C with a flow rate of 900 μL/min. Chromatography was as follows: solvent consisted of H20 with 0.1% (v/v) formic acid for channel “A” and acetonitrile with 0.1% formic acid for channel “B”. Following column equilibration at 98% B, the sample was injected via autosampler, and the column was flushed for 2.0 min to waste. From 2.0 min to 14.5 min, the gradient was linearly ramped from 98% to 65% B. From 14.5 min to 16.0 min, the gradient was ramped from 65% to 25% B. From 14.5 to 18.0 min the column was held at 25% B, and from 18.0 to 18.2 minutes the gradient was ramped from 25% to 98% B. From 18.2 to 20.0 minutes the column was re-equilibrated with 98% B. An Agilent 6538 Q-TOF with dual-ESI source mass spectrometer was used for all analyses. Resolution was approximately 20,000 and accuracy was 1 ppm. Source parameters were: drying gas 12 L/min, nebulizer 60 psi, capillary voltage 3500V, capillary exit 100V. Spectra were collected in positive mode from 50 to 1700 m/z at a rate of 1 Hz.
Raw spectral data in .d format where uploaded to XCMS Online (Version 1.0.42) and processed as pairwise comparisons using parameters optimized for data acquired with UPLC on an Agilent 6538 MS [44]. Data has been deposited in the XCMS Public archive (https://xcmsonline.scripps.edu/landing_page.php?pgcontent=mainPage) under the identifier Lassa_Serum_PLOS_NTD.
All statistical analyses were carried out using the R statistical software package [45]. Raw mass spectral intensity values and a unique identifier for specific spectral features were extrapolated from XMCS output and compiled into .csv files for machine learning analysis with predefined outcome. Machine learning algorithms built into the R package were utilized with outputs quantifying the sensitivity, specificity, accuracy, and/or receiver operating character computed depending on diagnostic (binary) or prognostic (multi-outcome) analysis. The Random Forest algorithm was employed for all analyses reported. The Random Forest algorithm was set to select features through permutation of the data set yielding the strongest indicators of the input features. The datasets where run with 10-fold cross validation ensuring that ranked output features where selected on importance after predefined, multiple rounds of random training and testing. FacoMineR was utilized for a Principle Components Analyses [46].
A panel of 49 serum samples from patients presenting to Kenema Government Hospital (KGH) in Sierra Leone and triaged to the Viral Hemorrhagic Fever Ward (LFW) with febrile illnesses was analyzed by LCMS. These subjects presented with varying serological status and are representative of the spectrum of illnesses during Lassa fever from acute disease to convalescence (Fig 1). Serum samples where drawn upon admittance. Diagnostic tests for patients were performed the same day the sample was drawn. Twenty subjects tested positive for the presence of LASV in their blood by either antigen-capture ELISA or RT-PCR and were considered to have acute Lassa fever. Five patients died and were classified as having fatal Lassa fever (FL). 15 of these patients survived and were classified as having non-fatal Lassa fever (NFL). 21 subjects without measurable LASV in there blood tested positive for the presence of anti-LASV Immunoglobulin M (IgM) and/or anti-Lassa IgG by ELISA, and were considered to have survived infection with LASV. This group was subdivided into post-Lassa fever patients that were acutely ill (PLAFI, n = 9) or those whose illness was non-acute and presented after an extended period of illness (PLNAFI, n = 11). Nine patients presenting with a febrile illness, but testing negative for the presence of LASV and either anti-Lassa IgM or IgG and were classified as non-Lassa febrile illness (NLF) controls.
There were 28 female and 21 male patient sera screened (57% female) with gender information not available for one patient (S1 Table). The mean age was 26.0±13.9 years with age information not available for one patient. 7 patients died (14% total sample group) with a mean terminal time point (time in days since the onset of symptoms) of 11.4±5.1 days (with time since symptom onset not available for two patients). The mean age for patients who died was 24.6 ±9.7 years.
The antiviral drug ribavirin has been reported have some efficacy in the treatment of Lassa fever, particularly if treatment is begun early during the course of the illness. In this cohort eleven patients with Lassa virus viremia at the time of admission, as well as three IgM positive patients received ribavirin. Spectral features of protonated, sodiated, and potassiated (m/z = 245.0852, 267.0775, & 283.0408; rt = 4.30, 4.32, 4.34, respectively) ribavirin adducts where detected in these samples. Three of the 11 (21%) ribavirin treated acute Lassa fever patients died.
Individual samples produced between 3100 and 6900 different spectral features. LCMS analyses allowed for the putative identification of small molecules in serum from the cohort of febrile patients presenting to KGH. Principle Components Analysis (PCA) of these features indicated that Lassa fever patients with different outcomes and patients at various stages during and after LASV infection segregated according to their serum small molecule profiles (Fig 2). Patients with active or prior LASV infection had profiles that were distinct from febrile patients without serological evidence of current or prior LASV exposure. The small molecule profiles of patients with fatal and non-fatal Lassa fever were also distinct. The overall serum small molecule profile of patients with evidence of prior exposure to LASV and presenting to the Lassa fever Ward after an extended period of illness (post-LASV non-acute) most closely resembled the profile of nonfatal Lassa fever patients. Patients with evidence of prior exposure to LASV, but presenting with an acute illness, showed a distinct small molecule PCA profile.
We used the LCMS data to perform cluster analysis of serum metabolites in subjects with different outcomes following Lassa virus infection, survivors of Lassa virus infection and febrile controls (Fig 3). Platelet activity is depressed during Lassa fever, particularly in terminal patients [29,47]. 24 platelet-activating factor/platelet-activating factor-like molecules were putatively identified and expressed at variable levels in the serum of febrile patients presenting to KGH (Fig 3A, S2 Table). Protonated and sodiated adducts of phosphatidylcholine, platelet-activating factor (PAF) C-16, its metabolic precursors Lyso-PAF C-16 and Arachidonoyl PAF C-16, and 9 additional PAF-like lipids were putatively identified via manual m/z screening. Heat maps of the levels of PAF or PAF-like species illustrate the levels from low (red), intermediate (black) to high (green) in the patient groups. The cluster analysis indicated that nearly all PAFs or PAF-like molecules were present in lower amounts in the serum of patients with fatal Lassa fever than in patients that survived the acute infection (nonfatal Lassa fever). Post-Lassa patients had higher levels of PAF or PAF-like molecules than fatal Lassa fever patients, with the subgroup of patients presenting with an acute illness displaying higher levels than the non-acute group. Non-Lassa febrile illness patients had the highest overall levels of PAF or PAF-like molecules. Extracted ion chromatograms of selected metabolites were analyzed (Fig 4). This analysis confirms the lower levels of two PAFs, PAF4 (PC(O-16:1(11Z)/2:0) H+, m/z 522.3504 and PAF 7 (PC(O-18:2(9Z,12Z)/2:0) Na+, 570.3463) in patients with fatal Lassa fever compared to patients with non Lassa fever (Fig 4A and 4B).
Products of hemoglobin breakdown and various nucleosides were among other spectral features that were putatively identified in the LCMS data set (Fig 3B, S3 Table). Certain of these metabolites were expressed at variable levels in the serum of febrile patients presenting to KGH. For example extracted ion chromatograms confirm that the hemoglobin breakdown products D-urobilinogen and I-urobilin sharing m/z 591.3195 were reduced in patients with fatal Lassa fever compared to Lassa Negative patients (Fig 4C). A spectral feature consistent with the protonated adduct of 7-methylinosine is detected with m/z 283.1016 (theoretical m/z = 283.1037) significantly elevated in the sera of Lassa fever patients who died compared to patients with a non-Lassa febrile illness and other patient groups (Fig 4D). There were several spectral features that could not be putatively identified by their precise mass, and were designated as unknown metabolites.
Lipids constituted the most abundant class of molecules assigned putative identifications in serum samples from the cohort of subjects presenting with febrile illnesses to KGH. 153 substituents of the primary lipid classes were putatively identified included fatty acids and conjugates, fatty esters, glycerophosphocholines, glycerolipids, diacyglycerols, glycerophospholipids, prenol, sterol, sphingolipids, vitamin D3 and derivative species (Fig 3D, S5 Table). Approximately half of the lipids were present in lower amounts in the serum of patients with fatal Lassa fever than in patients with non-fatal Lassa fever. Lipids as a class were generally higher in post-LASV group of patients presenting with an acute illness or in patients with a non-Lassa febrile illness than in the other patient groups.
Random Forest machine learning provided a quantitative assessment of the ability for metabolomics data to discriminate between patients in different serological groups. Several metabolites showed significantly different levels in different groups of patients (Fig 5). For example, PAF4 (PC(O-16:1(11Z)/2:0) H+, m/z 522.3504) and PAF6 (PC(O-18:1(10E)/2:0) H+, m/z 550.3808) were found in significantly lower levels in patients that succumbed to Lassa fever compared to those that survived acute infection (Fig 5A). PAF8 (PC(O-18:2(9Z,12Z)/2:0) H+, m/z 548.3552) and M5 (Unknown 2, m/z 187.0693) were found in significantly higher levels in non-Lassa febrile illness patient than in patients that succumbed to Lassa fever (Fig 5B). M4 (Fibrin monomer breakdown product Na+, m/z 168.075) and M12 (D-Urobilinogen/I-Urobilin Na+, m/z 613.3223) among other metabolites was higher in non-Lassa febrile illness patients compared to the combined groups of patients that presented with acute Lassa fever (Fatal plus nonfatal, Fig 5C). Other PAFs or PAF-like molecules, including PAF7 (PC(O-18:2(9Z,12Z)/2:0) Na+, m/z 570.3463) and PAF12 (PC(O-14:0/2:0) Na+, m/z 613.3223) distinguished acute Lassa fever patients (fatal plus nonfatal) from patients with prior LASV infection (Acute and non-acute presentations, Fig 5D).
The random forests machine-learning algorithm is also able to rank the power of the input variables in predicting sensitivity and specificity for placement in a given serological group. For example, M9 (Mesobilirubinogen H+, 593.3334) provided a sensitivity of 1 and a specificity of .89 when comparing fatal Lassa fever and nonfatal Lassa fever (Table 1). PAF6 (PC(O-18:1(10E)/2:0) H+, 550.38080) provided a sensitivity of 1 and a specificity of .78 when comparing the same groups. Receiver operator characteristic (ROC) curves for both comparisons had a value of 1 (S1A Fig). Among several metabolites with promising diagnostic potential four, PAF4 (PC(O-16:1(11Z)/2:0) H+, 522.3504), PAF8 (PC(O-18:2(9Z,12Z)/2:0) H+, 548.3552), M2 (Unknown 1 H+, 102.0537) and M5 (Unknown 2, 187.0693), showed sensitivities, specificities and ROC of 1 when comparing patients with a fatal outcome from Lassa fever versus nonLassa febrile illness (Table 2, S1B Fig). The algorithm was also able to identify a number of metabolites with possible diagnostic potential for discriminating between acute Lassa fever patients (fatal plus nonfatal) and non-Lassa febrile illness patients (S6 Table, S1C Fig) or patients with prior LASV infection (acute or non-acute presentation, S7 Table, S1D Fig).
LASV induces a dynamic physiological dysregulation within the circulatory system of infected humans, which is manifest in changes in the levels of numerous metabolites. Pathways mediating blood coagulation, hemoglobin breakdown and lipid, amino acid, nucleic acid metabolism are affected during or following LASV infection. Further investigation of these metabolic pathways may inform discovery of novel therapeutic targets for Lassa fever. Metabolites that differentiate Lassa fever patients at various stages of disease, as well those that differentiated these patients from other febrile illness patients presenting to KGH, have been identified. Several compounds, including PAF, PAF-like molecules and products of heme breakdown emerged as candidates that may prove useful in diagnostic assays to inform better care of Lassa fever patients.
Several PAFs or PAF-like molecules demonstrated high sensitivity and specificity for discriminating between patients that ultimately succumbed to fatal Lassa fever and those that survived. The primary physiological role of platelets is to aggregate at the site of endothelial injury where they initiate the clotting cascade to block circulatory leak [48]. Human and nonhuman primates infected with LASV develop clotting abnormalities that manifest in abnormal in vitro platelet aggregation [29,39,49,50]. Levels of platelets in the blood and platelet survival times are normal or only slightly depressed in Lassa patients. Abnormal platelet aggregation correlated with the presence of hemorrhage and with the severity of disease. Here, we demonstrate that levels of PAF or PAF-like molecules were decreased in Lassa fever patients that succumbed to their infection. An as yet to be identified inhibitor of platelet aggregation was identified in the blood of patients with Lassa fever as well as the in patients with Argentine hemorrhagic fever, which is caused by Junin virus, an arenavirus related to LASV [47,51]. The contribution of PAF and PAF-like molecules to hemorrhagic fever pathogenesis appears to be complex. In dengue virus infected patients, PAF appears to be a contributing factor to vascular leakage [52] and higher expression of PAF-degrading acetylhydrolase (PAF-AH) correlates with lower frequency of dengue fever, but not dengue hemorrhagic fever in two ethnically distinct populations [53]. In a murine model of dengue genetic knockout or chemical inhibition of the platelet-activating factor receptor (PAFR) resulted in a less severe disease and increased survival in those animals deficient or inhibited for PAFR [54]. Additional studies will be required to determine if decreased PAF mediated platelet activation contributes to the hemorrhagic manifestations of severe Lassa fever, and whether or PAFs or PAF-like molecules can serve as diagnostic or prognostic markers.
Hemoglobin breakdown products were identified as potential prognostic biomarkers. Mesobilirubinogen (M9) exhibited a specificity of 1 and sensitivity of >86% in discriminating between fatal Lassa fever and either non-fatal Lassa fever or non-Lassa febrile illness. Two confounding factors concerning the presence of heme breakdown products are worth noting for consideration in future studies. First is the malaria endemic locale where over 75% of febrile patients test positive for Plasmodium spp [25]. Second, 35% of the LF patients received ribavirin treatment, a drug attributed to development of anemia [55].
In the present study, serum lipids were the most frequently identified molecular class and also the most frequently identified as decreased in fatal Lassa fever. Similar results were obtained previously in a study of lymphocytic choriomeningitis virus (LCMV) infected mice, a small animal model of arenavirus infection. Stearoyl lysophosphocholine (18:0), identified with positive ion m/z 524.37057, had reduced signal intensity in the serum of infected animals [43]. Proteolytic breakdown products are also observed in this murine arenavirus infection model. The dipeptides γ-glutamyl-Valine, γ-glutamyl-Leucine, and prolyl-hydroxyproline were present in lower amounts in the plasma over the course of LCMV infection in mice [43]. We observed both increased and decreased amounts of a several peptide species. However, no consistent pattern emerged to suggest a mechanism that might account for differences in Lassa fever patients, survivors or febrile controls. Despite the observed differences amongst patient groups there were no peptide or lipid species identified that showed diagnostic sensitivity and specificity approaching that of PAFs or heme breakdown products. In this regard, additional features in the present dataset merit continued investigation to obtain a definite chemical identification. Unknown 1 (H+ m/z 102.0537, NH4+ m/z 119.08) was significantly elevated in serum samples from patients with fatal Lassa fever compared to those with non-Lassa febrile illness. A second unknown spectral feature detected at m/z 187.0693 showed a significant reduction in sera from Lassa fever patients that died compared to non-Lassa febrile patients.
Virus load, the levels of liver enzymes and certain cytokines have predictive value in the outcomes of Lassa fever [28,35,56,57]. Assays to measure these parameters are generally not feasible in austere environments. Many rural health posts across the Lassa fever zone in West Africa are challenged by lack of electricity, minimal lab infrastructure and lack of access to training. LCMS technology is also not feasible in field clinics in West Africa where Lassa fever patients are prevalent. Therefore, simple assays such as dipstick style chromatographic assays, including lateral flow immunoassays, have gained acceptance, and can be conducted with minimal resources and training [35,58–60]. Several small molecules with high biomarker potential were identified including adducts of the modified nucleoside 1-methylinosine. The protonated, sodiated, and potassiated adducts of 1-methylinosine are elevated in the urine of cancer patients [61,62] and in the plasma of patients in renal failure whereupon removal possesses biomarker utility for effective hemodialysis [63]. Reagents used in urinalysis sticks for assaying 1-methylinosine, and similar approaches for quantifying heme breakdown products and other metabolites potentially could be adapted for a Lassa fever prognostic/diagnostic panel.
While these studies have putatively identified a number of compounds with altered serum levels during LASV disease, unequivocal identity of a compound by LCMS requires comparison of the spectrum of the metabolite with a reference standard, which will be pursued for molecules with diagnostic potential. Furthermore, the panel of candidate biomarkers, including PAF, PAF-like molecules and heme breakdown products, must be investigated for diagnostic efficacy alone and in combination in prospective clinical studies. Another limitation of the current study is the relatively low numbers of patients samples analyzed. Additional metabolomics profiling using a larger number of patient samples should be conducted, including LCMS analysis of metabolites in noninvasive specimens such as saliva or urine. Larger panels including other bodily fluids may expand the panel of potential metabolites with diagnostic potential. Metabolic markers for early diagnosis and prognosis of patients at high risk for development of fatal Lassa fever could be integrated into existing clinical and laboratory algorithms for LASV diagnosis and prognosis and improve outcomes of this often fatal disease by identifying cases at greatest risk of death. In addition, the application of metabolomics to reveal fundamental LASV pathogenic mechanisms will potentially provide new targets for therapeutic interventions.
|
10.1371/journal.pgen.1000059 | Coordinated Regulation of Intestinal Functions in C. elegans by LIN-35/Rb and SLR-2 | LIN-35 is the sole C. elegans representative of the pocket protein family, which includes the mammalian Retinoblastoma protein pRb and its paralogs p107 and p130. In addition to having a well-established and central role in cell cycle regulation, pocket proteins have been increasingly implicated in the control of critical and diverse developmental and cellular processes. To gain a greater understanding of the roles of pocket proteins during development, we have characterized a synthetic genetic interaction between lin-35 and slr-2, which we show encodes a C2H2-type Zn-finger protein. Whereas animals harboring single mutations in lin-35 or slr-2 are viable and fertile, lin-35; slr-2 double mutants arrest uniformly in early larval development without obvious morphological defects. Using a combination of approaches including transcriptome profiling, mosaic analysis, starvation assays, and expression analysis, we demonstrate that both LIN-35 and SLR-2 act in the intestine to regulate the expression of many genes required for normal nutrient utilization. These findings represent a novel role for pRb family members in the maintenance of organ function. Our studies also shed light on the mechanistic basis of genetic redundancy among transcriptional regulators and suggest that synthetic interactions may result from the synergistic misregulation of one or more common targets.
| Genetic or functional redundancy is a widespread feature of eukaryotic genomes and may be largely attributable to the complex and overlapping nature of many regulatory networks. Despite the prevalence and importance of this phenomenon, the mechanistic bases underlying genetic redundancy have remained elusive, particularly within the context of multicellular developing organisms. To gain a deeper understanding of this phenomenon, we have focused on a synthetic genetic interaction between lin-35, a C. elegans member of the pRb/pocket-protein tumor suppressor family and slr-2, a Zn-finger protein. Whereas single mutants in either lin-35 or slr-2 develop normally, lin-35; slr-2 double mutants display a highly penetrant synthetic growth arrest during early larval development. We show that this arrest is specifically due to an inability to utilize standard nutrient sources. Using a wide range of approaches including transcriptome profiling, mosaic analysis, and expression analysis, we demonstrate that both LIN-35 and SLR-2 act within the intestine to regulate the expression of many genes that function in nutrient utilization. We also show that the basis of the synthetic interaction between lin-35 and slr-2 is not merely due to regulon overlap, but likely results from the synergistic misregulation of multiple shared transcriptional targets in the intestine.
| The Retinoblastoma protein, pRb, was among the first recognized tumor suppressor proteins [1]–[3], and loss or repression of pRb function is thought to play a causative role in most human cancers [4]–[8]. The role of pRb as a tumor suppressor has been largely attributed to its functions in cell cycle regulation, which it carries out in conjunction with its two family members, p107 and p130, collectively known as the pocket proteins [9]–[11]. Pocket proteins act primarily as transcriptional repressors and physically associate with diverse array of transcription factors [12]. The most thoroughly characterized of these interactions is with E2F family members, which leads to the repression of E2F-target genes, a group that includes many genes required for entry and progression through S-phase [13]–[16]. Correspondingly, LIN-35, the sole pocket protein ortholog in C. elegans, carries out analogous cell cycle functions during larval stages of development [17]–[21].
In addition, a growing number of studies have demonstrated non−cell cycle roles for pRb family members, which in some cases may prove relevant to the tumor-suppressing activity of pocket proteins [12],[19],[22],[23]. In the case of LIN-35, the majority of these functions are revealed only when LIN-35 activity is compromised in specific mutant backgrounds. This phenomenon can be explained on the basis of genetic or functional redundancy, a widespread feature of eukaryotic genomes, which is attributable to the complex and overlapping nature of many regulatory networks. The first described, and still most thoroughly characterized, genetically redundant function of LIN-35 is its role restricting epidermal cells from inappropriately acquiring vulval cell fates [22],[24],[25]. More specifically, when lin-35, a member of the class B group of synthetic multivulval (SynMuv) genes, is simultaneously inactivated with individual members of the SynMuv A or C classes [24],[26], hyperinduction of vulval cells is observed. In contrast, single mutants in most SynMuv genes, including lin-35, do not display observable defects in vulval development.
LIN-35 also redundantly regulates pharyngeal and vulval morphogenesis [27]–[29], asymmetric cell divisions [30], cell fates in the somatic gonad [31], larval growth and development [30],[32],[33], and the promotion of cell death [34]. Furthermore, lin-35 functions non-redundantly in the control of germline gene repression [35] and germline apoptosis [36] and to modulate sensitivity to RNAi [35],[37]. In addition, transcriptome profiling has suggested potential roles for LIN-35 in intestinal and neuronal development, although direct evidence for functions in these tissues has been lacking [21]. Here we describe a novel role for LIN-35 in the intestine of C. elegans. Specifically, we find that LIN-35, in conjunction with the Zn-finger protein SLR-2, acts within intestinal cells to regulate the expression of genes required for proper nutrient utilization.
A previously described genetic screen was used to identify genes that function redundantly with lin-35 [20]. Briefly, we chemically mutagenized lin-35(n745) mutants that carry an unstable extrachromosomal array (kuEx119), which expresses wild-type lin-35 together with the sur-5::GFP marker. Following F2 clonal selection, we identified strains with synthetic interactions by the presence of visible phenotypes in progeny that failed to inherit the array. One allele, ku297, defines a locus that we have designated as slr-2 (for synthetic with lin-35/Rb). slr-2 single mutants are largely indistinguishable from wild type, although we observed weak-to-moderate elongation defects at low frequencies (the Dpy phenotype). In contrast, lin-35; slr-2 double mutants exhibit uniform early-larval arrest (Figure 1A, 1B, Table 1). To verify that phenotypic alleviation by kuEx119 was specifically due to rescue of lin-35 activity, we used RNAi to knock down the expression of lin-35 from the array. This resulted in animals that arrested in early larval development despite the presence of the array (Figure 1A, 1B, inset), confirming that the interaction was specific to lin-35 and slr-2.
We mapped ku297 to an ∼82-kb region of LGV that contains nine genes, including a predicted C2H2-type Zn-finger protein, Y59A8B.13 (Figure 2A). Given the established role of lin-35/Rb in transcriptional regulation as well as our previous findings that lin-35 displays synthetic genetic interactions with other transcriptional regulators, we focused on Y59A8B.13 as a candidate locus. The large size of the Y59A8B.13 genomic locus along with the presence of multiple repetitive elements within this region precluded our amplification and cloning of the Y59A8B.13 locus from C. elegans. We therefore turned to the Y59A8B.13 ortholog from C. briggsae, CBG05648, which though strongly conserved at the amino acid level with Y59A8B.13, comprises a smaller and non-repetitive genomic region (Figure 2B). Injection of a PCR product spanning the complete predicted CBG05648 locus together with a sur-5::RFP plasmid into lin-35; slr-2; kuEx119 hermaphrodites led to the generation of RFP-marked extrachromosomal arrays in seven independent strains. Strikingly, all seven strains were strongly rescued for the lin-35; slr-2 larval-arrest phenotype by arrays containing the C. briggsae Y59A8B.13 ortholog (Figure 1C–E). This result is consistent with previous findings demonstrating the ability of C. briggsae genes to rescue corresponding mutants in C. elegans [38]. Based on these findings, as well as additional data presented below, we conclude that Y59A8B.13 is SLR-2. Our results also underscore the utility of trans-species rescue approaches in cases where technical limitations may preclude the use of the endogenous locus.
Using a combination of approaches, we identified a full-length cDNA corresponding to slr-2 that was trans-spliced at the 5′ end to the SL1 spliced leader sequence (Figure S1) [39]. We note that our cDNA-derived sequence for slr-2 differed from WormBase predictions in the location of several exons. That the translation of slr-2 is likely to proceed from the second 5′ ATG is suggested by sequence alignments with the closely related C. briggsae and C. remanei orthologs (Figure 2B). All three proteins show strong sequence identity within the N-terminal Zn-finger regions but are divergent in their C-terminal domains (Figure 2B). The translation of slr-2 also predicts what appears to be an acidic domain in the C terminus, consistent with SLR-2 functioning as a transcriptional regulator.
Sequencing of the entire slr-2 genomic region in ku297 mutants revealed a single mutation within the splice acceptor site preceding exon six (Figure 2A) [40]. This mutation, which affects the terminal invariant nucleotide, would be predicted to strongly disrupt splicing between exons five and six (e5–e6), leading to aberrant splicing between exons five and seven (e5–e7). The resultant transcript would contain a frameshift followed by a premature stop codon at the eleventh nucleotide of exon 7, leading to the deletion of the C-terminal 250 amino acids of SLR-2. We confirmed this prediction by quantitative real-time PCR analysis, in which we observed a ∼1000-fold decrease in transcript abundance of the e5–e6 product versus the e4–e5 control (data not shown). Furthermore, we observed an e5–e7 splicing product that was ∼80 bp shorter than wild type, consistent with the absence of the 83-bp sixth exon (data not shown). Taken together, these data demonstrate that processing of the normal slr-2 transcript is dramatically reduced in ku297 mutants, suggesting that ku297 likely represents a strong loss-of-function or null allele. Consistent with this interpretation, when placed over a regional deficiency that removes the entire slr-2 locus, slr-2/yDf4, transheterozygotes displayed no exacerbation of the weak Dpy phenotype associated with slr-2 single mutants and were viable and fertile.
As our analysis of arrested lin-35; slr-2 larvae failed to reveal any obvious morphological defects, we undertook transcriptome profiling as a means for shedding light the basis of the double-mutant phenotype. Our rationale for this approach stemmed in part from the known roles of LIN-35 and pRb family members in transcriptional control, as well as the presence of four Zn fingers and an acidic domain in SLR-2, which strongly suggest that it too may function as a transcriptional regulator. Thus, we reasoned that the observed genetic redundancy could be due to the misregulation of targets that are common to both regulators. More specifically, we had previously observed two major classes of genes affected in lin-35 mutants at larval stages: cell cycle control and intestinally expressed genes [21]. We therefore hypothesized that SLR-2 may co-regulate genes in either or both of these classes.
Because lin-35; slr-2 mutants arrest in early larval development, we focused on the late L1 stage for our microarray analysis. Transcriptome profiling was carried out on three independent biological replicates using Affymetrix GeneChips and established procedures [21]. Our analysis identified ∼1,700 genes that are differentially regulated in slr-2(ku297) mutants as compared with identically staged wild-type animals (Figure S2). We further verified expression changes for 29 of these targets by qRT-PCR, thus validating findings from the microarray data (Figure S3 and Table S1).
In common with our previous findings for lin-35, the slr-2 data set showed strong overrepresentation of intestine-enriched/intestine-specific genes (p<0.001) as previously identified using a serial analysis of gene expression (SAGE) approach [41]; 261 genes were common to both data sets (Figure 3A, categories I and II; Figure S4, also see Materials and Methods). A comparison of differentially regulated genes in slr-2 and lin-35 single mutants also revealed a statistically significant overlap (p<0.001); 261 genes were present in both data sets (Figure 3, categories I and IV; Figure S5). Furthermore, although lin-35 and slr-2-responsive genes showed opposite trends in their directionality of regulation (70% of lin-35 targets were upregulated versus 20% of slr-2 targets), the correlation coefficient calculated for common targets was 0.58, indicating a moderate-to-strong correlation. In addition, 76% of genes common to both data sets showed expression changes in the same direction; the common data set contained an approximately equal mixture of up- and downregulated genes.
Importantly, lin-35 and slr-2-responsive genes displaying overlap with the SAGE dataset (Figure 3, categories I, II, and III) also showed evidence of intestinal gene enrichment and intestine-associated functions according to several additional lines of evidence. Based on meta-array functional clustering [42], the only mountains showing strong overrepresentation (p<0.001) were those associated with the intestine, amino acid metabolism, and lipid metabolism (Figure S6A). In addition, available data from the C. elegans expression database (http://gfpweb.aecom.yu.edu), showed that, on average, 78% of genes were expressed in the intestine and 38% showed intestine-specific expression (Figure S6B). Similar results were also obtained from an examination of in situ hybridization data available on the NEXTDB database (http://nematode.lab.nig.as.jp/db2/index.php) (Figure S6C). Moreover, genes found to overlap between lin-35 and slr-2 only (Figure 3, category IV), also showed a specific overrepresentation of intestine and metabolic mountains (p<0.001), a finding further corroborated by data available through the expression databases (Figure S6A-C). Thus, genes implicated in intestinal and metabolic functions are statistically and uniquely overrepresented among the common targets of lin-35 and slr-2, including many genes not previously identified by SAGE analysis.
A previous analysis of intestine-specific/enriched genes identified a single over-represented motif (TGATAA), corresponding to the binding site of the intestinal regulator, ELT-2 [41]. This motif is present in the proximal enhancer regions of 23.3% of genes (using multiple random sampling) in the intestine-specific/enriched dataset used in our above analysis. Using cluster analysis, we independently identified this motif among lin-35 responsive genes [21], and have observed a high frequency of this motif (33–62%) in categories I–IV of overlapping genes from our current analysis (Figure 3A and Figure S6D). This finding is further consistent with our above analysis, indicating that intestine-associated genes our enriched in our dataset. Consistent with the misregulation of intestinal genes, slr-2 mutants exhibited repression of several important metabolic pathways, including the TOR and insulin signaling networks (Table S1), suggesting that slr-2(ku297) mutants experience metabolic stress [43].
In contrast to intestinal genes, genes with cell cycle functions were not overrepresented in the slr-2 data set (Figure S3). Therefore, the genetic redundancy observed for lin-35 and slr-2 mutants is consistent with their combined effects on intestinal gene misregulation and not to cell cycle defects (also see below). We also note that in addition to intestinal genes, the lin-35 and slr-2 data sets showed upregulation of ∼15 common genes with attributed neurological functions. Based on data presented below, it is unlikely, however, that the misregulation of these genes contributes strongly to the double-mutant phenotype. Finally, we observed the downregulation of several dpy genes in slr-2 mutants (dpy-11, dpy-18, and dpy-21), which may account for the variable morphogenetic defects observed in these animals.
The larval-arrest phenotype of lin-35; slr-2 mutants, along with the results of our transcriptome analysis, are consistent with LIN-35 and SLR-2 acting within the intestine to control the expression of gut-associated genes. To determine the precise tissue focus for LIN-35 and SLR-2, we carried out a mosaic analysis [44]–[47]. This method takes advantage of the inherent mitotic instability of most extrachromosomal arrays and allows for the identification of particular mosaic species, thereby enabling direct correlations to be drawn between gene function and localized expression.
To assess the roles of lin-35 and slr-2 in the intestine, we first measured the frequencies by which the C. elegans lin-35 (Ce–lin-35; kuEx119) and C. briggsae slr-2 (Cb–slr-2; fdEx25) rescuing extrachromosomal arrays are spontaneously lost within the intestinal lineage in lin-35 and slr-2 single mutants, respectively. Under these non-selective conditions, intestinal loss was observed in 3.4% and 3.0% of adults carrying the Ce–lin-35 and Cb–slr-2 arrays, respectively (Figure 4H). In contrast, loss of the Ce–lin-35 array in intestinal cells was never observed in adult lin-35; slr-2 double mutants, suggesting that expression of lin-35 from the array is required within intestinal cells for rescue of larval arrest (Figure 4H). Similar results were obtained for the Cb–slr-2 array, indicating that slr-2 also acts within the intestine (Figure 4H).
One limitation to the above analysis is that the absence of intestinal expression in mosaic animals could reflect loss of the array in either the E blastomere (Figure 4D, 4E), which gives rise exclusively to intestinal cells, or in EMS (Figure 4B, 4C) or P1, which are progenitors of E but produce additional cell types (Figure 4A). Thus we determined the frequency of E-specific losses by examining array expression in additional relevant lineages of intestinal mosaic animals. E-specific losses were found to account for 36% of Ce–lin-35 intestinal mosaic animals (n = 11), which is close to the expected frequency based on the lineage (i.e., the failure to segregate the array during one of three possible cell divisions or 33%). Thus of the total number of kuEx119 intestinal mosaic animals (3.4%), we would expect that 36% had experienced loss specifically within the E-cell lineage. This is equivalent to ∼1.2% (or 0.034×0.36) of all animals carrying the Ce–lin-35 array. Thus, if lin-35 is not required within the E-lineage, we would expect to observe ∼40 viable intestinal-mosaic adults among the 3,228 lin-35; slr-2 animals assayed. The total absence of intestinal-mosaic adults strongly indicates that lin-35 function is indeed required in the intestine for rescue of lin-35; slr-2 mutants. Correspondingly, E-specific losses accounted for 40% of Cb−slr-2 intestinal mosaic animals (n = 12). Thus, among the 1,254 lin-35; slr-2; fdEx25 animals assayed, ∼15 E-specific mosaic animals would have been expected. As for lin-35, the complete absence of intestinal mosaics in the adult population demonstrates that slr-2 activity is also required in the intestine (Figure 4H).
The above results conclusively demonstrate that lin-35 and slr-2 are required in the E-cell lineage for rescue of double-mutant lethality. However, these findings do not rule out the possibility that these genes may be simultaneously required in another lineages, such as MS and AB, which are essential for formation of the foregut (Figure 4A). To address the role of MS, we specifically screened for mosaic double-mutant animals in which the Ce–lin-35 array was absent from the MS lineage but was present in E. In doing so, we identified five lin-35; slr-2; kuEx119 viable adults in which the array was absent from the entire MS lineage (Figure 4F, 4G). In addition, we identified a number of rescued adults in which the Ce–lin-35 array was missing from within sub-lineages of MS. Given that viable adults were identified that lacked both lin-35 and slr-2 within the MS lineage, these data demonstrate that neither lin-35 nor slr-2 are required within MS for rescue of lin-35; slr-2 larval lethality. Similarly, we identified viable double-mutant adults where expression patterns indicted the absence of the Ce–lin-35 array within the AB.a (n = 4) and AB.p (n = 3) lineages, suggesting that neither lin-35 nor slr-2 activity are required within the AB lineage for the rescue of double mutants. Taken together, these results indicate that lin-35 activity is not required in the foregut (or other lineages produced by AB and MS) for rescue of lin-35; slr-2 double-mutants.
Previously, we and others have observed a redundant role for lin-35 in the control of intestinal cell proliferation [17],[20]. Given the above indicated intestinal focus of the lin-35; slr-2 phenotype, we examined staged double mutants for abnormalities in intestinal nuclei number and DNA ploidy. We observed wild-type numbers of intestinal nuclei in 54% of lin-35; slr-2 mutants (20±1), although 46% contained 1–14 extra nuclei (Figure S7). This result demonstrates that lin-35; slr-2 mutants undergo larval arrest despite, in most cases, having normal numbers of intestinal nuclei. As a control, we also examined intestinal nuclei numbers in larvae containing a loss-of-function mutation in the APC component, fzr-1, and observed 1–14 extra nuclei in 72% of these animals (Figure S7). Given that only 2% of fzr-1 mutants undergo arrest during larval development (n = 237), this result demonstrates that extra intestinal nuclei per se do not cause larval arrest, a result that is consistent with previous reports [17],[20]. We also failed to detect any difference in DNA ploidy between intestinal cells in wild type and lin-35; slr-2 mutants (Figure S7C-D). Thus, growth arrest in lin-35; slr-2 mutants due to intestinal-associated defects cannot be attributed to overt cell cycle abnormalities.
The above findings indicate that lin-35; slr-2 double mutants are likely to undergo early larval arrest as a result of intestinal-specific gene misregulation. More precisely, malfunctioning of the intestine in double mutants may lead to nutrient deprivation and subsequent arrested growth. To determine whether lin-35; slr-2 animals are defective at nutrient utilization, we made use of the DAF-16::GFP translational fusion reporter. DAF-16::GFP exhibits diffuse cytoplasmic expression throughout the body in animals that are well fed but rapidly translocates to nuclei following their removal from a food source [48]. Accordingly, in wild type we observed diffuse GFP expression in fed animals at all stages and nuclear localization in populations after nutrient deprivation (Figure 5A, 5D, and data not shown). In contrast, lin-35; slr-2 double-mutant larvae displayed high levels of nuclear DAF-16::GFP expression on plates with ample food as early as 12 hours into larval development (Figure 5C, 5D). Nuclear GFP localization further increased in double mutants at later time points, such that nearly 100% of animals displayed punctate fluorescence by 36 hours (Figure 5D). This result suggests that lin-35; slr-2 larvae experience starvation in the presence of a food source, consistent with defects in intestinal functions. We also note that as compared with wild type, both lin-35 and slr-2 single mutants exhibited enhanced nuclear DAF-16::GFP localization on plates containing food at all time points (Figure 5B, 5D). This result is perhaps not unexpected given that both single mutants exhibit mild growth retardation (data not shown) and independently show misregulation in many intestine-related genes.
Because nuclear localization of DAF-16::GFP is known to occur in response to several other forms of environmental stress [48], we performed an independent assay to test whether or not lin-35; slr-2 mutants specifically experience nutritional deprivation. Previous studies have demonstrated that intestinal cell UV-induced autofluorescence, which facilitates the visualization of lysosomal gut granules that serve as sites of fat storage [49], provides a reliable marker for starvation in eating-defective mutants [50]. We therefore examined gut granule autofluorescence in wild-type, lin-35, slr-2, and lin-35; slr-2 staged larvae propagated in the presence or absence of the OP50 food source. Strikingly, by this assay lin-35; slr-2 mutants grown in the presence of food displayed a punctate pattern of gut autofluorescence that was identical to wild-type animals propagated in the absence of the food (Figure 6). In contrast, both single mutants were effectively indistinguishable from wild type under all conditions and time points tested Figure S8). Thus, by two independent assays, lin-35; slr-2 mutants show evidence of experiencing nutritional deprivation.
Based on our evidence that lin-35; slr-2 larvae have intestinal-associated defects and undergo starvation in the presence of their normal bacterial food source, E. coli (OP50), we hypothesized that suppression of larval arrest might be achievable by supplying double mutants with a preprocessed synthetic food source. To test this, we grew lin-35; slr-2 animals in liquid axenic medium, a nutreint source consisting of peptides, fatty acids, hydrolyzed yeast and soy, dried milk, and hemoglobin [51]. Notably, 19% of double mutants grown in axenic medium were capable of bypassing early larval arrest and 5% of the total population went on to become gravid adults (n = 243). Furthermore, when fertile lin-35; slr-2 adults were placed back onto OP50 plates, progeny from these animals arrested uniformly as L1 larvae. We note that although the observed frequency of suppression of larval arrest was relatively modest in these experiments, the ability of the synthetic medium to bypass what is otherwise a completely penetrant phenotype is highly significant. In addition, we examined the expression pattern of DAF-16::GFP in lin-35; slr-2 mutants rescued from larval arrest by growth on axenic media. Notably, we observed cytosol-specific expression in 95% of rescued double-mutant adults (n = 21; Figure S9). This later finding indicates that the nuclear localization of DAF-16::GFP observed in previous experiments (Figure 5) occurs most likely as a consequence of nutritional deprivation and is not due to non-specific effects conferred by the lin-35 and slr-2 mutations.
It has previously been shown that developmental arrest resulting from nutrient deprivation is mediated during the L1 stage by a pathway that includes the DAF-18/PTEN lipid phosphatase [52]. One hallmark of this developmental arrest is the cessation of all germ cell proliferation, leading to diminutive gonads. Consistent with lin-35; slr-2 mutants undergoing starvation-induced arrest, gonad size in double mutants was indistinguishable from wild-type starved L1s (Figure S10E). To determine whether or not germline proliferation is inhibited in lin-35; slr-2 mutants by the DAF-18 pathway, we used RNAi feeding to inhibit daf-18 activity in double mutants (See Materials and Methods). Most notably, average gonad size increased by 2.1-fold in double mutants exposed to daf-18(RNAi) (n = 55) versus OP50 controls (n = 59; Figure S10). This result is consistent with our findings indicating that growth arrest in double mutants is due to nutritional deprivation and further demonstrates that the developmental arrest exhibited by lin-35; slr-2 mutants depends, at least in part, on the DAF-18 pathway.
Although data presented above, including the mosaic and transcriptome analyses, strongly implicates the intestine as the focus of the lin-35; slr-2 growth-arrested phenotype, defects in bacterial cell uptake and mechanical disruption by the foregut (pharynx) could theoretically account for the starvation-induced arrest. To test for this possibility, we directly assayed the ability of lin-35; slr-2 larvae to ingest a food analog (fluorescent beads) and observed normal uptake in 90% of double mutants at all time points tested (Figure S11). Furthermore, based on a GFP-marked OP50 strain, mechanical disruption of bacteria was completely normal in double mutants twelve hours into larval development, at which time double mutants show strong evidence of starvation-induced growth arrest (Figure S12; Figures 5 and 6). Taken together, our cumulative results strongly indicate that lin-35; slr-2 double mutants undergo starvation-induced growth arrest and that this arrest is specifically attributable to defects associated with the intestine.
Our results indicate that both LIN-35 and SLR-2 function redundantly within the intestine to promote nutrient utilization. Consistent with this, LIN-35 is expressed in many cell types throughout early development including cells of the intestine [25],[53]. To determine the pattern of slr-2 expression during development, we constructed a slr-2::GFP transcriptional reporter using an ∼900-bp region upstream of the slr-2 start codon. Based on the location of a nearby adjacent gene, Y59A8B.12 (Figure 2A), this sequence is likely to encompass the complete 5′ regulatory region of slr-2. slr-2::GFP expression was first detected in most or all embryonic cells beginning at around the 100-cell stage (Figure S13). Beginning in late embryonic development and continuing through L1, slr-2::GFP expression was largely restricted to the intestine (Figure 7A, 7B; Figure S13), with highest levels of expression observed in posterior gut cells. A similar expression pattern was also observed in L4 larvae and adults, although expression was detected in additional tissues including the foregut (Figure S13). This expression pattern is consistent with our findings that slr-2 acts during the L1 stage to control the expression of many genes associated with intestinal functions.
As described in the Introduction, lin-35 limits vulval cell induction in cooperation with numerous genes of the SynMuv network. To determine whether or not the observed genetic interaction with slr-2 is specific to lin-35 or is common among other SynMuv family members, we tested seven Class B and one Class A mutant for genetic interactions with slr-2. Interestingly, only one of the tested SynMuv genes, dpl-1, showed strong interactions with slr-2 (Table 1). dpl-1 encodes an ortholog of mammalian DP, which functions as a binding partner for the E2F family of transcriptional regulators. This finding therefore implicates DP and E2F as co-partners of LIN-35 in the transcriptional regulation of intestinal-associated genes. In contrast, the remaining mutants showed at most very weak interactions with slr-2, a finding that underscores the fundamental differences between the role of LIN-35 in vulval cell induction and nutrient utilization (Table 1).
To gain mechanistic insight into the observed differences between the interactions of slr-2 with lin-35 and dpl-1, and the non-interacting SynMuv genes, we used qRT-PCR to assay expression levels of 29 genes identified previously by our transcriptome analysis. This included seven genes with intestine-specific expression, ten with known roles in energy production and metabolic regulation, and eight associated with cell cycle, germline, and RNAi functions (Table S2). We first examined expression levels in staged L1 single mutants to determine the frequency of co-regulation between slr-2 and the SynMuv genes. We define co-regulation as the occurrence of a simultaneous increase or decrease in target gene expression levels by ≥1.5 fold relative to wild type. Interestingly, similar levels of co-regulation with slr-2 were observed between the strong interacting (lin-35 and dpl-1) and weak or non-interacting (lin-9 and hpl-2) SynMuv genes tested; lin-35, dpl-1, lin-9, and hpl-2 showed co-regulation of 12, 17, 16, and 17 genes, respectively (Figure 8). This result suggests that target co-regulation per se is not sufficient to cause a synthetic genetic interaction. Interestingly, co-regulation was only observed among the intestine-specific and metabolic regulator gene classes.
To better understand the basis of the observed genetic interactions, we performed qRT-PCR analysis in staged L1 double mutants (also see Materials and Methods). Specifically, we looked for synergistic misregulation effects on co-regulated genes identified by our analysis of single mutants. Genes were scored positive for synergistic effects if the fold change observed in double mutants was ≥1.95-fold higher than the greatest fold-change observed in either single mutant. Thus, for single mutants displaying fold changes of 3.0 and 6.0, a fold change of 11.7 or greater would be defined as synergistic (greater than additive). Notably, we observed synergistic effects on 7/12 and 8/17 lin-35–slr-2 and dpl-1–slr-2 co-regulated targets, respectively (Figure 8). In contrast, only 2/16 lin-9–slr-2 and 0/17 hpl-2–slr-2 co-regulated targets displayed synergistic misregulation (Figure 8). To see if the above differences were statistically significant, we applied a chi square test and found no significant difference between lin-35; slr-2 and dpl-1; slr-2 double mutants. Comparison of lin-9; slr-2 and hpl-2; slr-2 with lin-35 slr-2 double mutants, however, showed that the observed differences were statistically significant (p = 0.01 and p<0.001, respectively). These results suggest that the strong synthetic phenotypes observed in lin-35; slr-2 and dpl-1; slr-2 double mutants may be due to the synergistic misregulation of common targets.
We have demonstrated a novel role for the pocket protein ortholog lin-35/Rb in nutrient utilization by the intestine. To our knowledge, these findings constitute the first direct demonstration of a role for pRb family members in the basic maintenance of organ functions. Furthermore, this function is carried out redundantly with SLR-2, a putative transcription factor and previously uncharacterized member of the C. elegans Zn-finger protein family. Multiple lines of evidence support the model that LIN-35 and SLR-2 act within intestinal cells to control the expression of genes required for the proper functioning of the digestive tract. (1) The early growth and developmental arrest of lin-35; slr-2 mutants, coupled with an absence of morphological defects, is consistent with an inability to utilize nutrients (Figure 1, Figure S10). (2) Transcriptome profiling of lin-35 and slr-2 single mutants revealed extensive involvement in the regulation of intestine-associated genes (Figure 3; Figure S6). (3) Mosaic analysis demonstrated a requirement for both LIN-35 and SLR-2 within intestinal cells (Figure 4). (4) Based on UV-induced intestinal autofluorescence (Figure 6) and the DAF-16::GFP reporter (Figure 5), lin-35; slr-2 larvae experience nutrient deprivation when grown on a standard bacterial food source. (5) The larval arrest of lin-35; slr-2 mutants can be partially suppressed by growth on a processed synthetic medium. (6) Both lin-35 and slr-2 are expressed in cells of the intestine during early larval development and the expression of slr-2 in the intestine is highly specific at this stage (Figure 7).
Currently, our data do not distinguish between specific classes of intestinal defects such as those affecting nutrient uptake, processing, or the dissemination of nutrients to other tissues. However, the sizeable number and diverse array of intestinal genes affected in both lin-35 and slr-2 mutants could indicate that multiple, and potentially additive, defects may be involved. We note that foregut-based defects are effectively ruled out as a primary cause of starvation in lin-35; slr-2 mutants based on data from our mosaic analysis as well as additional results indicating that double mutants can both internalize and mechanically disrupt OP50 bacteria (Figures 11–12).
A central question posed by our analysis concerns the mechanistic basis of the observed synthetic genetic interaction between lin-35 and slr-2. This question is most directly addressed by our studies of synthetic genetic interactions between slr-2 and nine SynMuv genes (including lin-35), and by our subsequent analysis of transcriptional effects in a subset of both single and double mutants. Somewhat surprisingly, among the 29 genes assayed in single mutants, we found equivalent levels of transcriptional co-regulation between slr-2 and the non-interacting SynMuv genes (hpl-2 and lin-9) and between slr-2 and the strong interactors (lin-35 and dpl-1). This result suggests that regulon overlap alone is not sufficient for the induction of a synthetic genetic interaction (Table 1 and Figure 8). In contrast, only lin-35; slr-2 and dpl-1; slr-2 double mutants showed a high frequency of synergistic misregulation of common targets (Figure 8). This finding suggests that it is the synergistic misregulation of common targets that may specifically predispose animals to displaying strong synthetic phenotypes. It is also worth noting that enhanced effects on targets (equivalent to changes ≥1.95-fold) in double mutants were generally only observed in cases where both single mutants showed evidence of co-misregulation (Figure 8). Although these trends are compelling, our data does not address whether or not the synthetic phenotype occurs as a result of the synergistic misregulation of one or multiple genes. Certainly, the finding that a potentially sizeable number of genes are synergisitically misregulated in both lin-35; slr-2 and dpl-1; slr-2 mutants is consistent with a mechanism involving combined effects on many targets. Nevertheless, in the case of the SynMuv phenotype, it has been shown that it is the ultimate misregulation of a single gene, lin-3, that likely accounts for the excess in vulval cell induction [54].
It is also striking that of the seven Slr mutants identified by our laboratory, four (spr-1, xnp-1, psa-1, and slr-2) encode transcriptional regulators [27],[30],[31]. Furthermore, results from other laboratories reveal a strong overrepresentation of transcription factors among genes that display genetic redundancy with lin-35/Rb, including many of the discovered SynMuv genes [22],[25],[33],[34],[55]. Thus, a common theme among lin-35-synthetic interactors appears to be transcriptional regulation. These findings are further consistent with large-scale screens in S. cerevisiae indicating that genetic redundancy is more frequently observed between genes with similar predicted molecular functions [56], and suggests that similar trends will be observed in C. elegans and higher organisms.
Our discovery of a role for LIN-35 in promoting nutrient utilization by the intestine also complements several recent reports describing roles for pRb family members in intestinal cell proliferation, morphogenesis, and differentiation. Recently, Haigis and coworkers reported that that simultaneous loss of pRb and either p107 or p130 leads to both structural aberrations and incomplete cellular differentiation in intestinal villi [57]. There are conflicting reports, however, as to whether single mutations in pocket protein members are sufficient to disrupt intestinal development [57]–[59]. Notably, one study using a conditional Rb knockout approach observed dramatic hyperproliferation of intestinal epithelium, leading to villi filling up the luminal space [59]. Consistent with this, we and others have observed pronounced hyperproliferation of intestinal nuclei in double mutants of lin-35 and either fzr-1 [20] or cki-1 [17]. In contrast to the above studies, our current findings uniquely implicate LIN-35 in nutrient acquisition or utilization by the intestine (organ function), and show that this defect is not due to overt cell cycle abnormalities. This latter finding is also consistent with an absence of cell cycle genes from the slr-2 regulon (Figure 3; Figure S6). Interestingly, the human ortholog of another lin-35-synthetic gene identified by our screen, xnp-1/ATR-X, was recently implicated in intestinal functions, as mutations in ATR-X led to a wide range of gastrointestinal abnormalities [60].
Our study also further links Zn-finger proteins to pRb-related functions, as a sizeable number of C2H2-type proteins have been shown to physically or genetically interact with pocket protein family members [27], [34], [55], [61]–[66]. In C. elegans, lin-35-interacting Zn-finger transcription factors include lsy-2 and zfp-2, which function coordinately with lin-35 to promote fertility [55]; mcd-1, which acts with lin-35 to promote apoptosis [34]; and spr-4, which redundantly regulates vulval morphogenesis with lin-35 [27]. In addition, the THAP-domain protein GON-14 functions redundantly with LIN-35 to promote larval growth [33], and a mammalian THAP-domain protein, THAP1, acts in parallel to pRb/E2F to control the expression of E2F target genes required for G1–S-phase progression [66]. Thus, interactions between LIN-35 and Zn-finger proteins (frequently C2H2-type proteins) in various organisms have important and versatile roles in development.
C. elegans strains were maintained according to established procedures [67], and all experiments were carried out at 20°C. Strains used in these studies include the following: N2, wild type; MT10430, [lin-35(n745) further backcrossed by our laboratory 5×] [25]; MH1461, [lin-35(n745); kuEx119]; MH1620, [lin-35; slr-2(ku297); kuEx119]; MT1799, [lin-36(n766); unc-32(e189)]; MT5470, [lin-37(n758)]; MT1806, [lin-15A(n767)]; MT2495, [lin-15B(n744)]; MT8840, [dpy-5(e61); lin-53(n833)]; MT8879, [dpl-1(n2994)]; PFR40, [hpl-2(tm1489)]; TJ356, [daf-16::GFP; rol-6]; TY903, [yDf7/unc-76(e911)]; WY53, [lin-35(n745); unc-76(e911); rol-9(sc148)]; WY286, [slr-2(ku297); fdEx25 (CBG05648+sur-5::GFP)]; WY447, [lin-35(n745); slr-2(ku297); fdEx25]; WY471, [lin-35; daf-16::GFP; rol-6]; WY472, [daf-16::GFP; rol-6; slr-2]; WY473, [lin-35; daf-16::GFP; rol-6; slr-2; kuEx119].
Genetic mapping of the slr-2(ku297) locus was performed using established procedures [for details see [68]]. Briefly, slr-2 was mapped between unc-76 and rol-9 on LGV. SNP mapping (using the online SNP database at http://genome.wustl.edu/genome/celegans/celegans_snp.cgi) was then used to place the slr-2 locus in an 82.3-kb region defined by SNP:Y59A8B:89830 and SNP:Y59A8B:172162, which contains nine predicted genes.
The following primers were used to amplify the C. briggsae slr-2 ortholog, CBG05648: 5′-GTGGCATTGTAGGACGATACCC-3′ and 5′-GGAATTCGGAGGGAATTTGAAC-3′. The resulting PCR product, together with the sur-5::RFP marker, was injected into lin-35; slr-2; kuEx119 mutants to generate lines carrying an RFP-marked extrachromosomal array. Rescue was inferred by the ability of the RFP-encoding array to confer viability to worms lacking the GFP-marked lin-35 rescuing array (kuEx119). Seven of seven lines isolated from independently injected P0s yielded strains that could be propagated in the absence of kuEx119.
3′ RACE was performed using domains of Y59A8B.13 that are conserved between C. elegans, C. briggsae, and C. remanei. 5′ RACE was performed using a primer complementary to the SL1 trans-splice leader sequence and a primer specific to Y59A8B.13.
RNA was extracted from staged L1 larvae, purified, and used for microarray analysis as previously described [21]. Differentially expressed genes were identified by comparison with identically staged N2 worms using RMA software, as previously described [21]. RNA was extracted from staged L1 larvae, purified, and used for qRT-PCR as previously described [21]. Intestine-specific/enriched genes referenced in our studies were obtained from a previously published SAGE analysis and displayed a minimum intestine:whole worm tag ratio of >2.6 as previously determined [41]. Our P-values were calculated using either t-tests or chi-square tests, where appropriate, using the statistic language R. Pearson correlation coefficients between lin-35 and slr-2 coregulated genes as well as standard errors of mean (or deviation) for all other experiments was calculated using Microsoft Excel. A staged (L1) population of lin-35; slr-2 double mutants (lacking the kuEx119 array) was obtained using a COPAS worm sorter from Union Biometrica.
A transcriptional reporter was generated by amplifying an ∼900-bp region upstream of the slr-2 start codon using the following primers: 5′-CCCATTATCGGCCATTTTTGCTG-3′ and 5′-GGTGCAGGTCGACACTTTTCGACATTTCCGGTGGTCTG-3′. Based on the location of the confirmed gene, Y59A8B.12, which is located <1 kb upstream of the translational start site for slr-2, this sequence is likely to encompass the complete 5′ regulatory region of slr-2. Following digestion with XhoI and SalI, the resulting PCR product was inserted into pPD95.69 (gift of A. Fire), and the obtained plasmid was injected in N2 animals to establish multiple independent extrachromosomal arrays, all of which showed similar patterns of GFP expression. One extrachromosomal array line was then chosen for integration using standard irradiation methods [69].
RNAi was carried out using standard procedures [70]. MH1620 L4 hermaphrodites were placed on daf-18(RNAi) feeding plates and F1 progeny were assayed 4–5 days later for gonad size. |
10.1371/journal.pntd.0001402 | Effect of a Control Project on Clinical Profiles and Outcomes in Buruli Ulcer: A Before/After Study in Bas-Congo, Democratic Republic of Congo | Buruli ulcer (BU) is a necrotizing bacterial infection of skin, subcutaneous tissue and bone caused by Mycobacterium ulcerans. Although the functional impairment caused by BU results in severe suffering and in socio-economic problems, the disease remains largely neglected in Africa. The province of Bas-Congo in Democratic Republic of Congo contains one of the most important BU foci of the country, i.e. the Songololo Territory in the District of Cataractes. This study aims to assess the impact of a BU control project launched in 2004 in the Songololo Territory.
We used a comparative non-randomized study design, comparing clinical profiles and outcomes of the group of patients admitted at the General Reference Hospital (GRH) of the “Institut Médical Evangélique” (IME) of Kimpese 3 years before the start of the project (2002–2004) with those admitted during the 3 years after the start of the project (2005–2007).
The BU control project was associated with a strong increase in the number of admitted BU cases at the GRH of IME/Kimpese and a fundamental change in the profile of those patients; more female patients presented with BU, the proportion of relapse cases amongst all admissions reduced, the proportion of early lesions and simple ulcerative forms increased, more patients healed without complications and the case fatality rate decreased substantially. The median duration since the onset of first symptoms however remained high, as well as the proportion of patients with osteomyelitis or limitations of joint movement, suggesting that the diagnostic delay remains substantial.
Implementing a specialized program for BU may be effective in improving clinical profiles and outcomes in BU. Despite these encouraging results, our study highlights the need of considering new strategies to better improve BU control in a low resources setting.
| Buruli ulcer (BU), which is caused by Mycobacterium ulcerans, is an important disabling skin disease. However, BU has been neglected in many endemic African countries, including in the Democratic Republic of Congo. The province of Bas-Congo contains one of the most important BU foci of the country, i.e. the Songololo Territory in the District of Cataractes. In 2004 a specialized BU control program was launched in that area. The present study aims to evaluate the impact of the above-mentioned program, by comparing clinical profiles and outcomes of the group of patients admitted at the General Reference Hospital (GRH) of the “Institut Médical Evangélique” (IME) of Kimpese 3 years before the start of the project (2002–2004) with those admitted during the 3 years after the start of the project (2005–2007). The project implementation was associated with a strong increase in the number of admitted BU cases at the GRH and a fundamental change in the profile of those BU patients. Despite these encouraging results, our study provides some limitations of such program, and highlights the need of considering new strategies to better improve BU control in a low resources setting.
| Buruli ulcer (BU) is a necrotizing bacterial infection of skin, subcutaneous tissue and bone, caused by an environmental pathogen, Mycobacterium ulcerans [1]. Although the functional impairment caused by BU results in severe suffering and in socio-economic problems [2], the disease remains largely neglected by health authorities in Africa [3]. BU is considered as one of the Neglected Tropical Diseases with a poorly known global prevalence [4].
The province of Bas-Congo (Lower Congo) in the Democratic Republic of Congo (DRC) contains one of the most important BU foci of the country, i.e. the Songololo Territory in the District of Cataractes [5]–[10]. Meyers et al. reported that BU existed in that region before 1935 on the basis of interviews of former patients [7]. The first BU case reports were published in the sixties [5]–[7] followed by a long period without reported cases. Since 1999, the general reference hospital (GRH) of the Institut Médical Evangélique (IME)/Kimpese, located in the Songololo Territory, 220 km southwest of Kinshasa, regularly admits BU cases. Between 2002 and 2004 this hospital admitted 64 patients, 95% of them in the ulcerative stage.
During this period, 48 patients out of 64 (75%) were referred by government health centers or other health professionals, 9 (14.1%) by family members, and 7 (10.9%) presented spontaneously. Surgery was the main method of treatment applied amongst these patients (93.7%). An abnormally high case fatality rate (18.7%) was observed among these 64 patients, and whereas 36% presented already a functional limitation at the time of diagnosis, 23% were discharged with permanent disability. The median length of hospitalization was 89 days and, -noteworthy- 90% of the patients were not able to pay their hospitalization costs.
To address these poor clinical outcomes, the American Leprosy Mission and the IME hospital launched a BU control project in Songololo Territory in 2004. The principal aims of this project were (i) the improvement of the patient care of BU patients admitted at the GRH IME/Kimpese and (ii) the promotion of early community-based detection of suspected BU cases. The aim of this study is to evaluate the impact of this specialized BU control program on clinical profiles and outcomes.
Ethical clearance for this study was obtained from the Institutional Review Board of IME. All patients, or their guardian in the case of minors, provided informed consent for all diagnostic and treatment procedures and publication of any or all images derived from the management of the patient, including clinical photographs that might reveal patient identity.
The BU control project started at the end of 2004 and introduced free patient care for BU patients during their admission at GRH IME/Kimpese, whereas this was hitherto to be paid on a fee-for-service basis. Furthermore, the patients benefited from a free daily nutritional supplement, and specific antibiotherapy was introduced in accordance with WHO recommendations [11], as well as a physiotherapy program for prevention of disabilities. Simultaneously the project organized awareness raising campaigns in the endemic communities, based on a mass-media approach targeting the general public, followed by active case-finding and referral of suspected cases to the specialized BU care centre. The project was based on the following five components: Improving facilities' management and treatment skills; Prevention of disabilities and physical rehabilitation; Feeding patients and psychological and social support for those affected; Stepping up Information, Education and Communication for the general public and community-based surveillance, and Training and research.
To evaluate the effect of this control project, we used a comparative non-randomized study design, comparing patient demographic profiles and clinical outcomes of the group of patients admitted at the GRH IME/Kimpese in the 3 years before the start of the project (2002–2004) with those admitted during the 3 years after the start of the project (2005–2007).
We have included all consecutive patients clinically diagnosed as BU and admitted to the Surgical Department of GRH IME/Kimpese from January 2002 to December 2007. The clinical case definition elaborated by the World Health Organisation (WHO) was used to diagnose BU [12]. Additionally for the second period, as recommended by the WHO [11], we introduced patients' categorization as follows: A single lesion <5 cm (Category I); A single lesion 5–15 cm (Category II); A single lesion >15 cm, multiple lesions, lesions at critical sites (face, breast and genitalia) or osteomyelitis (Category III). For all patients included in this study, the diagnostic confirmation process consisted of swabs from ulcerative lesions and biopsies for the laboratory confirmation (bacteriology and/or histopathology) of suspected cases according to WHO recommendations [12]. The initial direct smear examinations for acid-fast bacilli and histopathologic analyses were made at the IME/Kimpese laboratory. Other specimens from the same patient were sent in a transport medium to the Mycobacteriology Unit of the Institute of Tropical Medicine (ITM) in Antwerp, Belgium [13], where Ziehl-Neelsen (ZN) staining, in vitro culture on Löwenstein-Jensen medium, and PCR for the detection of M. ulcerans DNA were performed according to WHO recommendations [12]. Formalin-fixed tissues were sent to the Department of Infectious and Parasitic Diseases Pathology of the Armed Forces Institute of Pathology in Washington DC, for the histopathological confirmation of diagnosis [10].
Throughout the whole study period, clinical data of BU patients were recorded on standardized Case Report Forms elaborated by WHO (known as form BU01) and the data were entered in a standardized case registry form (BU02) [14]. Next, these data were entered into an Excel database (Microsoft Corporation, Redmond, WA) and analyzed with Epi-Info version 3.3.2 (Centers for Diseases Control and Prevention, Atlanta, GA). The Pearson chi-square test was used to compare proportions with a significance level set at 5%, as well the Fisher's exact test when an expected cell value was less than 5.
To evaluate the relevance and the effect of the BU control project, we used the conceptual framework to evaluate public health programs proposed by Habicht et al. [15]. The principal indicators considered for the data analysis are the number of recorded cases for each period, the number of new cases and relapses, the proportion of cases with functional limitation of joints at diagnosis, the proportion of cases confirmed by at least one laboratory test, the proportion of ulcerative forms at diagnosis, the type of treatment applied, the proportion of discharged cases with functional limitation of joints, the median duration of hospitalization, and the case-fatality rate. Relapse was defined in both study periods as a new confirmed diagnosis of BU less than one year after being declared cured from BU after treatment (surgical only in the first period, antibiotic and/or surgical in the second period). Functional limitation was defined as any reduction in the range of motion of one or more joints, and was assessed based on clinical observation.
Lesions were considered as mixed forms when simultaneous presence of different forms of disease including bone and joint involvement in the same patient was noticed. Besides, we defined as simple ulcerative forms (SUF) the ulcerative lesions not associated with other clinical lesions such as papule, nodule, plaque, edema or osteomyelitis at the same site.
The number of suspected BU cases admitted at GRH IME/Kimpese strongly increased after the start of the BU control project. The average number of annual admissions for BU tripled, from 21 cases per year for the period 2002–2004, to 63 cases per year for 2005–2007 (Figure 1). The clinico-epidemiological features and the results of patient management are shown in Tables 1 and 2. The origin of patients remains mainly the Songololo Territory, Cataractes District, where the GRH IME/Kimpese is located (Figure 2). The median age of patients (20 years) was similar for both periods. The proportion of female patients increased significantly from 30% before to 49% after the project was initiated (p = 0.005).
In both periods, the majority of BU patients were new cases, yet the proportion of relapse cases amongst all admissions reduced from 32.8% to 11.6% (p<0.001) after 2004.
The proportion of ulcerative forms at admission decreased from 95.3% to 85.8% after 2004 (p = 0.041), and the proportion of SUF increased from 32.8% to 60.7% amongst the ulcers (p<0.001) (Figure 3). There was no change in the proportion of confirmed osteomyelitis nor in the proportion of patients presenting with joint movement limitations. The reported median duration of the disease since the appearance of first symptoms increased from 6 to 8 weeks. Globally, the proportion of patients who healed with complications did not change significantly from 23.4% to 19.5% (p = 0.496), even amongst patients declared cured only, from 31.3% to 21.0% (p = 0.136).
However, the number of cases that healed without complications increased significantly from 51.6 to 73.2% (Figure 4) (p = 0.001). The proportion of cases confirmed by at least one laboratory test positive for M.ulcerans remained the same (70% in 2002–2004 versus 61% in 2005–2007, p = 0.183).
Antibiotic therapy was introduced as part of the control project, and was prescribed to 56.3% of patients, although most patients continued to receive surgery (93.7% previously compared to 84.2% after 2004, p = 0.052). Ninety patients (47.4%) were treated by a combination of antibiotics (rifampicin and streptomycin) and surgery. Seventy patients (36.8%) were treated with surgery alone, seventeen patients (8.9%) only with antibiotics, and thirteen (6.8%) were treated with daily wound dressing.
The median duration of hospitalization, around 90 days, was approximately similar during both periods (Table 2) and varied by disease category during the second period, respectively 60 days for category I (Figure 5 and 6), 81 days for category II, and 118 days for category III.
The case fatality rate was significantly decreased from 18.7% during the previous period (12 out of 64 patients) to 3.2% (6 out of 190 patients) during the second period (p<0.001). Conditions associated with mortality among BU patients in the previous period were as follows: sepsis in four patients out of twelve (33%), malnutrition and anaemia in nine patients (75%), edematous disseminated disease in two patients (16.6%), postsurgical shock in one patient (8%), and cancerization in two patients (16.6%).
The BU control project was associated with a strong increase in the number of admitted BU cases at GRH IME/Kimpese and a fundamental change in the profile of those BU patients. Since the implementation of the control project we observed equal numbers of men and women presenting with BU, significant decrease in the proportion of relapse cases and significant increases in the proportion of early lesions and simple ulcerative forms, and in the proportion of patients healed without complications. Importantly, the case fatality rate decreased significantly from 18.7% to 3.2%.
While those parameters indicate a positive impact of the project, we are aware of the limitations of our study. For our evaluation, we used a historical control group: BU patients admitted at the hospital before the project (2002–2004) were compared to those who benefited from the implementation of the control project (2005–2007). Although such before/after evaluation design does not provide conclusive evidence that the observed changes are attributable to the control project itself, it is usually considered sufficient by policy makers to conclude to a beneficial effect [15]. The threefold increase in the number of BU cases admitted annually can to a large extent be explained by the active case-finding and the reduction of the financial barrier, as patient care was free after 2004, but is probably also due to the improvement of patient management and the quality of clinical results. While both aspects are likely partially involved in the observed results, the observational study design will not allow us to distinguish between the two.
The capacity strengthening of medical staff on the surgical management of BU patients through local and international training, the introduction of specific antibiotherapy (rifampicin and streptomycin), and implementation of a program for the prevention of disabilities have contributed to improvement of clinical outcomes (the increased proportion of patients healed without complications, the reduction of the proportion of relapses, and the reduction of the case fatality rate). Furthermore, we assume that the improved access to adequate and prompt BU treatment in the second period through the free patient care, and the free daily nutritional supplement offered played a major role in the improvement of clinical outcomes during the second period. Indeed, in Africa, the challenge for health care professionals working with BU patients is to break up the cycle of poor clinical outcomes leading to loss of confidence of the affected communities in the hospital [16]. Debacker et al. reported that in the Centre Sanitaire et Nutritionnel Gbemoten (CSNG), Zagnanado, Benin, 68.3% of patients were referred to the hospital by a former BU patient. The improved quality of care at CSNG resulted in a reduction of the median duration of hospitalization from 9 months in 1989 to 1 month in 2001, and the median delay in seeking medical care dropped from 4 months in 1989 to 1 in 2001 [17]. The introduction of a BU program was an important factor in the marked reduction in patient delay. Furthermore, after promotional sessions on BU organized in 2000 by the National BU program in the Zou, Oueme, and Atlantique Departments, patients reported earlier than in 1999 [17]. We are hopeful that similar results will develop at the Territory of Songololo in Bas-Congo.
Awareness raising campaigns followed by active case-finding have contributed to the dissemination of information on BU among the communities in Songololo during the intervention period. We assume that the active case-finding activities have contributed to the change of the Male/Female ratio from 2.4/1 before the project to 1.02/1 during the project period, and thus, the project seems to have contributed to equilibrate the gender balance. During the first period, male BU patients were more frequent probably due to sociocultural barriers for women to seek care, whereas during the second period the active case-finding activities helped the female patients to overcome these barriers.
Progressively, more early lesions and more SUF were diagnosed at the hospital. However, rather surprisingly, the median duration since the onset of first symptoms remained high after the project was launched. Reasons why the median delay in seeking medical attention was higher during the second period compared to the first one remain unclear, and need to be assessed. This may explain the fact that the number of confirmed osteomyelitis cases, limitations of joint movement, both at diagnosis as well as at healing, and patients needing surgery, remained similar. This is problematic, as the huge clinical impact of BU is mainly due to the late detection of cases [18]. Indeed, an extended delay before presentation to the hospital has been identified as one of the most important risk factors for bone involvement. Between 1996 and 2007, out of 930 confirmed and treated BU patients at Zagnanado, Benin, 106 (11.4%) presented an osteomyelitis caused by M. ulcerans. The median delay between onset of symptoms and consultation was 167 days for patients with bone involvement and 61 days for those with cutaneous lesions (p<0.001) [19]. In most endemic regions, consulting the hospital seems to be the last resort when other attempts were unsuccessful and when the disease has reached an advanced stage with large cutaneous ulcerations or other complications, such as joint contractures or osteomyelitis [19]. Stienstra et al. reported in their study on the beliefs and attitudes towards BU in Ghana that in 59% of cases, witchcraft was mentioned as cause of the disease. Among the interviewed patients, 52% applied herbs on their lesions and consulted a hospital as last resort. The reasons evoked were (i) financial difficulties [30]% of patients), (ii) the fear of treatment at the hospital and in particular amputation, and (iii) expectations of a spontaneous healing [20]. Recently, a study conducted by Renzaho et al. in Ga West district in Ghana demonstrated that 71.8% of BU patients consulted a traditional practitioner first and that the hospital was consulted as last resort [16]. Meyers and others noted that in the Songololo Territory, DRC, the reasons for which many BU patients delayed seeking medical assistance were obviously complex, but cultural, economic, and transportation factors were especially important [7]. Recently, a study conducted in the same area showed that all interviewed patients first adopted a “wait and see” attitude which lasted on average 2 months [21]. Similar observations were reported in other African countries as Cameroun [22], and Benin [23]–[25]. Those studies were realized when surgery was still the treatment of choice; the recent introduction of specific antibiotherapy as first line treatment may alter this behavior [26]–[28].
These social, economical, geographical and cultural reasons, that limit the access to health care in endemic regions, suggest that the number of admitted patients at GRH IME/Kimpese may represent only the emerged part of the iceberg. The free of charge policy offered to patients does not resolve completely the problem of financial barriers related to the patient management of BU. The study conducted by Grietens et al. in two hospitals with a specialized program for BU in Cameroun, similar to ours, has shown that in spite of the reduction of the treatment costs, the hospitalization for BU remains financially and socially untenable for patients and their households, leading to the abandonment of biomedical treatment or a complete refusal [29]. Therefore, there is a need to consider new control strategies which are both socially and financially acceptable and appropriate for the concerned communities.
Overall, the results after 3 years of implementation of BU control activities in Songololo Territory are encouraging. However, the morbidity and disabilities due to BU remain high among our patients. The burden of BU in terms of human suffering, long duration of hospitalization, the development of disabling sequelae, and socio-economic repercussions, is mainly attributable to the late detection of cases. For this reason, secondary prevention through earlier case detection and treatment remains one of the key measures in the control of BU [30].
To reduce the burden and to increase the coverage of the population at risk, we consider that a dedicated BU control program at central and provincial level, that operates in close collaboration with the existing polyvalent health services, would be the most efficient way to organize the control of BU in Songololo Territory. The aforesaid program should involve education of the population in the endemic areas, training of healthcare workers, early detection by active case-finding and adequate case management provided free of charge. Further decentralization and integration of BU control activities may improve access to diagnosis and care at the most peripheral level of the health system. A close collaboration between the BU control project and the health zones is essential for the implementation of a simple, functional, and efficient active surveillance system in a resource-limited context.
|
10.1371/journal.pcbi.1002443 | Replication Fork Polarity Gradients Revealed by Megabase-Sized U-Shaped Replication Timing Domains in Human Cell Lines | In higher eukaryotes, replication program specification in different cell types remains to be fully understood. We show for seven human cell lines that about half of the genome is divided in domains that display a characteristic U-shaped replication timing profile with early initiation zones at borders and late replication at centers. Significant overlap is observed between U-domains of different cell lines and also with germline replication domains exhibiting a N-shaped nucleotide compositional skew. From the demonstration that the average fork polarity is directly reflected by both the compositional skew and the derivative of the replication timing profile, we argue that the fact that this derivative displays a N-shape in U-domains sustains the existence of large-scale gradients of replication fork polarity in somatic and germline cells. Analysis of chromatin interaction (Hi-C) and chromatin marker data reveals that U-domains correspond to high-order chromatin structural units. We discuss possible models for replication origin activation within U/N-domains. The compartmentalization of the genome into replication U/N-domains provides new insights on the organization of the replication program in the human genome.
| DNA replication in human cells requires the parallel progression along the genome of thousands of replication machineries. Comprehensive knowledge of genetic inheritance at different development stages relies on elucidating the mechanisms that regulate the location and progression of these machineries throughout the duration of the DNA synthetic phase of the cell cycle. Here, we determine in multiple human cell types the existence of a new type of megabase-sized replication domains across which the average orientation of the replication machinery changes in a linear manner. These domains are revealed in 7 somatic cell types by a U-shaped pattern in the replication timing profiles as well as by N-shaped patterns in the DNA compositional asymmetry profile reflecting the existence of a replication-associated mutational asymmetry in the germline. These domains therefore correspond to a robust mode of replication across cell types and during evolution. Using genome-wide data on the frequency of interaction of distant chromatin segments in two cell lines, we find that these U/N-replication domains remarkably correspond to self-interacting folding units of the chromatin fiber.
| Comprehensive knowledge of genetic inheritance at different development stages relies on elucidating the mechanisms that regulate the DNA spatio-temporal replication program and its possible conservation during evolution [1]. In multi-cellular organisms, there is no clear consensus sequence where initiation may occur [2], [3]. Instead epigenetic mechanisms may take part in the spatial and temporal control of replication initiation in higher eukaryotes in relation with gene expression [4]–[9]. For many years, understanding the determinants that specify replication origins has been hampered by the small number (approximately 30) of well-established replication origins in the human genome and more generally in mammalian genomes [1], [7], [10]. Recently, nascent DNA strands synthesized at origins were purified by various methods [11]–[14] to map a few hundreds putative origins in 1% of the human genome. For unclear reasons, the concordance between the different studies is very low (from to ) [12]–[15]. In a completely different approach to map replication origins, previous in silico analyses of the nucleotide compositional skew of the human genome showed that the sign of abruptly changed from to when crossing known replication initiation sites. This allowed us to predict putative origins at more than a thousand sites of sign inversion (-jumps) along the human genome [16], [17]. Further analyses of patterns identified 663 megabase-sized N-domains whose skew profile displays a N-like shape (Fig. 1A), with two abrupt -jumps bordering a DNA segment whose skew linearly decreases between the two jumps [16]–[21]. Skew N-domains have a mean length of Mb and cover 29.2% of the human genome. The initiation zones predicted at N-domains borders would be specified by an open chromatin structure favorable to early replication initiation and permissive to transcription [21], [22]. The determination of HeLa replication timing profile [23] and the analysis of available timing profiles in several human cell lines [24]–[26] allowed us to confirm that significant numbers of N-domains borders harbor early initiation zones active in germline as well as in somatic cell types [18], [27].
Recent studies have shown that replication induces different mutation rates on the leading and lagging replicating strands [27]. This asymmetry of rates acting during evolution has generated the skew upward jumps that result from inversion of replication fork polarity at N-domain extremities. The skew profile along N-domains would result from superimposed effects of transcription and of replication [19], [20], [28]–[31]. Accordingly, the linear decrease of the skew (Fig. 1A) may reflect a decrease in the proportion of replicating forks propagating from the left (5′) to the right (3′) N-domain extremity. This organization of replication in a large proportion of the genome contrasts with the previously proposed segmentation of mammalian chromosomes in regions replicated either by multiple synchronous origins with equal proportion of forks coming from both directions (0.2–2.0 Mb Constant Timing Regions) or by unidirectional replication forks (0.1–0.6 Mb Transition Timing Regions) [25], [32]–[34].
Here, to determine the existence of a new type of replication domains presenting gradients of replication fork polarity, we establish (i) that the replication fork polarity and the compositional skew are proportional to each other, (ii) that the replication fork polarity can be directly extracted from the derivative of the replication timing profile. Taking advantage of replication timing profiles in several human cell types [23], , we show that the derivative of the replication timing profile of N-domains is shaped as a N. The corresponding U-shape of the replication timing profile is not specific to the germline but is generally observed in all replication timing profiles examined, thus establishing these “U-domains” as a new type of replication domains, consistent with the recent experimental observation of multiple replication initiations in most Transition Timing Regions in several human cell lines [35]. As observed with the early initiation zones bordering N-domain extremities, those specific to the U-domains are significantly enriched in open chromatin markers as well as insulator-binding proteins CTCF [36], [37] and are prone to gene activity. Analysis of recent Hi-C data [38] reveals that U-domains correspond to self-interacting structural chromatin units. These data make a compelling case that the “islands” of open chromatin observed at U-domains borders are at the heart of a compartmentalization of chromosomes into chromatin units of independent replication and of coordinated gene transcription.
To establish the existence of replication domains associated with replication fork polarity gradients, we first demonstrate the relations between replication fork polarity, nucleotide compositional skew and derivative of the replication timing profile. Under appropriate hypotheses, the skew resulting from mutational asymmetries associated with replication is proportional to the fork polarity at position on the sequence (Material and Methods):(1)where (resp. ) is the proportion of forks replicating in the (resp. ) direction on the Watson strand. The linear decrease of in N-domains from positive ( end) to negative ( end) values thus likely reflects a linear decrease of the replication fork polarity with a change of sign in the middle of the N-domains. This result strongly supports the interpretation of N-domains (Fig. 1A–C) as the signature of a higher-order organization of replication origins in germline cells.
The replication fork polarity can also be directly deduced from replication timing data under the central hypotheses that the replication fork velocity is constant and that replication is bidirectional from each origin. Note that recent DNA combing experiments in HeLa cells have shown that replication fork velocity does not significantly vary during S phase which strongly supports the former hypothesis [35]. We demonstrate that the replication fork polarity is the product of the derivative of the mean replication timing (MRT) and the replication fork velocity (Material and Methods):(2)The fork polarity should therefore provide a direct link between the skew and the derivative of the replication timing profile in germline cells. To test this relationship, we used a substitute for germline MRT, the replication timing profiles of seven somatic cell lines (one embryonic stem cell, three lymphoblastoid, a fibroblast, an erythroid and HeLa cell lines) (Material and Methods). We first correlated the skew with , in the BG02 embryonic stem cells, over the 22 human autosomes (Fig. 1D). The significant correlations observed in intergenic (, ), genic (, ) and genic (, ) regions are representative of the correlations observed in the other 6 cell lines (Table 1). These correlations are as strong as those obtained between the profiles in different cell lines (Supplementary Table S1), as well as those previously reported between the replication timing data themselves [26], [34], [39]. The correlations between and are even stronger when focusing on the 663 skew N-domains (Table 1). The correlations obtained in intergenic regions () are recovered to a large extent in genic regions () where the transcription-associated skew was hypothesized to superimpose to the replication-associated skew [18]–[20]. Further evidence of this link between and was obtained when averaging, for the different cell lines, the profiles inside the 663 skew N-domains after rescaling their length to unity (Fig. 1E). These mean profiles are shaped as a N, suggesting that some properties of the germline replication program associated with the pattern of replication fork polarity are shared by somatic cells.
According to Equations (1) and (2), the integration of the skew is expected to generate a profile rather similar to the replication timing profile. In segments of linearly changing skew, the integrated function is thus expected to show a parabolic profile. The integrated function when estimated by the cumulative skew (Fig. 1B) along N-domains of a 11.4 Mb long fragment of human chromosome 10, indeed displays a U-shaped (parabolic) profile likely corresponding the replication timing profile in the germline. Remarkably, the 6 N-domains effectively correspond to successive genome regions where the MRT in the BG02 embryonic stem cells is U-shaped (Fig. 1C). The 7 putative initiation zones ( to ) corresponding to upward -jumps (Fig. 1A), co-locate (up to the kb resolution) with MRT local extrema which supports that they are highly active in BG02. These initiation zones can present cell specificity as exemplified by the putative replication origin which is inactive (or late) in both the K562 erythroid and GM06990 lymphoblastoid cell lines (Fig. 1C) resulting in domain “consolidation” [40]. Two neighboring U-domains ( and ) in BG02 merged into a larger U-domain in the K562 and GM06990 cell lines. Note that the other 3 N-domains (,, and ) are replication timing U-domains common to BG02, K562 and GM06990. To detect U-domains in replication timing profiles at genome scale, we developed a wavelet-based method (Material and Methods, and Supplementary Text S1) which allowed us to identify in the 7 human cell lines from 664 (TL010) up to 1534 (BG02) U-domains of mean size ranging from 0.966 Mb (HeLa R2) up to 1.62 Mb (TL010) and covering from 39.6% (TL010) to 61.9% (BG02) of the genome (Table 2). For each cell line, the average MRT profile of U-domains has an expected parabolic shape (Fig. 2A) representative of individual U-domains (Fig. 2C and Supplementary Figs. S1A–S9A). Inside the U-domains, the derivative is N-shaped (Fig. 2D and Supplementary Figs. S1B–S9B) like the skew profile inside N-domains (Supplementary Figs. S1F–S9F). When rescaling the size of each U-domains to unity for a given cell line, these profiles superimpose onto a common N-shaped curve well approximated by the average profile (Fig. 2B).
To determine the amounts of U-domains conserved in different cell types, we computed for each cell type pair the mutual covering of the corresponding sets of U-domains (two U-domains are shared by two different cell lines if each domain covers more than of the other domain (Table 3)). Taking as reference the matching obtained for the two BJ (68.6% and 74.3%) and HeLa (51.8% and 54.6%) cell replicates, the matchings between the other cell lines were statistically significant and comparable (from 40% to 65% for the mutual covering of lymphoblastoid cell lines). The number of U-domain shared by cell type pairs were all significantly larger than the number expected by chance (, Supplementary Table S2). For example BG02 shares 197 and 189 U-domains with K562 and GM06990 respectively, when only and are expected by chance (Supplementary Table S3). This corresponds to a significant proportion () of the U-domains of the individual cell lines (Table 3), as compared to the matchings () expected by chance (Supplementary Table S4). A significant percentage of N-domains correspond to U-domains (e.g. from 12.5% in BJ R1 up to 23.7% in BG02). This explains that when representing the MRT profile of BG02 instead of the skew , along the set of N-domains ordered according to their size, we can recognize the edges of many N-domains (Supplementary Figs. S1D–S9D). The same observation can be made when comparing the profiles (Supplementary Figs. S1E–S9E) to the corresponding skew profiles (Supplementary Fig. S1F). Note that the N-domains match only of the U-domains of various cell lines due to the very stringent N-domain selection criteria [19], [20] that yielded only 663 N-domains (29.2% of the genome) as compared to much larger U-domain numbers ( of the genome; Table 2). Replication timing U-domains are robustly observed in all cell lines, covering of the human genome. For each cell type, about half U-domains are shared by at least another cell line, namely BG02 (38.4%), K562 (61%), GM06990 (59.2%), BJ R1 (51.6%), HeLa R1 (44.7%). This is also true for the skew N-domains (50.2%) that likely correspond to replication timing U-domains in the germline. However about half of the genome that is covered by U-domains corresponds to regions of high replication timing plasticity where replication domains may (i) reorganize according to the so-called “consolidation” scenario (merging of two U-domains into a larger one) (Fig. 1C), (ii) experience some boundary shift and (iii) emerge in a late replicating region as previously observed in the mouse genome during differentiation [40].
Genome-wide investigation of chromatin architecture has revealed that, at large scales (from 100 kb to 1 Mb), regions enriched in open chromatin fibers correlate with regions of high gene density [41]. Moreover there is a growing body of evidence that transcription factors are regulators of origin activation (reviewed in Kohzaki and Murakami 2005). We ask whether the remarkable genome organization observed around N-domain borders [19] is maintained around replication timing U-domain borders and to what extent it is mediated by a particular chromatin structure favorable to early replication origin specification [22].
When mapping DNase I sensitivity data (Material and Methods) [42] on the U-domains, we observed that the mean coverage is maximal at U-domain extremities and decreases significantly from the extremities to the center that is rather insensitive to DNase I cleavage (Fig. 3A and Supplementary Fig. S10). This decrease, from values significantly higher than the genome-wide average value, extends over 150 kb, whatever the size of the replication timing U-domain (Supplementary Fig. S11A–C) suggesting that, for all examined cell lines, early replicating U-domains borders are at the center of kb wide open chromatin regions. We observed a significant anti-correlation between DNase I cleavage sensitivity data and replication timing data in BG02 (DNase H1-hESC: , ), K562 (, ) and GM06990 (, ) cell lines as well as in the other four cell lines (data not shown; note that this was still observed when controlling for the GC content). This is further supported by open over input chromatin ratio data obtained from human lymphoblastoid cells [41]. We observed that the regions presenting an open/input ratio also decreased significantly (3-fold) from U-domain borders to centers (Fig. 3B).
Cytosine DNA methylation is a mediator of gene silencing in repressed heterochromatic regions, while in potentially active open chromatin regions, DNA is essentially unmethylated [43]. DNA methylation is continuously distributed over mammalian chromosomes with the notable exception of CpG islands (CGIs) and in turn of certain CpG rich promoters and transcription start sites (TSSs). Along the observation that the hypomethylation level of CGIs extends to about 1 kb in flanking regions, we used 1 kb-enlarged CGI coverage as an hypomethylation marker (Material and Methods) [22]. When averaging over the U-domains detected in BG02, we robustly observed a maximum of CGI coverage at U-domain borders as the signature of hypomethylation and a decrease over a characteristic distance of kb (Fig. 3C), similar to what we found for DNase I sensitivity coverage (Fig. 3A). This contrasts with the GC-content profile that strongly depends on the U-domain size and decreases very slowly toward the U-domain center without exhibiting any characteristic scale (Supplementary Fig. S11D–F). These observations are consistent with the hypothesis that early replication origins at U-domain borders are associated with CGIs that are possibly protected from methylation by colocalization with replication origins [44].
Open chromatin markers have been associated with genes. For example 16% of all DNase I hypersensitive sites (HS) are in the first exon or at the TSS of a gene and 42% are found inside a gene [45]. Also, more than 90% of broadly expressed housekeeping genes have a CpG-rich promoter [46]. Remarkably, the mean profiles of Pol II binding Chip-Seq tag density (Material and Methods) along U-domains detected in BG02, K562 and GM06990 cell lines strongly decay over kb away from U-domain borders (Fig. 3D). This indicates that, whatever the cell line, the open chromatin regions around replication U-domains are prone to transcription whereas U-domain central regions appear, on average, transcriptionally silent.
Importantly, we have reproduced the analyses of open chromatin markers near U-domain borders that do not match with a N-domain border (at 100 kb resolution) and confirmed that the results reported in Fig. 3 apply to the initiation zones at U-domains borders of every cell line (Supplementary Fig. S12).
It is widely recognized that the 3D chromatin tertiary structure provides some understanding to the experimental observation of the so-called replicon and replication foci [2], [47]. In particular, replicon size, which is dictated by the spacing between active origins, correlates with the length of chromatin loops [8], [47], [48]. The chromosome conformation capture technique [38] has provided access to long-range chromatin interactions as a footprint of the different levels of chromatin folding in relation with gene activity and the functional state of the cell. From a comparative analysis of replication timing data and Hi-C data correlation matrix in the human genome, some dichotomic picture has been proposed where early and late replicating loci occur in separated compartments of open and closed chromatin respectively [34], [38]. Here, instead of considering the partitioning of the chromosomes derived from all intrachromosomal interactions of each locus (using a principal component of the principal component analysis of the Hi-C data over each chromosome), we focused on interactions between loci separated by short genomic distances ( Mb) over which the contact probabilities are the highest [38]. First, we performed this zoom in the Hi-C contact matrix in the K562 cell line at the 100 kb resolution (Material and Methods) for the 11.4 Mb fragment of human chromosome 10 which contains four U-domains in K562 (Fig. 1; , , and ). We found that these four U-domains remarkably correspond to four matrix square-blocks of enriched interactions (Fig. 4A). Hence, we recover that early replicating zones that border a U-domain (e.g. and separated by 3.9 Mb), have a high contact probability as the signature of 3D spatial proximity. However, we also observe a high contact probability of the two early replicating borders with the late replicating U-domain center and interactions appear sparse for loci in separate U-domains (e.g. and separated by 3.6 Mb). Further examination of the average behavior of intrachromosomal contact probability as a function of genomic distance for the complete genome corroborates these observations. We found that the mean number of interactions between two 100 kb loci of the same U-domain decays when increasing their distance as observed genome-wide (Fig. 4B). Importantly, the mean number of pairwise interactions is significantly higher inside the U-domains than genome-wide and this seems to depend on the U-domain length. In particular, we found that the smaller the domain, the higher the mean number of interactions which is probably a signature of a more open chromatin structure. When comparing the contact probability between two loci inside a U-domain or lying in neighboring U-domains (Fig. 4C), we observed that the latter is higher than the former for distances smaller than the characteristic size ( kb) of the open chromatin structure at U-domain borders (Fig. 3). Above this characteristic distance, the tendency is reversed and the ratio increases up to 2 for distances Mb (Fig. 4C). These data suggest that the segmentation of the genome into replication timing U-domains corresponds to some spatial compartmentalization into self-interacting structural chromatin units insulated by two boundaries of open, accessible, actively transcribed chromatin. This conclusion is strengthened by the observation that U-domain borders are significantly enriched in the insulator binding protein CTCF (Fig. 5), that is known to be involved in chromatin loop formation conditioning communication between transcriptional regulatory elements [36], [37], [49], [50]. Quantitatively similar results were obtained for the lymphoblastoid GM06990 cell line for which both replication timing and Hi-C data were available (Supplementary Fig. S13).
The mapping of open chromatin marks along U-domains revealed that they are bordered by early replication initiation zones likely specified by a kb wide region of accessible, open chromatin permissive to transcription. Such a strong gradient of open chromatin environment was not observed around a large fraction of the 283 replication origins identified in ENCODE regions [12]; only 29% overlap a DNase I hypersensitivity site and half of them do not present open chromatin marks and are not associated with active transcription [22]. Furthermore, the typical inter-origin distance in human cells is 50–100 kb [12], [4], a much smaller value than the mean U-domain size (1–1.5 Mb). These data can be reconciled in a model [51], [52] where replication origins fire independently and their properties (intrinsic firing time probability, efficiency) are specified by the chromatin state: efficient early replicating origins in euchromatic regions (U-domains borders) and late replicating or less efficient origins in heterochromatic regions (U-domains centers). A more dynamical model can also be proposed in which replication first initiates at U-domain borders followed by a chromatin gradient-mediated succession of secondary origin activations. These origins may be remotely activated by the approach of a center-oriented fork that may stimulate initiation due to changes in DNA supercoiling in front of the fork or to association of chromatin remodelers or origin triggering factors with replication fork proteins [35]. This “domino” model could explain why replication progresses from U-domain borders much faster (3–5 times) than the known speed of single fork [8], [35], [48]. Indeed the U-shape of the replication timing profile indicates that the replication wave accelerates (effective velocity equals the inverse of the replication timing derivative, Equation (2)) as the signature of an increasing origin firing frequency during the S-phase [53]. It will be essential to determine to what extent the chromatin state influences fork progression and origins activations and whether outside of U-domains, the genome replicates according to a similar or completely different scenario.
We use the formalism of Markov processes to prove that replication-associated asymmetries between the substitution rates of the two DNA strands induce, in the limit of small asymmetries, a nucleotide compositional skew proportional to the replication fork polarity (the average direction of a locus' replication). Models of DNA composition evolution are usually written in the form of an autonomous and homogeneous system of first-order differential equations [54]:(3)where is the vector which represents the state of the system, i.e. for , is the frequency of at time , and for , is the substitution rate of . A general and well-known property of a Markov process like Equation (3) is that tends exponentially towards the equilibrium value , defined as . The evolution on the complementary strand is given by the same equation but for and , defines the frequency vector on the complementary strand, is the substitution rate matrix on the complementary strand, and denotes the complementary base of . Under no-strand-bias conditions [55], the same substitution rates affect the two strands, i.e. leading to the so-called parity rule of type 2 (PR2): and [56]–[59]. Departure from this symmetry condition can thus be quantified by decomposing into symmetric and antisymmetric parts, the latter accounting for the establishment of a nucleotide compositional strand asymmetry during evolution.
According to our previous studies of the skew in mammalian genomes [16]–[20],[29]–[31], we can reasonably suppose that replication and transcription are the main mechanisms responsible for deviations in PR2. If we concentrate on the effect of replication on DNA composition, we may consider intergenic regions only: then the substitution rate matrix can be written as(4)where is a substitution rate matrix satisfying the no-strand bias conditions (), is the substitution rate matrix associated with replication and (resp. ) the proportion of forks replicating the region of interest in the (resp. ) direction. can be easily decomposed into a symmetric part:(5)and an antisymmetric part:(6)which turns out to be proportional to the fork polarity:(7)
Under the assumption that is significantly smaller than , namely(8)we can use perturbation theory to solve Equation (3) and to show that if the compositional skews:(9)are initially null (), then the total skew will be proportional to the fork polarity at all times up to terms of order (Equation (8)):(10)where is a function that depends only on and . Using the mean nucleotide substitution rate matrix computed in the intergenic regions on each side (300 kb windows) of the -upward jumps [27], the coefficients of were found to be much smaller than those of with (Supplementary Text S1). Thus, according to Equation (10), the observed linear decrease of the skew in N-domains from positive ( end) to negative ( end) values likely reflects the progressive linear decrease of the replication fork polarity with a change of sign in the middle of the skew N-domains. These results provide strong support to the interpretation of skew N-domains (Fig. 1A) as independent replication units in germline cells.
As previously pointed out in [52], the derivative of the replication timing profile does not provide a direct estimator of the replication fork velocity as it also depends on the fork polarity. Here, we demonstrate that the replication fork polarity can be directly deduced from replication timing data under the central hypothesis that the replication fork speed is constant and that replication is bidirectional from each origin. For a given cell cycle, let be the number of activated origins, their positions along the genome and their initiation times. Then the configuration (where and when the origins of replication fire during the S-phase) completely specifies the spatio-temporal replication program (Fig. 6) [51], [52]. If we denote the event “the fork coming form meets the fork coming from ” whose space-time coordinates are:(11)then the replication timing and fork orientation at spatial position are given by (Fig. 6):(12)We clearly see that since then the fork orientation is equal to times the derivative of the replication timing:(13)Under the hypothesis of constant fork velocity , this relationship holds in whole generality in each cell cycle and at every locus without any specific asumption on the distribution of initiation events. By definition, the replication fork polarity is the population average over cell cycles of the fork orientation: . Hence, when averaging over cell cycles, Equation (13) yields:(14)where we have used the fact that the spatial derivative commutes with the population average and that by definition . The replication fork polarity therefore provides a direct link between the skew and the derivative of the MRT (Equations (10) and (14)) in germline cells.
Sequence and annotation data were retrieved from the Genome Browsers of the University of California Santa Cruz (UCSC) [60]. Analyses were performed using the human genome assembly of March 2006 (NCBI36 or hg18). As human gene coordinates, we used the UCSC Known Genes table. When several genes presenting the same orientation overlapped, they were merged into one gene whose coordinates corresponded to the union of all the overlapping gene coordinates, resulting in 23818 distinct genes. We used CpG islands (CGIs) annotation provided in UCSC table “cpgIslandExt”.
The coordinates of the 678 human replication N-domains for assembly NCBI35/hg17 were obtained from the authors [19] and mapped using LiftOver to hg18 coordinates; we kept only the 663 N-domains that had the same size after conversion.
We determined the mean replication timing profiles along the complete human genome using Repli-Seq data [23], [26] (Supplementary Text S1, and Supplementary Fig. S14). For embryonic stem cell line (BG02), three lymphoblastoid cell lines (GM06990, H0287, TL010), a fibroblast cell line (BJ, replicates R1 and R2), and erythroid K562 cell line, Repli-Seq tags for 6 FACS fractions were downloaded from the NCBI SRA website (Studies accession: SPR0013933) [26]. For the HeLa cell line we computed the mean replication timing (MRT) instead of computing the S50 (median replication timing) as in [23].
We developed a segmentation method of the MRT profile into U-domains based on the continuous wavelet transform. This method amounts to perform objective (U-) pattern recognition in 1D signals where the U-motif is picked out from the background signal variations (Supplementary Text S1, and Supplementary Fig. S15).
For the analysis of correlations, we reported the Pearson's product moment correlation coefficient and the associated P-value for no association (). All statistical computations were performed using the R software (http://www.r-project.org/).
We used the DNaseI sensitivity measured genome-wide [42]. Data corresponding to Release 3 (Jan 2010) of the ENCODE UW DNaseI HS track, were downloaded from the UCSC FTP site: ftp://hgdownload.cse.ucsc.edu/goldenPath/hg18/encodeDCC/wgEncodeUwDnaseSeq/.
We plotted the coverage by DNase Hypersentive Sites (DHSs) identified as signal peaks at a false discovery rate threshold of 0.5% within hypersensitive zones delineated using the HotSpot algorithm (“wgEncodeUwDnaseSeqPeaks” tables). When several replicates were available, data were merged.
We used ChIP-seq data using antibody for Pol II and CTCF from Release 3 (Mar 2010) of the ENCODE Open Chromatin track [11], [61]. Data were downloaded from the UCSC FTP site: ftp://hgdownload.cse.ucsc.edu/goldenPath/hg18/encodeDCC/wgEncodeChromatinMap.
We plotted coverage by regions of enriched signal in ChIP experiments, called based on signals created using F-Seq [62] (“wgEncodeUtaChIPseqPeaks” tables). Significant regions were determined at an approximately 95% sensitivity level. We always used the most recent version of data.
We used the spatial proximity maps of the human genome generated using Hi-C method [38]. We downloaded 100 kb resolution maps for GM06990 and K562 cell lines from the GEO web site (GSE18199_binned_heatmaps): http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE18199.
Open over input chromatin ratio data from human lymphobastoid cells were obtained from the authors [41].
Coordinates of N-domains and U-domains in the investigated 7 cell lines can be downloaded from: http://perso.ens-lyon.fr/benjamin.audit/ReplicationDomainsPLoSComputBiol2012/.
|
10.1371/journal.pcbi.1002469 | The Impact of Imitation on Vaccination Behavior in Social Contact Networks | Previous game-theoretic studies of vaccination behavior typically have often assumed that populations are homogeneously mixed and that individuals are fully rational. In reality, there is heterogeneity in the number of contacts per individual, and individuals tend to imitate others who appear to have adopted successful strategies. Here, we use network-based mathematical models to study the effects of both imitation behavior and contact heterogeneity on vaccination coverage and disease dynamics. We integrate contact network epidemiological models with a framework for decision-making, within which individuals make their decisions either based purely on payoff maximization or by imitating the vaccination behavior of a social contact. Simulations suggest that when the cost of vaccination is high imitation behavior may decrease vaccination coverage. However, when the cost of vaccination is small relative to that of infection, imitation behavior increases vaccination coverage, but, surprisingly, also increases the magnitude of epidemics through the clustering of non-vaccinators within the network. Thus, imitation behavior may impede the eradication of infectious diseases. Calculations that ignore behavioral clustering caused by imitation may significantly underestimate the levels of vaccination coverage required to attain herd immunity.
| Both infectious diseases and behavioral traits can spread via social contacts. Using network-based mathematical models, our study addresses the interplay between these two processes, as disease spreads through a population and individuals copy their social contacts when making vaccination decisions. Imitation can produce clusters of non-vaccinating, susceptible individuals that facilitate relatively large outbreaks of infectious diseases despite high overall vaccination coverage. This may explain, for example, recent measles outbreaks observed in many countries with universal measles vaccination policies. Given that vaccine decisions are likely to be influenced by social contacts and that such imitation can have detrimental epidemiological effects, it is important that policy makers understand its causes, magnitude and implications for disease eradication.
| Vaccination is the primary public health measure for preventing transmission of infectious diseases as well as reducing morbidity and mortality from infections [1]. An individual's decision-making with respect to vaccination may depend on perceived risk of infection, cost of infection, cost of vaccination, and the vaccinating behaviors of other individuals [2], [3], [4]. Game theory has been integrated into epidemiological models to investigate vaccination behaviors [5], [6], [7], [8]. Previous game-theoretic studies on vaccination dynamics typically assume that the population is homogeneously mixed and fully rational, defined as making decisions that yield the highest personal utility based on their perceived risks. In reality, there is individual heterogeneity in the number of contacts [9]–[22] and individuals frequently imitate behaviors of their contacts [23], [24], particularly those who appear to have adopted successful strategies [25], [26]. In addition, peer influence is a significant determinant of vaccine uptake in many populations [27].
An imitation vaccination model was previously developed under the assumption of a homogeneously mixed population [25], [28]. This model predicts that imitation is likely to generate oscillations in vaccine uptake, and that the oscillations tend to be large when the perceived risk of vaccination is high [25]. Since this model assumes that the population is homogeneously mixed, it cannot capture clustering of vaccination behaviors in a social network. The clustering of vaccination opinions can exacerbate disease outbreaks by interfering with herd immunity [29], [30], [31].
To evaluate the effect of imitation dynamics on vaccination and disease outbreaks, we develop social network models with imitation behavior. We consider three different contact network structures, a contact network based on a prior study of contact patterns within Vancouver [9], a relative homogeneous network with a Poisson degree distribution, and a heterogeneous scale free network (with a power law degree distribution). We assume that a portion of the population adopts vaccination based on a “payoff maximization” strategy that maximizes their perceived payoff, and the remaining population imitates the vaccination choices of their neighbors.
For all three networks considered, we find that imitation behavior increases the equilibrium level of vaccination coverage when vaccines are inexpensive and decreases vaccination coverage when vaccines are expensive. However, when imitation increases vaccination coverage, it simultaneously leads to connected clusters of unvaccinated individuals, which increase disease prevalence. The emergence of susceptible clusters and its detrimental epidemiological effects are most prominent when vaccination coverage is close to the herd immunity threshold.
We consider a social contact network where individuals can switch between decisions of vaccinating or not vaccinating. An individual's vaccination decision is a function of both the strategies their neighbors have adopted and the perceived benefits of vaccination. Individuals know only the vaccination opinions of immediate neighbors (i.e., whether they are in favor of or opposed to vaccinating), and update their strategies either by imitating one of their neighbors (i.e., following their opinion) or by maximizing their perceived benefits. The fraction of individuals with imitation behavior is indicated by , and the remaining individuals () follow a payoff maximization strategy. The population opinion configuration is denoted by , where indicates the vaccination opinion of individual i,(1)Let be the perceived payoff of an individual i with opinion , then(2)(3)where CV is the individual's cost of vaccinating, CI is the individual's cost of infection, and is perceived probability of infection. The payoffs are negative because maximizing a payoff in this context means minimizing a negative health cost/impact. Let be the relative cost of vaccination (cost of vaccination/cost of infection), . Without loss of generality, we can rewrite Eqs. 2 and 3 as:(4)(5)Let be the perceived probability of contracting the disease from an infectious neighbor at a given time step and be the number of non-vaccinators in the neighborhood of i, respectively. We assumed that the perceived probability of infection depends on the number of non-vaccinator neighbors (whose status as non-vaccinators is assumed known), with no correlation between degree and the number of non-vaccinator neighbors. From basic probability theory we can express as(6)where denotes the perceived probability that an unvaccinated neighbor will not become infected.
A payoff maximizer i will vaccinate if and will not vaccinate if . When , an individual i will adopt the vaccinator or non-vaccinator strategy with equal probability (50%). If the entire population adopts payoff maximization strategy, the system is expected to settle on the Nash equilibrium at steady state.
An imitator i randomly chooses a neighbor (‘role model’) j to imitate. Imitator i adopts j's vaccination decision according to predetermined rules:
We compare vaccination dynamics across three different classes of networks: a pseudo-empirical urban network based on contact patterns within Vancouver, Canada [9], a homogeneous random network with a Poisson degree distribution, parameterized so that the average degree is equal to that of the urban network, and a highly heterogeneous, power law network in which degrees follow a truncated power law distribution. Let denote the probability that a randomly selected individual in a network has degree k. The Poisson network is given by with mean contact number ; the power law network is given by with mean contact number of 4.5 (Figure 2). We calibrate epidemic parameters to ensure that infection risk in an unvaccinated population is equal across all network structures [33], [34]. More precisely, we calibrate the value of disease transmission probability to ensure that the average final epidemic size is equal across the population structures. We chose the final size to be equal to 90%, although results were found to be qualitatively robust for a range of final sizes. For each network, the population size N was equal to 5000. The contact networks are generated using the configuration model (CM) algorithm for constructing finite random networks with a specified degree sequence [35], [36]. We generated degree sequences by choosing random deviates from these degree distributions.
To investigate the effect of imitation behavior on vaccination and disease outbreaks, we assumed that the perceived transmission probability is equal to the transmission probability of the infectious disease.
We perform Monte Carlo simulations on vaccinating opinion formation and disease transmission according to these four steps:
We found that imitation () tends to increase vaccination coverage when the cost of vaccination (r) is low and to decrease vaccination coverage when r is high (Figure 3). However, the effect of imitation varies with the degree of distribution of the contact network (Figure 3). Comparing two extreme cases of (: fully payoff maximization and : fully imitation), imitation dynamics () can promote near-universal coverage when the cost of vaccination is very low compared to that of infection (small values of r) (Figure 3). This difference is particularly pronounced in the power law network.
Individuals have a high incentive to vaccinate when the relative cost of vaccination (r) is low, and not to vaccinate when it is high. Moreover, when most of an individual's neighbors adopt a given strategy, an individual has more incentive to adopt the opposite strategy. That is, if an individual is surrounded by vaccinators, their risk of infection and resulting incentive to vaccinate will both be low; if an individual is surrounded by non-vaccinators, their risk of infection and incentive to vaccinate will be high. However, imitators have a non-zero probability of copying the vaccination strategy that is adopted by most of their neighbors, even when such a strategy may be less suboptimal for them. For low values of r, payoff maximizers have a high incentive to vaccinate, and thus imitators are likely to have vaccinators as role models; the opposite should be true under high values of r. Therefore, for low values of r, imitators may have higher vaccination coverage than payoff maximizers, whereas for high values of r, imitation may lead to fewer vaccinators than anticipated by payoff maximization strategy (Figures 3 and 4).
The power law network was shown to be more sensitive to the effect of imitation behavior than Poisson and Urban networks (Figures 3 and 4). This is due to the fact that the power law network has a highly skewed degree distribution, with a small density of highly-connected individuals. Highly-connected individuals (hubs) have a high incentive to vaccinate, whereas individuals with few contacts have less incentive to vaccinate. By imitating their highly-connected neighbors, individuals with few contacts become more likely to vaccinate, which may substantially increase vaccination coverage (Figures 3 and 5). However, this increase of vaccination coverage overall decreases the incentive for hubs to vaccinate (Figure 5). Depending on the density of hubs and the value of the relative cost of vaccination, this decrease in the incentive of hubs to vaccinate may reduce the total vaccination coverage within the population (Figure 3).
We found that imitation () increases the final size of the outbreak (i.e., the fraction of the population infected) for intermediate costs of vaccination (r) (Figure 6). Numerical investigation showed that this range of values, which varies with the contact network (Figure 6), has an upper bound that represents the value of r above which it is disadvantageous for anyone to vaccinate, resulting in a full blown epidemic, and a lower bound which represents the value of r below which the average final epidemic size was less than twice the size of the initial inoculum of 10 infected individuals. Imitation dynamics can increase the vaccination coverage relative to a population with payoff maximization strategy, when the cost of vaccination is low, but can never decrease the final epidemic size (Figures 3 and 6). As a result of behavioral clustering that emerges from imitation dynamics, the size of the epidemic does not necessarily decrease as vaccination coverage increases. That is, vaccinators tend to contact vaccinators, and non-vaccinators tend to contact non-vaccinators (Figure 7). Because herd immunity is considerably high in these pockets of vaccinators, further vaccination within these pockets reduces transmission to a lesser degree than if vaccination were increased in regions of the network with relatively low vaccination coverage. The clusters of non-vaccinators fuel transmission and increase the probability of an outbreak. This effect of imitation is most prominent when vaccination coverage is close to the herd immunity threshold (Figures 3 and 6).
To investigate the sensitivity of our results to the degree to which imitators respond to payoff differences between themselves and their neighbors, we compared weak responsiveness to strong responsiveness (Figure 2). For strong responsiveness, individuals reliably copy the strategy of successful neighbors. However, if most neighbors of an imitator adopt a given strategy, then the opposite strategy becomes advantageous, and the imitator would be more likely to choose the opposite strategy. Strong responsiveness, relative to weak responsiveness, leads imitators to rarely copy unsuccessful neighbors (Figure 2). Therefore, as the degree of responsiveness increases (α = 15), vaccination coverage under pure imitation () tends to converge towards the vaccination level predicted by the payoff maximization equilibrium (Figures 3 and 8). A similar convergence occurs for final epidemic size (result not shown here).
Classic economic theory has not considered the reality that individuals frequently imitate others [2], [3], [24], [31], [37]. Imitation begins with simple behaviors in infancy and evolves into more complex behaviors in childhood and adulthood [31], [37]. In the context of epidemiology, imitation behavior can influence vaccination patterns and thus the dynamics of disease outbreaks [2], [3], [24]. In this work, we address the impact of imitation on vaccination coverage, disease prevalence, and the herd immunity threshold. We develop a model that allows contact patterns to be heterogeneous and individuals to incorporate varying degrees of imitation into decision-making. Individuals within a social contact network can switch between the strategies of vaccinating and not vaccinating. An individual's decision regarding whether to vaccinate is affected by the strategies that their neighbors have adopted or the perceived net benefits of vaccination. Monte Carlo simulations show that imitation dynamics increase the equilibrium vaccination coverage when vaccination cost is relatively low and may decrease vaccination coverage when vaccination is costly. In both cases, imitation actually exacerbates disease transmission when vaccination is inexpensive through the social clustering of non-vaccinators. The detrimental effects of imitation are most prominent when the vaccination is close to the herd immunity threshold.
Salathe and Bonhoeffer recently developed a vaccination opinion formation model to reveal that opinion clustering increases the size of an epidemic [29]. Their model assumed that opinions are determined by the proportion of neighbors that have the same opinion about vaccination, such that whenever an individual switches opinion, another individual has to switch the opinion in an opposite way in order to maintain constant vaccination coverage level [29]. Extending this previous seminal model, we consider both imitation (opinion formation) and payoff maximization consideration (individuals are not just blindly imitating neighbors; they are trying to optimize a payoff function). Our model thereby recognizes that vaccine decision-making is not a purely imitative process, and often depends on actual health considerations. Additionally, by incorporating payoffs, we are able to analyze the impact of vaccine cost on the dynamics of vaccination.
This analysis takes an initial step towards understanding the combined impacts of payoff maximization and imitative decision-making on vaccination coverage and epidemiological dynamics. The model, however, rests on several simplifying assumptions. For example, the contact networks are assumed to remain static throughout the epidemic, and to be identical for both disease and behavioral transmission. These assumptions could be relaxed by incorporating temporal changes in network structure [38], and modeling multiple different edge types (e.g. individual variation in susceptibility and infectivity) [39]. The model can also be extended to allow individuals to follow mixed vaccination strategies, or by incorporating the effects of past epidemics on vaccine decision-making [40].
We find that imitation leads to clustering of susceptible individuals, which may exacerbate outbreaks of infectious diseases. For example, imitation may explain how outbreaks of measles have occurred in countries with high overall vaccination coverage [26], [29], [41], [42]. Given that vaccine decisions are likely to be influenced by social contacts [29], [32], [41] and that such imitation can have detrimental epidemiological effects [29], it is important that policy makers understand its causes, magnitude, and implications for disease elimination.
Our findings indicate that the common assumptions of simple payoff maximization and homogeneous mixing can lead to misestimates of the level of vaccination coverage necessary to control a disease outbreak. Our model provides a general framework for investigating the effect of imitation on vaccination decision-making and disease outbreaks. The model can be applied to study the interactions between behavior, public health, and epidemic dynamics for specific infectious diseases. Data describing real world imitation behavior in vaccination decision-making will be critical to future public health applications of the model.
|
10.1371/journal.ppat.1002918 | Very Long O-antigen Chains Enhance Fitness during Salmonella-induced Colitis by Increasing Bile Resistance | Intestinal inflammation changes the luminal habitat for microbes through mechanisms that have not been fully resolved. We noticed that the FepE regulator of very long O-antigen chain assembly in the enteric pathogen Salmonella enterica serotype Typhimurium (S. Typhimurium) conferred a luminal fitness advantage in the mouse colitis model. However, a fepE mutant was not defective for survival in tissue, resistance to complement or resistance to polymyxin B. We performed metabolite profiling to identify changes in the luminal habitat that accompany S. Typhimurium-induced colitis. This analysis suggested that S. Typhimurium-induced colitis increased the luminal concentrations of total bile acids. A mutation in fepE significantly reduced the minimal inhibitory concentration (MIC) of S. Typhimurium for bile acids in vitro. Oral administration of the bile acid sequestrant cholestyramine resin lowered the concentrations of total bile acids in colon contents during S. Typhimurium infection and significantly reduced the luminal fitness advantage conferred by the fepE gene in the mouse colitis model. Collectively, these data suggested that very long O-antigen chains function in bile acid resistance of S. Typhimurium, a property conferring a fitness advantage during luminal growth in the inflamed intestine.
| Microbial communities inhabiting the intestinal tract have been characterized using high-throughput sequencing approaches, but little is known about factors that change their luminal habitat. Using metabolite profiling we discovered that luminal concentrations of total bile acids become significantly elevated during Salmonella-induced colitis. Very long O-antigen chains of Salmonella enterica serotype Typhimurium (S. Typhimurium) were required for luminal growth in the presence of such elevated bile concentrations during colitis. In contrast, very long O-antigen chains were dispensable for luminal growth in the presence of normal concentrations of bile acids or for growth in organs of mice. Thus, elevated bile concentrations during Salmonella-induced colitis change the luminal habitat and microbial growth in this environment requires increased bile resistance, which in case of S. Typhimurium is conferred by very long O-antigen chains.
| Salmonella enterica serotype Typhimurium (S. Typhimurium) is an important cause of human gastroenteritis [1]. Upon ingestion, a fraction of the S. Typhimurium population enters intestinal epithelial cells using the invasion-associated type III secretion system (T3SS-1) [2], which is followed by macrophage survival mediated by a second type III secretion system (T3SS-2) [3]. The deployment of T3SS-1 and T3SS-2 triggers acute intestinal inflammation [4] and the resulting changes in the environment enhance growth of the luminal fraction of the S. Typhimurium population [5], [6], [7], [8] despite the fact that concentrations of antimicrobial substances are elevated in this habitat [9], [10], [11].
The surface of S. Typhimurium, which is exposed to antimicrobial substances in the inflamed gut, is decorated with lipopolysaccharide (LPS). The oligosaccharide core connects the lipid A moiety, which anchors the LPS molecule in the outer membrane, to O-antigen repeat units that extend from the bacterial surface. Each O-antigen repeat unit is composed of a trisaccharide backbone, consisting of α-D-mannose-(1,4)-α-L-rhamnose-(1,3)-α-D-galactose, and a branching sugar (abequose) that is α-(1,3)-linked to D-mannose in the backbone. S. Typhimurium can produce short LPS species containing between 1 and 15 O-antigen repeat units. Additionally, S. Typhimurium synthesizes LPS species containing a greater number of O antigen repeat units, which requires the length regulators WzzB [12], [13] and FepE [14]. The WzzB protein regulates the assembly of long LPS species carrying between 16 and 35 O-antigen repeat units [15] while the FepE protein controls the biosynthesis of very long LPS species with more than 100 O-antigen repeat units [14]. These regulatory mechanisms give rise to a tri-modal distribution in LPS length.
There is general agreement that WzzB-dependent assembly of long O-antigen chains (16–35 repeat units) provides a fitness advantage by conferring resistance to complement [14], [16]. In contrast, the function of very long O-antigen chains (>100 repeat units) during host microbe interaction remains unclear. The absence of very long O-antigen species in a fepE mutant reduces in vitro translocation of SipA [16], a T3SS-1 effector protein that significantly contributes to the induction of inflammation in the mouse colitis model [17]. However, in vivo the O-antigen is not required for T3SS-1-dependent induction of intestinal inflammatory responses [18]. These data indicate that SipA translocation proceeds through an O-antigen-independent pathway in vivo. One study suggests that very long O-antigen inhibits macrophage phagocytosis [19], while others observed no inhibition [16]. A fepE mutant and its S. Typhimurium wild-type parent are equally resistant to complement in vitro, although a wzzB fepE mutant is more serum sensitive than a wzzB mutant [14], [16], [20]. Importantly, an S. Typhimurium fepE mutant retains full virulence in the mouse typhoid model [14], suggesting that complement resistance and resistance to phagocyte-mediated killing mechanisms are independent of very long O-antigen chains in vivo. Since no phenotype has been described for a S. Typhimurium fepE mutant in mouse models, it remains unclear which selective forces prevent loss of very long O-antigen chains.
In the absence of selective forces to maintain the costly production of very long O-antigen chains, the fepE gene is predicted to accumulate point mutations and eventually become inactive. Interestingly, fepE is a pseudogene in S. enterica serotype Typhi, the causative agent of typhoid fever [21]. We thus reasoned that fepE might be involved in aspects of host pathogen interaction that are important during gastroenteritis (caused by S. Typhimurium) but dispensable during typhoid fever (caused by S. Typhi). While S. Typhimurium transmission requires maximum growth in the lumen of the inflamed intestine [22], S. Typhi persists in the human population through chronic gall bladder carriage [23], [24]. Based on these considerations, we proposed that very long O-antigen chains might enhance the fitness of S. Typhimurium in the environment of the inflamed intestine during gastroenteritis. We tested this hypothesis using the mouse colitis model and elucidated the underlying mechanism.
To study the role of very long O-antigen chains, we generated a fepE mutant of the S. Typhimurium wild type strain IR715. The fepE mutant (RC31) was deficient for producing very long O-antigen chains, and this defect could be restored by introducing the cloned fepE gene (pCR37) on a plasmid (Fig. S1A). Consistent with previous reports [14], [16], the S. Typhimurium wild type (IR715) and the fepE mutant (RC31) exhibited similar levels of serum resistance (Fig. S1B). In contrast, a wzzB mutant that lacked long O-antigen chains but retained the ability to produce very long O-antigen chains (Fig. S1A) exhibited significantly (P<0.01) reduced serum resistance (Fig. S1B).
Genetically resistant mice infected with S. Typhimurium exhibit little intestinal inflammation during the first three days after inoculation (mouse typhoid model) [25]. In contrast, mice treated with streptomycin develop severe acute cecal inflammation within 2 days after infection (mouse colitis model). To determine whether very long O-antigen chains confer a fitness advantage during conditions of acute intestinal inflammation, we performed a pilot experiment in which streptomycin pre-treated, genetically susceptible mice (C57BL/6J) were infected intragastrically with an equal mixture of the S. Typhimurium wild type (IR715) and a fepE mutant (RC31) (mouse colitis model). By day four after infection, mice became moribund and the wild type was recovered in higher numbers than the fepE mutant from intestinal contents, but not from liver and spleen (Fig. S2).
To determine whether the luminal fitness advantage conferred by the fepE gene would increase over time, groups of streptomycin pre-treated, genetically resistant mice (129/SvJ) were infected intragastrically with an equal mixture of the S. Typhimurium wild type (IR715) and a fepE mutant (RC31) (mouse colitis model). For comparison, mice (129/SvJ) that had not been pre-treated with streptomycin were infected with the same mixture of S. Typhimurium strains (mouse typhoid model). In the mouse typhoid model, the wild type did not exhibit a luminal growth advantage over the fepE mutant as indicated by recovery of bacteria from feces (Fig. 1A) or colon contents (Fig. 1B). In contrast, the wild type was recovered in significantly higher numbers (P<0.05) from feces starting at day 3 after infection in the mouse colitis model (Fig. 1A). By day 7 after infection, the wild type was recovered in greater than 100-fold excess over the fepE mutant (Fig. 1B). In both the mouse typhoid model and in the mouse colitis model, the S. Typhimurium wild type and fepE mutant were recovered in similar numbers from the liver and spleen (Fig. 1C), suggesting that very long O-antigen chains are dispensable for growth in tissue. As expected, histopathogical analysis revealed a greater severity of cecal inflammation in the mouse colitis model compared to the mouse typhoid model seven days after infection (Fig. 1 and E). Bacteria recovered from intestinal contents of streptomycin pretreated mice exhibited strong expression of very long O-antigen chains, as indicated by Western blot analysis (Fig. S3).
Production of very long O-antigen chains had a similar effect on bacterial numbers in the mouse colitis model when we used CBA/J mice, a lineage that is also genetically resistant to S. Typhimurium infection (Fig. S4A and S4B). To assess complementation in vivo, streptomycin pre-treated CBA/J mice were infected with an equal mixture of a S. Typhimurium fepE mutant (RC61) and a derivative carrying the fepE gene inserted upstream of the phoN gene (phoN::fepE) (RC62). The complemented strain (RC62) was recovered in higher numbers from feces and colon contents (Fig. S4C), suggesting that an intact chromosomal copy of the fepE gene could restore the fitness defect of a fepE mutant. Next, we investigated whether the fitness advantage conferred by very long O-antigen chains was apparent in mice infected with individual S. Typhimurium strains (single infection design). Streptomycin pre-treated CBA/J mice were infected with either the S. Typhimurium wild type (IR715) or a fepE mutant (RC31) (mouse colitis model). Mice infected with the S. Typhimurium wild type shed significantly (P<0.01) higher bacterial numbers with their feces than mice infected with the fepE mutant (Fig. S4E), while similar bacterial loads were recovered from liver and spleen (Fig. S4D).
T3SS-1 and T3SS-2 cooperate to induce intestinal inflammation in the mouse colitis model [26], [27]. Inactivation of both T3SS-1 (through a mutation in invA) and T3SS-2 (through a mutation in spiB) renders the resulting S. Typhimurium double mutant strain unable to trigger intestinal inflammation [9]. To further investigate why very long O-antigen chains conferred a fitness advantage in the mouse colitis model, streptomycin pre-treated mice (129/SvJ) were infected with an equal mixture of an invA spiB mutant (SPN452) and an invA spiB fepE mutant (RC55). Mice infected with this mixture did not developed intestinal inflammation (Fig. 1A and 1B) and the invA spiB mutant did not exhibit a luminal growth advantage over the invA spiB fepE mutant (Fig. 1A and 1B). Due to the attenuation caused by mutations in invA and spiB, neither strain was recovered from the liver and spleen of mice seven days after infection (Fig. 1C). Collectively, these data pointed to the lumen of the acutely inflamed intestine as the habitat conferring a fitness advantage to S. Typhimurium expressing very long O-antigen chains. However, the mechanism by which very long O-antigen chains would provide a benefit in the environment of the inflamed gut remained unclear.
To identify potential small molecule candidates that might explain the fitness advantage conferred by very long O-antigen chains in the inflamed gut, groups of streptomycin pre-treated mice (n = 4) were inoculated with sterile LB broth (mock infection) or S. Typhimurium. Four days after infection, the ceca were collected and metabolites extracted from the mucosa with water. Metabolites in these cecal washes were analyzed after removing bacteria by filtration to avoid contamination of samples with bacterial intracellular metabolites. Samples underwent hydrophilic interaction chromatography - liquid chromatography/mass spectrometry (HILIC-LC/MS) metabolic profiling. Peak metabolite values were measured for cecal washes of each animal and data were transformed logarithmically (log2) for statistical analysis and determination of false discovery rates. More than 800 components were detected by HILIC-LC/MS. Principle component analysis revealed group clustering of samples from mock-infected mice and S. Typhimurium-infected mice (Fig. S5), suggesting that S. Typhimurium infection was accompanied by characteristic changes in the luminal environment.
We were able to assign metabolite identities to 67 components (Table S1). A total of 20 components met the criteria for at least a 2-fold change between samples from mock-infected and S. Typhimurium-infected mice and a P value of less than 0.1, which is illustrated in a volcano plot in Figure 2A. Of these, 15 components were considered significantly changed (P<0.05; q<0.1) between both groups. We were able to assign metabolite identities to four of the six components that were significantly (P<0.05; q<0.1) increased in samples from S. Typhimurium-infected mice compared to mock-infected animals (Fig. 2B). Interestingly, one of these metabolites, phosphatidylinositol 3-phosphate, is produced in host cells by the T3SS-1 effector protein SopB [28]. Two metabolites were breakdown products of phosphatidylethanolamine (lysophosphatidylethanolamine [14∶1]) and phosphatidylcholine (lysophosphatidylcholine [16∶1]), respectively. The fourth metabolite with significantly increased abundance (P<0.05; q<0.1) in samples from S. Typhimurium infected mice was the bile acid tauromurocholate. These data indicated that increased levels of bile acids might accompany S. Typhimurium infection, thereby identifying a potential candidate for an anti-microbial metabolite that is elevated during colitis.
To validate results from metabolite profiling, the concentration of total bile acids was determined in colon contents recovered from the mouse typhoid model and from the mouse colitis model using an enzymatic assay. The concentration of total bile acids was significantly (P<0.01) increased in the mouse colitis model, compared to the mouse typhoid model or to mock-infected mice (Fig. 2C). These data suggested that the concentration of total bile acids was increased during S. Typhimurium-induced colitis. Furthermore, these data established an inverse correlation between luminal concentrations of total bile acids (Fig. 2C) and the luminal fitness of the fepE mutant (Fig. 1A and 1B) but fell short of establishing cause and effect.
To investigate whether an increased concentration of total bile acids during S. Typhimurium-induced colitis was the mechanism responsible for the fitness advantage conferred by very long O-antigen chains (Fig. 1), we first determined the minimal inhibitory concentration (MIC) of S. Typhimurium strains to ox bile extract (sodium choleate) in vitro (Table 1). The MIC of sodium choleate was identical for the S. Typhimurium wild type (IR715) and a wzzB mutant (RC46). However, compared to the wild type, the MIC of sodium choleate for the fepE mutant (RC31) was reduced. Introduction of the cloned fepE gene (pCR37) into the fepE mutant restored resistance to sodium choleate. Lack of very long O-antigen chains in the fepE mutant did not reduce the MIC for the anionic detergent sodium dodecyl sulfate (SDS) or for the cationic antimicrobial peptide polymyxin B. In contrast, a mutation in pmrE, a gene necessary for 4-aminoarabinose lipid A modification [29], increased sensitivity to the antimicrobial peptide polymyxin B, but not to the anionic detergent SDS. Collectively, these data suggested that very long O-antigen chains, but not long O-antigen chains, are specifically required for full resistance to bile acids.
Since the MIC of the fepE mutant for sodium choleate (2%) was nearly 100-fold higher than the concentration of bile acids measured in intestinal contents (Fig. 2C), we compared growth of the S. Typhimurium wild type and a fepE mutant in medium containing different concentrations of bile acids. As expected, medium containing 3% sodium choleate supported growth of the S. Typhimurium wild type (MIC 12%) but fully suppressed growth of the fepE mutant (MIC 2%) (Fig. S6A). Growth of the fepE mutant was retarded compared to that of the S. Typhimurium wild type in medium containing 0.3% sodium choleate (Fig. S6B) or 0.03% sodium choleate (Fig. S6C). Finally, both strains grew equally in medium containing 0.003% sodium choleate (Fig. S6D). These data suggested that very long O-antigen chains conferred a fitness advantage at concentrations of bile acids that were approximately two orders of magnitude below the MIC of the fepE mutant (Table 1).
The concentration of total bile acids in intestinal contents can be lowered by bile sequestrants, such as cholestyramine resin, a compound used clinically to restrict cholesterol and fat intake. To investigate whether bile sequestrants could rescue growth of the fepE mutant in medium containing sodium choleate, MIC values were determined in vitro in the presence of cholestyramine resin (Table 1). The S. Typhimurium wild type and the fepE mutant exhibited identical MIC values for sodium choleate in the presence of cholestyramine resin.
To test whether increased concentrations of total bile acids are responsible for the fitness advantage conferred by very long O-antigen chains in the mouse colitis model, animals fed rodent chow containing cholestyramine resin or control chow were pretreated with streptomycin and inoculated with sterile LB broth (mock infection) or with an equal mixture of the S. Typhimurium wild type (IR715) and a fepE mutant (RC31). While the S. Typhimurium wild type was recovered in significantly higher numbers from feces (Fig. 3A) and colon contents (Fig. 3B) of mice fed control chow, this competitive advantage was abrogated in mice fed chow containing cholestyramine resin. Concomitantly, diet containing cholestyramine resin significantly (P<0.05) reduced the concentration of total bile acids in colon contents present during S. Typhimurium infection (Fig. 3C) but did not reduce the severity of intestinal inflammation (Fig. 3D and 3E). These data provided direct support for the idea that increased concentrations of total bile acids during S. Typhimurium-induced colitis reduced the fitness of bacteria lacking very long O-antigen chains.
S. Typhimurium-induced colitis is accompanied by elevated luminal concentrations of several antimicrobial proteins, including lipocalin-2 [9], calprotectin [11], regenerating islet-derived 3 gamma (RegIIIγ) [30] and RegIIIβ [10]. In some cases, specific resistance mechanisms against these antimicrobial proteins have been shown to confer a fitness advantage in the lumen of the inflamed gut. For example, the ability to synthesize the siderophore salmochelin confers resistance to lipocalin-2 [31], a property that boosts luminal growth of S. Typhimurium in the mouse colitis model [9]. Similarly, the high affinity zinc transporter ZnuABC confers resistance to the zinc-binding protein calprotectin, thereby enhancing luminal growth of S. Typhimurium during colitis [11].
Here we used metabolite profiling to determine whether small antimicrobial metabolites are elevated during S. Typhimurium-induced colitis. Our results identify bile acids as a class of small antimicrobial molecules whose concentration is elevated in the mouse colitis model. A previous metabolite profiling study detected elevated bile acid biosynthesis in the mouse typhoid model by analyzing acetonitrile extracts from feces [32]. Total concentrations of bile acids in feces and fecal water are elevated two to five times in patients with inflammatory bowel disease [33]. However, fecal concentrations of total bile acids are significantly decreased during dextran sulfate sodium (DSS)-induced colitis in rats [34]. Thus, an increased concentration of bile acids is not likely to be a general feature of intestinal inflammation, but might involve mechanisms that are only triggered in a subset of inflammatory disorders, including S. Typhimurium-induced colitis. It should be noted that due to absorption in the large intestine, bile concentrations might be higher in the ileum, the primary site of S. Typhimurium infection in humans. The mouse is not an ideal model to study luminal growth in the ileum, because the bulk of luminal S. Typhimurium are present in the cecum.
The absence of acute colitis during typhoid fever might explain why fepE is a pseudogene in S. Typhi [21]. However, S. Typhi requires bile resistance for persistence in the gall bladder. It is possible that other long surface carbohydrates, such as the Vi-capsular polysaccharide or the O-antigen capsule [35], [36], can compensate for the lack of very long O-antigen chains during growth in the environment of the gall bladder. Nonetheless, our data show that very long O-antigen chains increased the fitness of S. Typhimurium in the lumen of the acutely inflamed intestine by mediating bile resistance, which represents the first function ascribed to this surface structure. Inactivation of yrbK and rlpB, two genes implicated in LPS transport, increases bile resistance of S. Typhimurium and correlates with differences in the oligosaccharide units that form long-chain LPS [37]. However, our data suggest that long O-antigen chains do not confer resistance to bile. Conversely, long O-antigen chains confer serum resistance, while very long O-antigen chains are dispensable for this trait [14], [16]. Thus, long O-antigen chains and very long O-antigen chains do not have redundant functions.
The picture emerging from these studies is that very long O-antigen is important for growth of S. Typhimurium in the lumen of the inflamed intestine. Maximum growth in the intestinal lumen is of relevance, because it enhances transmission by the fecal-oral route [22], which is predicted to confer a selective advantage. We conclude that the role of very long O-antigen chains in bile resistance likely represents one of the selective forces that help to maintain the fepE gene in the genomes of non-typhoidal Salmonella serotypes.
Bacterial strains and plasmids used in this study are presented in Table 2. Cultures of S. Typhimurium and Escherichia coli were routinely incubated with aeration at 37°C in Luria-Bertani (LB) broth (10 g tryptone, 5 g yeast extract, and 10 g NaCl per liter) or on LB agar plates unless indicated otherwise. Antibiotics were added as necessary at appropriate concentrations: chloramphenicol (Cm), 0.03 mg/ml; carbenicillin (Cb), 0.1 mg/ml; kanamycin (Kan), 0.05 mg/ml; and nalidixic acid (Nal), 0.05 mg/ml.
Nucleotide sequences of oligonucleotide primers used for cloning are listed in Table 3. Internal fragment of S. Typhimurium the fepE and wzzB genes were polymerase chain reaction (PCR) amplified using primer pairs 62/63 and 150/151, respectively, and cloned into pCR2.1 (Invitrogen) to give rise to plasmids pRC30 and pRC48. The inserts were excised using restriction endonuclease digestion with SmaI and SalI (pRC30) or SacII and XhoI (pRC48) and cloned into suicide vector pGP704 digested with SmaI and SalI or into suicide vector pEP185.2 digested with SacII and XhoI to yield plasmids pRC31 and pRC49, respectively. Plasmids pRC31 and pRC49 were introduced into S. Typhimurium IR715 by conjugation with E. coli S17-1 λpir and exconjugants were designated RC31 and RC46, respectively. The wild-type fepE gene was PCR amplified using the primer pair 52/53 and cloned into pCR2.1 to yield plasmid pRC36. The insert of pCR36 was excised with EcoRI and cloned into low copy vector pWSK129 to yield pRC37. To construct a pmrE mutant, an internal fragment of the S. Typhimurium pmrE gene was PCR amplified using primers 263/264, cloned into pCR2.1 to yield plasmid pSW197. The insert of pSW197 was cloned into suicide plasmid pFUSE at XbaI and SmaI restriction sites to yield plasmid pSW198. Plasmid pSW198 was introduced into S. Typhimurium IR715 by conjugation to yield strain SW698. The fepE::pGP704 mutation was transduced from RC31 into the S. Typhimurium invA spiB mutant SPN452 using transduction with phage P22 HT int-105 to yield strain RC55.
To construct a fepE deletion mutant, primers 170 and 171 were used to PCR amplify the region upstream of S. Typhimurium fepE (flanking region 1) from base 3537 of entF to base −48 of fepE. Primers 172 and 173 were used to PCR amplify the downstream region of S. Typhimurium fepE (flanking region 2) from base 1136 of fepE to base 322 of fepC. PCR products for the fepE upstream and downstream regions were gel purified (Qiagen), digested with XbaI, and ligated together with T4 DNA ligase (NEB). The ligation reaction served as a template for PCR amplification of the fepE flanking region construct using primers 170 and 173, and this product was gel purified and cloned into pCR2.1 (Invitrogen) to generate pRC50. The cassette containing fepE flanking regions was excised from pRC50 using SalI, ligated into SalI-digested pRDH10 using the Quick T4 DNA Ligase kit (NEB) to generate pRC51. The KSAC cassette from pBS34 was excised by XbaI digestion and ligated into the XbaI site of pRC51 to yield pRC52. Plasmid pRC52 was conjugated into S. Typhimurium IR715. An exconjugant selected on plates containing LB+Nal+Km was designated RC61 and deletion of fepE was confirmed by PCR. For complementation, the S. Typhimurium wild type fepE allele was PCR amplified using primers 174 and 175 from base −350 of fepE to base 1136 of fepE, and cloned into pCR2.1 to give rise to plasmid pRC53. The insert was excised using restriction endonuclease digestion with SalI and SmaI and ligated into SalI- and SmaI-digested pSW84, a suicide plasmid carrying a DNA fragment upstream of the phoN open reading frame, to yield pRC54. Plasmid pRC54 was introduced into RC61 by conjugation with E. coli S17-1 λpir and an exconjugant carrying pRC54 integrated into the phoN locus was designated RC62.
Bacterial cultures grown 16 hours at 37°C in LB broth were suspended in 1 ml Dulbecco's phosphate buffered saline (DPBS, Gibco) at a concentration of approximately 1×109 CFU/ml. Samples were washed three times in DPBS (12,000× g for 4 minutes). The pellet was resuspended in 0.1 ml lysis buffer (0.5 M Tris-HCl at pH 6.8, 4% sodium dodecyl sulfate [SDS], 40% glycerol, 0.2% bromophenol blue, and 10% β-mercaptoethanol) and boiled for 10 minutes. We then added 0.1 ml of proteinase K solution (Invitrogen) (20 mg proteinase K/ml in 10 mM Tris-HCl, 50% glycerol, and 20 mM CaCl2) and incubated for 3 hours at 65°C. LPS preparations were boiled again for 10 minutes, and 0.02 ml of each sample loaded on a 15% SDS-polyacrylamide gel electrophoresis (SDS-PAGE), and visualized by silver staining as described previously [38].
LPS purified from colon and cecal contents was probed with anti-Salmonella O:4 serum (Becton, Dickinson and Company) for production of the somatic O antigen by Western blot analysis. CFU of S. Typhimurium were determined for each sample and samples loaded in each lane were normalized to represent equivalent amounts of bacteria. Preparations were separated by 15% SDS-PAGE, transferred to an Immobilon PVDF membrane (Millipore) using a Trans-Blot semidry transfer apparatus (Bio-Rad), and detected with anti-Salmonella O:4 serum and an alkaline phosphatase-conjugated goat anti-rabbit secondary antibody (Bio-Rad Laboratories). Bands were visualized following exposure and development on a BioSpectrum AC Imaging System (Ultra-Violet Products).
To determine MIC values of S. Typhimurium strains against polymyxin B (Sigma Aldrich), SDS (USB Corporation), sodium choleate (a crude ox bile extract from Sigma Aldrich, which contains the sodium salts of taurocholic, glycocholic, deoxycholic, and cholic acids), or with a combination of sodium choleate and cholestyramine resin (DOW Chemical Company), overnight cultures were washed twice in DPBS, diluted to 1×103 cells per ml with antimicrobial compounds in sterile LB broth (polymyxin B: 0–1.5 µg/ml, increments of 0.1 µg/ml; SDS and sodium choleate: 0–20% w/v, increments of 2%) and incubated for 15 hours at 37°C with aeration to determine the lowest concentration of each antimicrobial compound that prevented bacterial growth. Each experiment was repeated three times independently.
For analyses of serum sensitivity, 1×107 bacteria from cultures grown to stationary phase were washed three times in DPBS and incubated in 10% human serum for one hour at 37°C with slight agitation. The number of viable bacteria was quantified by spreading serial 10-fold dilutions on LB agar plates containing the appropriate antibiotics.
For competitive infection experiments, groups (n = 4) of female mice (C57BL/6J or 129/svJ Jackson Laboratory) aged 6 to 8 weeks were inoculated intragastrically with 20 mg streptomycin in a volume of 0.2 ml 48 hours prior to intragastric inoculation with either 0.1 ml sterile LB broth or 1×107 CFU of the indicated mixtures of S. Typhimurium strains. Alternatively, animals were inoculated intragastrically with either 0.1 ml sterile LB broth or 1×107 CFU of the indicated mixtures of S. Typhimurium strains without being pretreated with streptomycin. Fecal pellets were collected and organs were harvested at day 4 (C57BL/6J) or day 7 (129/svJ) after infection. Fecal and tissue samples were homogenized in PBS, serially diluted and plated on selective media to enumerate CFU. The competitive index (CI) was calculated by dividing the ratio of wild-type to mutant bacteria in homogenates (output) by the ration present in the inoculum (input). In some experiments, 129/SvJ mice were fed a normal rodent diet (Harlan-Teklad) supplemented with the bile acid sequestrant cholestyramine resin (2%, The Dow Chemical Company) starting the day before streptomycin treatment and continuing throughout the experiment.
For metabolite profiling, groups of 4 mice (C57BL/6J) received 20 mg streptomycin in a 0.1 ml volume intragastrically and were inoculated intragastrically 24 hours later with either 0.1 ml sterile LB broth or 1×109 CFU of S. Typhimurium in LB broth. The cecum was collected four days after infection, contents were removed and metabolites from the mucosal surface were extracted by five washes with 0.2 ml sterile UltraPure distilled water (Gibco). Debris was removed by centrifugation at 20,000× g for 5 minutes, the supernatant was filter sterilized to remove bacteria and submitted to metabolite profiling analysis.
All animal experiments were performed according to Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) guidelines. All animal experiments were approved by the Institutional Animal Care and Use Committee at the University of California, Davis.
Cecal tissues were formalin fixed, sectioned, stained with hematoxylin and eosin (H&E), and submitted to a veterinary pathologist for blinded scoring using a scale described previously [8]. Representative images of tissue sections were taken using an Olympus BX41 microscope.
Reagents and standards: Liquid chromatography/mass spectrometry (LC/MS) grade acetonitrile and acetone (Burdick and Jackson, VWR International, West Chester, PA, USA) and extra pure ammonium formate (EMD, Gibbstown, NJ, USA) were purchased. Ultrapure water was supplied by an in house Millipore system (Billerica, MA, USA). Each lot of organic solvents was investigated by LC/MS infusion. Utilizing ultrapure water, aqueous buffer for liquid chromatography/electrospray ionization mass spectrometry (LC/ESI-MS) and a stock solution (1 mg/ml) of tuning reference compound was prepared freshly on a daily basis in a solvent system identical to the initial mobile phase composition of LC.
Hydrophilic interaction chromatography (HILIC)-LC/ESI-MS profiling: HILIC-LC/ESI-MS analysis [39] was performed with the use of a modified silica-based column (Luna HILIC Diol, 150_3 mm, 3 mm particle size; Phenomenex, Torrance, CA, USA). The mobile phases were 100 mM ammonium formate (pH 4.0) (A) and acetonitrile (B) (flow rate 0.4 mL/min at 40°C). After a 2 minute isocratic run at 3% A, a sequential ramping scheme was followed up to 40% A for total injection time of 20 min. The injection volume was set to 10 ml. The entire effluent from the high-pressure liquid chromatography (HPLC) column was directed into the ESI source of a linear trap quadrupole (LTQ) linear ion trap (LIT) mass spectrometer (Thermo Fisher, San Jose, CA, USA) operated under Xcalibur software (v1.4, Thermo Fisher). The electrospray voltage was set to 5 kV. Nitrogen sheath and auxiliary gas flow were set at 60 and 20 arbitrary units, respectively. The ion transfer capillary temperature was set at 350°C with typical ion gauge pressure of 0.90×10−5. Full scan spectra were acquired from 100–1500 amu at unit mass resolution with maximum injection time set to 200 ms in one micro scan. Acquisition was performed in both positive/negative switching modes. A sucrose tune file in negative/positive modes at high LC flow rate was used during all the LC/ESI-MS acquisitions on the LTQ mass spectrometer.
Metabolite annotation: Annotation of the MS and MS/MS spectra was done using commercial NIST05/Wiley Registry; METLIN, Mass-Bank, Human Metabolome Database, Lipid Maps, and in-house built mass spectral libraries. Annotation was further validated with de novo structure identification. The unique elemental formula was searched against the CAS database using the strategy of Explore Substances – Chemical Structure (American Chemical Society, Washington, DC) for known compounds, or input into the MolGen 3.5,20 generating all of the possible structural isomers corresponding to the elemental formula. The chemical structures were saved and imported to Mass Frontier 5.0 (HighChem Ltd., Bratislava, Slovakia) for MS/MS fragmentation modeling analysis. The Mass Frontier Fragments and Mechanisms module is an expert system providing information about basic fragmentation and rearrangement processes based on literature, starting from a user-supplied chemical structure. The theoretical fragments generated by Mass Frontier were compared to those acquired from LC/MS. Parent compounds that had the best match of MS/MS fragmentation pattern were considered as the molecular structures of the potential biomarkers. For validation purpose, the proposed molecules were searched against chemical structure and property databases or search engines such as publicly available PubChem, Chemical Structure Lookup Service (CSLS), CRC Dictionary of Natural Products (DNP), ChemSpider, and proprietary Beilstein Database using MDL Crossfire Commander and Chemical Abstracts Database (CAS) using SciFinder Scholar.
Raw data processing: Prior to data processing, original Xcalibur LC/MS files were converted into netCDF format using the XConverter (Thermo Fisher), then converted into WIFF format usingAnalyst QS for use in MarkerView software (v1.2, Applied Biosystems, Foster City, CA, USA). For HILIC LC/MS data, peak finding options were set as follows: subtraction offset, 10 scans; subtraction multiplication factor, 1.3; noise threshold, 3; minimum spectral peak width, 5 amu; minimum retention time peak width, 5 scans; and maximum retention time width, 1000 scans. Peak alignment options were set as follows: retention time tolerance, 0.5 min; mass tolerance, 0.8 amu; and maximum number of peaks, 5000. Peaks found in fewer than 3 of the samples were discarded using filter setting. Peak area integration was performed using raw data. Datasets were normalized using Euclidean norm by scaling each sample-vector to unit vector norm, which can be interpreted geometrically as a projection of the samples x to a hyper-sphere with the length of this sample vector scaled to one. Peaks were, then, normalized to the total absolute area of all detected metabolites in each sample. Datasets were routinely processed in MarkerView for principal component analysis. The data that constituted retention time, mass-to-charge ratio, and peak areas of detected and aligned peaks were exported from MarkerView into into MetaboAnalyst and MetPA for further analysis.
Colon contents were weight, homogenized in 1 ml PBS and analyzed using a colorimetric Total Bile Acids Assay Kit (Bio-Quant). Briefly, conversion of bile acids and Thio-NADH to 3-keto steroids and Thio-NADH by 3-α.hydroxysteroid dehydrogenase was followed by measuring change of absorbance at 405 nm. Bile acid concentrations were determined using a conjugated cholic acids standard using a protocol provided by the manufacturer.
Competitive indices were converted logarithmically (log10) prior to statistical analysis using a one-tailed parametric test (Student's t tests). Concentrations of total bile acids were analyzed using a one-tailed Student's t tests. A P value of <0.05 was considered to be significant.
Fold differences in metabolites between mock-infected mice and S. Typhimurium infected mice were converted logarithmically (log2) and a Student's t was used to determine whether differences were statistically significant. False discovery rates were estimated using the q value method. A P value of <0.05 and q value of <0.1 were considered a significant change.
|
10.1371/journal.pcbi.1000004 | Stochastic Differential Equation Model for Cerebellar Granule Cell Excitability | Neurons in the brain express intrinsic dynamic behavior which is known to be stochastic in nature. A crucial question in building models of neuronal excitability is how to be able to mimic the dynamic behavior of the biological counterpart accurately and how to perform simulations in the fastest possible way. The well-established Hodgkin-Huxley formalism has formed to a large extent the basis for building biophysically and anatomically detailed models of neurons. However, the deterministic Hodgkin-Huxley formalism does not take into account the stochastic behavior of voltage-dependent ion channels. Ion channel stochasticity is shown to be important in adjusting the transmembrane voltage dynamics at or close to the threshold of action potential firing, at the very least in small neurons. In order to achieve a better understanding of the dynamic behavior of a neuron, a new modeling and simulation approach based on stochastic differential equations and Brownian motion is developed. The basis of the work is a deterministic one-compartmental multi-conductance model of the cerebellar granule cell. This model includes six different types of voltage-dependent conductances described by Hodgkin-Huxley formalism and simple calcium dynamics. A new model for the granule cell is developed by incorporating stochasticity inherently present in the ion channel function into the gating variables of conductances. With the new stochastic model, the irregular electrophysiological activity of an in vitro granule cell is reproduced accurately, with the same parameter values for which the membrane potential of the original deterministic model exhibits regular behavior. The irregular electrophysiological activity includes experimentally observed random subthreshold oscillations, occasional spontaneous spikes, and clusters of action potentials. As a conclusion, the new stochastic differential equation model of the cerebellar granule cell excitability is found to expand the range of dynamics in comparison to the original deterministic model. Inclusion of stochastic elements in the operation of voltage-dependent conductances should thus be emphasized more in modeling the dynamic behavior of small neurons. Furthermore, the presented approach is valuable in providing faster computation times compared to the Markov chain type of modeling approaches and more sophisticated theoretical analysis tools compared to previously presented stochastic modeling approaches.
| Computational modeling is of importance in striving to understand the complex dynamic behavior of a neuron. In neuronal modeling, the function of the neuron's components, including the cell membrane and voltage-dependent ion channels, is typically described using deterministic ordinary differential equations that always provide the same model output when repeating computer simulations with fixed model parameter values. It is well known, however, that the behavior of neurons and voltage-dependent ion channels is stochastic in nature. A stochastic modeling approach based on probabilistically describing the transition rates of ion channels has therefore gained interest due to its ability to produce more accurate results than the deterministic approaches. These Markov chain type of models are, however, relatively time-consuming to simulate. Thus it is important to develop new modeling and simulation approaches that take into account the stochasticity inherently present in the function of ion channels. In this study, we seek new stochastic methods for modeling the dynamic behavior of neurons. We apply stochastic differential equations (SDEs) and Brownian motion that are also commonly used in the air space industry and in economics. An SDE is a differential equation in which one or more of the terms of the mathematical equation are stochastic processes. Computer simulations show that the irregular firing behavior of a small neuron, in our case the cerebellar granule cell, is reproduced more accurately in comparison to previous deterministic models. Furthermore, the computation is performed in a relatively fast manner compared to previous stochastic approaches. Additionally, the SDE method provides more sophisticated mathematical analysis tools compared to other, similar kinds of stochastic approaches. In the future, the new SDE model of the cerebellar granule cell can be used in studying the emergent behavior of cerebellar neural network circuitry.
| Neurons express intrinsic bioelectrical activity which is known to be stochastic in nature. In order to understand this complex dynamic behavior, computational modeling is inevitable. But, how to develop models that are capable of mimicking the intrinsic dynamic behavior of the biological counterpart accurately? On the other hand, how can detailed models, possibly also incorporating some sort of stochasticity, be simulated in a reasonable time? These questions are crucial in creating computer models of neurons with better predictive capabilities.
It is well known that many components of a neuron and its membrane, including voltage-dependent ion channels, are essential for the dynamic behavior (see, e.g., [1]). Stochasticity may as well play an interesting role in the dynamic behavior of neurons [2],[3],[4],[5]. Recent studies have indicated that the primary source of stochasticity, or noise, in vivo is the synaptic input activity (see, e.g., [2],[6]). However, there are other noise sources as well (for a review, see, e.g., [7]), including extrasynaptic inputs and ion channel stochasticity, that can have significant implications on the dynamic behavior of neurons.
Several stochastic approaches have previously been developed for modeling the bioelectrical activity of neurons and excitable tissue. Monte Carlo simulations using discrete Markov chain type of models have been performed to understand the role of randomly opening ion channels (so called microscopic approach; [7],[8],[9],[10],[11],[12],[13],[14],[15],[16],[17],[18]). On the other hand, the so called “ODE plus white noise” approach (i.e., ordinary differential equation with additive white noise) and the Langevin equations have been exploited. In these approaches, noise has been incorporated into synaptic, conductance, or voltage equations of the deterministic models (so called macroscopic level; [2],[5],[7],[19],[20],[21] for synaptic, [16],[22],[23],[24],[25] for conductance, and [21],[26],[27] for voltage incorporation of noise). Regardless of the approach, all previous studies have emphasized the importance of stochasticity on firing (see [28]). Most of the previous studies have used simple deterministic model systems, including the Fitzhugh-Nagumo neuron model [27], the Morris-Lecar model, the Hindmarsh-Rose model [29], leaky integrate-and-fire model [5],[26],[30],[31], cable model [19], and the two-conductance Hodgkin-Huxley (H-H) model [7],[8],[10],[13],[14],[15],[16],[17],[20],[22],[23],[24],[25], as example systems to study the effects of stochasticity. Only a few previous studies [2],[9],[11],[12],[21],[32] have used more realistic deterministic models than the two-conductance H-H model.
Recent theoretical work has provided evidence that more emphasis should be put on ion channel stochasticity and its role in intrinsic dynamic behavior of neurons [8],[9],[10],[11],[12]. Ion channel stochasticity is due to the thermal interaction of molecules constituting an ion channel and it can be observed as random opening and closing (gating) of an ion channel at an experimentally fixed membrane potential. This probabilistic gating of an ion channel can be considered as “ion channel noise” or “ion channel stochasticity”. Several experimental studies have shown that the opening of a single ion channel can trigger action potentials in small excitable cells that have a high input resistance. These cells include small cultured bovine chromaffin cells [33], acutely isolated mouse [34] and rat [35] olfactory receptor neurons, small cultured hippocampal neurons [36], and small cultured cerebellar granule cells [37]. The total membrane current of a small neuron is influenced by ion channel stochasticity. This can change the transmembrane voltage dynamics at or close to the threshold of firing and affect action potential initiation and subthreshold membrane potential oscillations. Subthreshold oscillations may be important in determining the reliability and accuracy of action potential timing, as well as in coincidence detection and multiplication of inputs [10].
The well-established H-H formalism has formed, to a great extent, the basis for building biophysically and anatomically detailed models of neurons. Subsequently, the roles of conductances (and, ion channels) have been addressed using these models. It should be noted, however, that the deterministic H-H formalism does not take into account the fact that the behavior of ion channels underlying the whole-cell ionic currents is stochastic in nature. In other words, the ion channel stochasticity has been ignored, as also pointed out by White et al. [28] and Carelli et al. [12]. Instead, ionic conductances have been modeled as continuous, deterministic processes. In an effort to achieve a better understanding of the complex intrinsic dynamics of a single neuron, a new approach based on stochastic differential equations (SDEs) and Brownian motion is developed here. An SDE is a differential equation in which one or more of the terms are stochastic processes, thus resulting in a solution which is itself a stochastic process. The small, electrotonically compact cerebellar granule cell is used as an example to verify broader applicability of the SDE approach for modeling. For biophysical plausibility, the stochasticity is incorporated into the gating variables of all conductances in the compartmental H-H type of model for the cerebellar granule cell. Preliminary results of the work have been presented in [38].
In this study, we use cerebellar granule cell as a test case and examine how the behavior of a small-size neuron is altered when stochasticity is introduced into the deterministic H-H type of model. In short, granule cells are glutamatergic excitatory neurons which translate the mossy fiber input into parallel fiber input to Purkinje cells [39],[40]. Granule cells are the smallest and the most numerous neuron type in the mammalian brain and have a simple morphology with an average of four short dendrites [39],[40], each receiving a single mossy fiber input. Previous experimental and modeling studies have shown that the granule cell has an electrotonically compact structure [41],[42]. This cell can thus be represented using only one compartment. Moreover, the basic single-neuron firing properties and the electroresponsiveness to various types of inputs, including intrasomatic pulses of currents and synaptic currents, have been extensively studied in vitro [43],[44],[45],[46] and in vivo [47] using the patch-clamp technique [48].
Several deterministic models have been presented for the cerebellar granule cell during the past few years [42],[46],[49],[50]. When studying the behavior of these deterministic models (see also [51]), it has become clear that, with the given parameter values, the deterministic single-cell models are not capable of reproducing the experimentally observed irregular behavior in vitro in response to depolarizing current pulses. For example, the irregularity in interspike intervals during firing, or the subthreshold membrane oscillations observed in vitro in response to current pulses ([46], see also in vivo [47]), cannot be reproduced with the existing deterministic models in a straightforward manner.
In this study, we select to use the deterministic model of [50],[51] as the basis of our new stochastic model. The deterministic model is a parallel conductance, one-compartmental model previously developed for a cultured cerebellar granule cell. The model includes six different voltage-, time- and calcium-dependent ionic currents (NaF, KDr, KA, Kir, CaHVA, and BKCa; NaF stands for the fast inactivating sodium channel, KDr for the delayed rectifier potassium channel, KA for the transient A-type potassium channel, Kir for the inward rectifier potassium channel, CaHVA for the high-voltage-activated calcium channel, and BKCa for the large-conductance calcium- and voltage-activated potassium channel) and simple calcium dynamics to describe the changes in the membrane potential. The model is based on the theory of equivalent electrical circuits, as is conventionally done in neuronal compartmental modeling. The change in membrane potential, Vm(t), is described using the ordinary differential equationwhere Iapp(t) is the applied current (for the description of parameters, see Table 1). The behavior of ionic currents is described using algebraic equations according to the H-H formalism [52]. For example, for the NaF channels, we have(2) is the maximal conductance of the NaF channels, and are the time- and voltage-dependent gating variables for the activation and inactivation processes of the NaF channels, respectively. Furthermore, constants and are the exponentials for the corresponding activation and inactivation processes, and the equilibrium potential for Na+. The processing of calcium ions is assumed to take place in small volume close to cell membrane and is linked to BKCa channel function. The change in intracellular calcium concentration, [Ca2+], is described by(3)where is the current of the CaHVA channels and v is the volume in which calcium ions are processed. For the parameters B, [Ca2+]rest, and τCa, see Table 1.
The parameter values of the original deterministic model have been selected based on data taken from in vivo and in vitro experimental records (for references see [50],[51]) on cerebellar granule cells. The original deterministic model has been verified in detail against the electrophysiological data recorded from in vitro granule cells (cf. Figures 5.3, 5.4, 5.5, 5.6, and 5.7 in [51]; cf. Figures 1, 2, and 3 in [50]). A semi-automatic parameter estimation procedure to fit the model to in vitro current clamp data is presented in [50],[51]. See [50],[51] for more details of the construction and fine-tuning of the original deterministic model. It has been shown that the deterministic model reproduces the basic firing properties of an in vitro granule cell, such as the frequent firing, the correct frequency-current (f-I) curve with different depolarizing current pulses, and the realistic single action potential waveform in response to intrasomatic current pulses [50],[51]. The deterministic model has been previously simulated using GENESIS simulator software [53]. In summary, we employ (i) a realistic one-compartmental H-H type of model, (ii) six voltage-dependent ionic conductances, (iii) simplified calcium dynamics, and (iv) stochasticity in the gating variables of ionic conductances. Item (iv) is further described in the next section.
The random nature of synaptic activity, including the probabilistic release of neurotransmitters from synaptic vesicles, is one of the main sources of noise causing variability of firing. When modeling neuronal dynamics, stochasticity has thus been typically incorporated in the model input (see, e.g., [2],[5],[19],[26]), not integrated into the model. The role of synaptic processes, however, is not covered in the present study. Instead, we concentrate on studying the random behavior of voltage-gated ion channels in shaping the input-output relations and the intrinsic dynamics of a neuron.
There are alternative ways of introducing stochasticity in the behavior of the voltage-gated ion channels. In this work, we approximate the randomness in the operation of voltage-dependent ion channels as Brownian motion, i.e., as a Gaussian process with independent increments. Therefore, we convert the complete deterministic model for the cerebellar granule cell into a stochastic model. We describe the activation and inactivation of the six different ionic conductances using stochastic differential equations of the form(4)
Here, the original deterministic equation [52] is modified by adding the stochastic component σdW. In the Equation 4, X(t) denotes the gating variable for the ion channel type in question, αX and βX the rate functions of activation or inactivation processes, and W Brownian motion. Brownian motion thus models the effects of random openings and closings of ion channels known to contribute to the very delicate subthreshold membrane dynamics in neurons. In our stochastic model, the parameter σ allows us to take into account the intensity of random fluctuations. Equation 4 is a short-hand notation of the corresponding integral equation of the form(5)where the last stochastic integral is interpreted as Itô-integral with respect to Brownian motion. To our knowledge this mathematical approach has not been presented before for realistic compartmental models of neurons, other than the cerebellar granule cell [38].
Using the common alternative notation, Equation 4 could also be given in the form(6)which includes the theoretically problematic variable, the “white noise process” ξ(t). In this paper, however, we interpret Equation 6 as a short-hand notation for Equation 5 and give an example how Equation 5 is used in the previous stochastic expansions of the original H-H model. For example, Fox [22] uses, in contrast to our model (Equation 4), a specific form of autocorrelation function to characterize the dynamics of ξ(t). This autocorrelation function has the form(7)where N is the number of specific ion channels on a given area. This form of autocorrelation function implies that ξ(t) is no longer white noise, and the solution to corresponding equation (Equation 5) can no longer be interpreted as an Itô-integral with respect to Brownian motion.
Specific types of autocorrelation functions have been used to avoid values for the gating variables which are not in the interval [0,1]. Autocorrelation function has been constructed so that it decreases the variance of the stochastic component when the value of a gating variable approaches 0 or 1. Although this approach decreases the probability of obtaining values outside the desired range, there is still a possibility that in a given point the realization of the stochastic component results in a value of the gating variable not in the interval [0,1].
It is possible to completely avoid values for the gating variables which are not in the interval [0,1]. The use of reflecting boundaries (i.e., the values under 0 or over 1 are reflected back to interval [0,1]) prevents the undesired values, but results in a model which does not correspond to the original stochastic integral equation (Equation 5).
In our model, we use a constant parameter σ and increments of Brownian motion, which ensures that the produced realizations are truly solutions of the corresponding integral equation. Another reason for selecting a constant parameter σ to our model is that, in the future, we are able to estimate its value using maximum likelihood estimation methods. This kind of estimation would be more difficult for a time-dependent parameter σ.
We have to be concerned about the undesired values of the gating variables, because the stochastic component in our model has now constant variance. This would result in problems when the values of the gating variables are close to 0 or 1. However, we are able to almost completely avoid undesired values for the gating variables by properly controlling the value of parameter σ. During depolarization only the gating variable for the KA channel inactivation approaches zero and large negative values of the stochastic component would result in negative values of the gating variable. Hence, we have to use small values of parameter σ or use a separate parameter describing the stochastic fluctuations in the KA channel inactivation process. For this paper, we choose the former approach and use the same, small value of parameter σ for all activation and inactivation processes. When the model is not depolarized, some of the gating variables are fluctuating relatively close to zero or one. This also limits our choice of proper value for the parameter σ.
In Figure 1, we present the gating variables for KA activation and inactivation process. From Figure 1 it can be seen that the model behavior is stable when the model is not depolarized, and during depolarization a properly selected value for the parameter σ ensures that the values for the gating variable are in the interval [0,1].
The complete stochastic model used in this work is described with Equation 8. We use our independently developed simulation software in the MATLAB programming environment to make the calculations. The random numbers required in the simulations are generated with MATLAB's random number generators. The following equations are used to calculate the change in membrane potential, Vm, in intracellular calcium concentration, [Ca2+], and in the gating variables for activation and inactivation processes at each time point(8)The parameters for the equations are given in Table 1 and the rate functions for the gating variables in Table 2. The selection of parameter values, including those in the rate functions, is explained in the Deterministic Model section and in [50],[51].
In the model, Wi = {Wi(t),t≥0} is Brownian motion (sometimes called the standard Wiener process to distinguish between the mathematical and physical processes), that is a Gaussian process with independent increments. This means that all finite-dimensional distributions of Brownian motion are Gaussian, Wi(0) = 0 almost surely, E(Wi(t)) = 0 for all t≥0, and Var(Wi(t)−Wi(s)) = t−s for all t≥s≥0. In addition, dWi stands for the infinitesimal increment of Brownian motion. In the simulations, the increments of Brownian motion are created by sampling a zero-mean, unit-variance normal distribution after which the sample is scaled using the time-step of the simulation. Details on discretizing Brownian motion and stochastic differential equations can be found in [54],[55].
In stochastic simulation, we use the same parameter values as for the original deterministic model (Tables 1 and 2) to elucidate the effects of addition of parameters σi on the dynamic behavior of the granule cell. For the parameters σi, we assume that σi = σ for i = 1,…,9. We use the Euler-Maruyama method [55] for simulating different realizations of the system. All simulations are carried out using the time-step Δt = 10−5 s.
Using this stochastic H-H type of model (see Equation 8), we are able to simulate, by intrinsic properties of the model, the following dynamic behavior (i)–(xii). The properties (i) through (iv) can be reproduced with both the deterministic and the stochastic model, while the properties (v) through (xii) only with the stochastic model. The stochastic expansion of the deterministic model retains all the properties of the deterministic model.
In the simulations, we observe (i) normal firing (Figure 2) that produces (ii) linear f-I curve with small depolarizing currents (Figure 3). The linearity of the f-I curve is an important requirement for a model of the cerebellar granule cell when small depolarizing current pulses are used [45],[46],[47],[51],[50]. Both the deterministic and stochastic models start firing when a small depolarizing current pulse of 11 pA is applied to the neuron soma, the value which is close to the experimentally observed threshold of firing found in vitro (cf. Figure 1B in [45]), see also in vivo (cf. Figure 1G in [47]). The f-I curves of the models are linear up to a frequency of 125 Hz, with no dampening of action potential amplitudes. With relatively strong depolarizing current pulses the models are still firing but show saturation of the f-I curves, due to high firing frequency of a small neuron.
The highest firing rate the models can attain is approximately 300 Hz. Firing frequencies of up to 250 Hz have been observed with little or no adaptation of firing in response to strong depolarizing current pulses in in vivo granule cells [47]. Furthermore, both models are capable of reproducing (iii) the KA effect (Figure 2), which is a delay in the firing caused by the KA current shown to exist in in vitro granule cells [44],[45], see also in vivo [47]. Also (iv) fast afterhyperpolarizations (fAHP) are reproduced realistically with both models.
Experimental findings have indicated that irregularities in the firing of cerebellar granule cells are at least partly driven by intrinsic mechanisms, not exclusively by synaptic mechanisms. Irregularity in firing, as well as random subthreshold membrane oscillations, have been measured in the presence of 10 µM bicuculline blocking GABA-ergic inhibition [46]. Moreover, spontaneous excitatory postsynaptic potentials (EPSPs) have rarely been detected in these experiments [46]. Similarly, irregularity in firing has been measured after application of the glutamate receptor blockers (10 µM CNQX, 100 µM APV, and 50 µM 7-Cl-kyn) [46]. Also, subthreshold membrane oscillations have been found to be independent from synaptic activity [45].
As an improvement to the deterministic granule cell model considered in this work [50],[51], and to the other previously presented deterministic models for cerebellar granule cells [42],[46],[49], we are now able to reproduce with fixed parameter values (v) irregularity in firing, including clusters of action potentials, (vi) random subthreshold membrane oscillations, and (vii) variability in heights of action potentials (Figure 2). These firing properties have been shown to be present in vitro (cf. Figures 2A and 2B in [45]; cf. Figures 1A and 1B in [46]), see also in vivo (cf. Figures 1C, 1D, and 1F in [47]). Furthermore, (viii) afterdepolarizations (ADP) and (ix) slow afterhyperpolarizations (sAHP) are reproduced realistically with small depolarizing current pulses (Figure 4; cf. Figure 1B in [46]).
Occasional (x) spontaneous firing can also be observed with current pulses smaller than 11 pA, due to the stochastic nature of the model (Figure 2 (upper panel) and Figure 5). Granule cells have been shown not to be spontaneously active in in vitro slice preparation [45]. However, in vitro granule cells in culture [37], as well as in vivo granule cells [39],[47], have been shown to be able to generate spontaneous activity when tonic inhibition of Golgi cells is reduced.
A comparison between the responses obtained by the deterministic and stochastic model is shown in Figure 6. As can be seen from Figure 6, the deterministic model (right panels) does not reproduce the experimentally observed irregularity in firing. The responses simulated by the stochastic model of this study, on the other hand, very closely resemble the experimentally obtained responses. To show variability, traces from three independent simulations with the same initial conditions are shown. The stochastic model thus expands the dynamic range of one-compartmental multi-conductance model for the cerebellar granule cell in vitro. The term “dynamic range” used in this work does not refer only to the range of firing frequencies of the model, but to the whole range of different dynamic behaviors the model is capable of attaining. Furthermore, the use of SDE approach and the presence of Brownian motion does not lead to unstable results when simulating the stochastic granule cell model. As a demonstration of this two examples showing a longer, continuous simulation are plotted in Figure 5 and Figure 7.
Variability in the firing caused by the parameter σ can be assessed by examining the histograms of interspike intervals with different values of depolarizing current pulses and different values of parameter σ (Figure 8). The histograms reveal that the value of parameter σ has a major effect on the firing with current pulses near the threshold of firing. With larger current pulses firing becomes more regular and the value of σ does not have as clear an effect. This can be observed from the histograms as a smaller deviation in the interspike intervals.
The existence of spontaneous firing can also be observed from Figure 8 (first row) where the applied current is below the threshold of firing. The increase in the value of parameter σ generates more and more spontaneous spikes which can be observed as an increase in the amount of small interspike intervals in Figure 8.
The coefficient of variation (CV) of the interspike intervals is often employed to quantify the regularity/irregularity of action potential firing. A completely regular firing has a CV of zero. In this work, the dependence of CV on different values of parameter σ and different depolarizing current pulses is studied. For the parameter values of σ = 0.1, 0.3, and 0.5, the results obtained for the mean, standard deviation (std), and CV are given in Table 3. Examination of the results shows variability in the mean firing rate when changing the value of parameter σ with depolarizing current pulses near the threshold of firing. Larger depolarizing current pulses cause the stochastic model to fire similarly as the deterministic model. With depolarizing current pulses above the threshold of firing (Iapp = 12 pA and Iapp = 29 pA; see Table 3), the increase in the value of parameter σ increases the irregularity of firing measured with the CV. However, with depolarizing current pulses below the threshold of firing (Iapp = 11 pA), the increase in the values of parameter σ enhances spontaneous activity, thus making the firing more regular. In other words, the increase in the value of parameter σ causes the membrane potential to pass the firing threshold more frequently thus decreasing the variability in the lengths of interspike intervals. This results in smaller values of CV when the value of parameter σ is increased. This can be seen from the CVs in Table 3.
Bursts of action potentials have been recently recorded in in vivo granule cells in response to sensory stimuli using patch-clamp technique (cf. Figures 3B and 3F in [47]). We are interested if these bursts can be generated intrinsically in in vitro cells, specifically in the light of recent findings by D'Angelo et al. [45]. In their study on in vitro granule cells, D'Angelo et al. [45] have concluded that bursting in cerebellar granule cells persists after NMDA receptor block (100 µM APV+50 µM 7-Cl-Kyn is used), indicating that the NMDA currents are not involved. By incorporating time dependency in the parameter σ, we are able to simulate (xi) bursts of intrinsic origin (Figures 4 and 5). In this study, we induce random changes in the parameter σ between two specified values. These values enable us to take into account two intensity levels of random fluctuations to obtain bursts. In the future, these changes can be implemented in such a way that they are linked with the experimentally observed fluctuations of, for example, voltage-dependent ion channels or synaptic currents, depending on which source(s) the bursting behavior arises.
The (xii) variability in spike timing can be observed in repeated simulations with the same initial condition. As can be seen from Figure 9, the value of parameter σ affects spike timing. Figure 9 shows that the main variability does not arise only from the timing of the first action potential, but that there is significant variability also after the first spike.
Based on the simulation results presented in the last four sections, it can be concluded that our new stochastic model is capable of reproducing the details of the firing shown for granule cells in vitro [45],[46], see also in vivo [47].
In addition to putting emphasis on choosing the correct noise model, there is a need to consider computational efficiency, especially with realistic neuron models. Using the same simulation environment, the computation time of the SDE model is only two times the computation time of the deterministic model. In other words, the simulations of the SDE model can be run in a time-scale of seconds with a standard desktop PC (in our simulations, 1.86 GHz processor with 2 GB of RAM). For example, simulating a five-second trace for Figure 7 (i.e., 500,000 time-points) using MATLAB (version 7.4.0.287 (R2007a)) programming environment takes ca. 15 seconds, in comparison to ca. 8 seconds of the deterministic model in the same simulation environment. Detailed benchmarking of different stochastic methods is demanding, being a topic of another study. It will require a careful implementation of methods using a suitable test case such as the H-H model of squid axon (see also Computation Time section in Discussion).
We have shown here that, by using stochastic differential equations and Brownian motion to incorporate ion channel stochasticity, it is possible to reproduce with high precision the intrinsic electrophysiological activity of a neuron. The method presented here has several advantages over deterministic and other stochastic approaches. First, the approach provides models of neurons with realistic irregular behavior better than the deterministic approaches commonly used in computational neuroscience. Second, it decreases the computation time in comparison to discrete stochastic approaches. Additionally, the method provides more sophisticated mathematical analysis tools compared to other, continuous stochastic approaches. In the following, we discuss these advantages as well as the limitations of the proposed method and point out some possible extensions for future work.
In general, there are a number of ways to improve deterministic compartmental models and to make them more accurate and realistic, as has also been pointed out by Carelli et al. [12]. One can include new conductances characterized for the neuron in question or introduce new dynamics for the existing conductances. Also calcium dynamics, among others, can be compartmentalized, and internal calcium stores can be added. We have strong confidence that it is equally important to consider alternative ways, such as the inclusion of stochasticity, to improve the compartmental models.
As there are experimental findings showing that irregular behavior observed in an in vitro granule cell may be driven by intrinsic mechanisms ([45],[46], see also the section Electroresponsiveness Obtained by the Stochastic Model Only), it is critical to consider ways to improve the deterministic model of the granule cell. With our new SDE model, irregularities in firing, inherent variability in electroresponsiveness and spike timing, as well as random subthreshold membrane oscillations, can be reproduced accurately. This is achieved by incorporating a stochastic component σdW in the deterministic equation for the gating variables and without changing any of the parameter values of the original deterministic model. In other words, the SDE model is able to reproduce the experimentally observed irregular behavior with the same parameter values for which the membrane potential of the original deterministic model exhibits regular behavior. Proper inclusion of stochastic elements in the operation of voltage-dependent ionic conductances should therefore be considered important, at the very least, for modeling the intrinsic electrophysiological activity of a small-size neuron.
Although several stochastic methods have been presented for describing the intrinsic activity of neurons (for a review, see, e.g., [28]), these methods have not been widely utilized in computational neuroscience, most probably due to computational reasons. At the microscopic level, a typical approach has been to use a chain of Markovian states with transition probabilities given directly by the H-H voltage-dependent transition rates (see, e.g., [8],[12],[18]). This kind of approach needs to be employed when the goal of the modeling study is to understand the biophysical mechanisms of ion channel gating. The SDE approach, on the other hand, can be used to describe the irregular behavior of a small neuron using the macroscopic measurements of ionic currents as such, thus avoiding the computationally demanding descriptions of single ion channel gating.
The computationally fast, yet accurate SDE model of the granule cell could be useful in studying the emergent behavior of cerebellar neural network circuitry. There are several interesting, experimentally observed phenomena that have to be addressed in the future, including the low-frequency oscillations observed in the cerebellar granule cell layer of awake, freely behaving rats [56] and anesthetized cats [39]. Furthermore, the tuning mechanisms controlling oscillations, resonant synchronization, and learning are of interest [47],[57],[58]. The SDE approach, in general, will help in simulating stochastic large-scale models in a relatively fast manner compared to many other stochastic approaches and in linking more tightly the molecular (see also [59]), cellular, network, and behavioral correlates of information processing in neural systems [60].
In addition to accurate reproduction of experimental findings, it is important to consider the computation time required by a specific stochastic approach. In many cases, lack of computing resources has prevented the use of stochasticity in detailed compartmental modeling. Moreover, there are very few studies reporting actual computation times to benchmark existing stochastic methods and to guide the selection of suitable method. Carelli et al. [12] have made a conclusion that intensive computation is needed to study the stochastic Markov chain model of crustacean stomatogastric ganglion neuron and the simulation of long time-series can thus become infeasible. Faisal and Laughlin [18] have studied stochastic effects of action potential propagation in thin axons where ion channel gating has been described by discrete-state Markov processes, thus directly capturing the kinetics of ion channels from patch-clamp experiments. The calculation of stochastic effects, however, has been shown to require several months of computation time on a workstation cluster.
The computation time of our SDE model is, in contrast, only two times the computation time of the deterministic model. Therefore, the computation time is considerably decreased in comparison to discrete-state stochastic approaches in which the ion channels' transition rates are described as discrete-state Markov processes. The SDE method thus makes it possible to simulate long time series, similarly as in Figure 3, in a reasonable time.
One advantage of the SDE approach is that the approach provides more sophisticated theoretical tools for analysis of models in comparison to other previously presented continuous-state stochastic approaches (see, e.g., [54],[55]). For example, the computationally fast “ODE plus white noise” approach is limited to simulation purposes and does not provide as sophisticated mathematical tools as the SDE method. Examples of the theoretical tools for the SDE approach include Sequential Monte Carlo (SMC) simulation based maximum-likelihood (ML) estimation of the model parameters. SMC methods offer, in general, a set of methods which are very flexible, relatively easy to implement, parallelizable, and applicable universally.
SMC simulation based ML estimation is a Bayesian type of estimation technique which relies on transforming the probability distributions of the estimation problem into distributions which are easy to sample. This transformation allows us to use SMC approach when drawing samples from the desired posterior distributions. Based on these samples, a maximum-likelihood estimation technique is utilized for producing ML estimates for the selected model parameters. As an example, these parameters can include maximal conductances of ionic currents and the intensity of random fluctuations in the current-clamp data. This kind of fitting makes it possible to use irregular learning data in the estimation. Our ongoing work using the SDE version of the H-H model for a squid axon has shown that accurate ML estimates can be obtained for the selected model parameters based on irregular learning data [61]. Moreover, the approximation of the likelihood function allows us also to study the sensitivity of the model parameters and the effects of the changes in their values to the model behavior. The sharper the peak is in the likelihood, around the correct parameter value, the more sensitive the model behavior is with respect to value of that parameter.
The SDE approach, inevitably, has certain challenges that need to be addressed in the future. First, the gating variables of the H-H type of models may have undesired values if no attention is paid to the selection of the value for the parameter σ. This problem may be corrected by implementing stochasticity into gating variables in such a way that the level of fluctuations is dependent on the value of the gating variable. This way we would be able to decrease the fluctuations when the value of the gating variable is approaching 0 or 1 thus decreasing the probability of obtaining values not in the interval [0,1]. This approach is, however, a matter of a future study. Second, none of the freely available neural simulation tools include the possibility to use stochastic differential equations. Presently, self-made simulation software is required which may hinder the use of SDEs in compartmental modeling. Inclusion of a variety of deterministic and stochastic methods in the simulation tools would greatly benefit neuroscientists in simulating the functions of neurons and, ultimately, of neural networks.
In the future, more work will be needed to clarify the roles of different types of noise sources for small, intermediate-size, and large-size neurons, both from experimental and theoretical points of view. As an example, when studying the effects of synaptic input noise the response dynamics of a nerve has been shown to be sensitive to the details of noise model [5]. Moreover, tools from nonlinear dynamics have to be applied to make detailed comparisons between different stochastic methods. Technologies for speeding-up the computations are equally important to develop. The proper addressing of the above-mentioned challenges will enhance our understanding of the role stochasticity has at both microscopic and macroscopic levels. |
10.1371/journal.pgen.1002741 | Stratifying Type 2 Diabetes Cases by BMI Identifies Genetic Risk Variants in LAMA1 and Enrichment for Risk Variants in Lean Compared to Obese Cases | Common diseases such as type 2 diabetes are phenotypically heterogeneous. Obesity is a major risk factor for type 2 diabetes, but patients vary appreciably in body mass index. We hypothesized that the genetic predisposition to the disease may be different in lean (BMI<25 Kg/m2) compared to obese cases (BMI≥30 Kg/m2). We performed two case-control genome-wide studies using two accepted cut-offs for defining individuals as overweight or obese. We used 2,112 lean type 2 diabetes cases (BMI<25 kg/m2) or 4,123 obese cases (BMI≥30 kg/m2), and 54,412 un-stratified controls. Replication was performed in 2,881 lean cases or 8,702 obese cases, and 18,957 un-stratified controls. To assess the effects of known signals, we tested the individual and combined effects of SNPs representing 36 type 2 diabetes loci. After combining data from discovery and replication datasets, we identified two signals not previously reported in Europeans. A variant (rs8090011) in the LAMA1 gene was associated with type 2 diabetes in lean cases (P = 8.4×10−9, OR = 1.13 [95% CI 1.09–1.18]), and this association was stronger than that in obese cases (P = 0.04, OR = 1.03 [95% CI 1.00–1.06]). A variant in HMG20A—previously identified in South Asians but not Europeans—was associated with type 2 diabetes in obese cases (P = 1.3×10−8, OR = 1.11 [95% CI 1.07–1.15]), although this association was not significantly stronger than that in lean cases (P = 0.02, OR = 1.09 [95% CI 1.02–1.17]). For 36 known type 2 diabetes loci, 29 had a larger odds ratio in the lean compared to obese (binomial P = 0.0002). In the lean analysis, we observed a weighted per-risk allele OR = 1.13 [95% CI 1.10–1.17], P = 3.2×10−14. This was larger than the same model fitted in the obese analysis where the OR = 1.06 [95% CI 1.05–1.08], P = 2.2×10−16. This study provides evidence that stratification of type 2 diabetes cases by BMI may help identify additional risk variants and that lean cases may have a stronger genetic predisposition to type 2 diabetes.
| Individuals with Type 2 diabetes (T2D) can present with variable clinical characteristics. It is well known that obesity is a major risk factor for type 2 diabetes, yet patients can vary considerably—there are many lean diabetes patients and many overweight people without diabetes. We hypothesized that the genetic predisposition to the disease may be different in lean (BMI<25 Kg/m2) compared to obese cases (BMI≥30 Kg/m2). Specifically, as lean T2D patients had lower risk than obese patients, they must have been more genetically susceptible. Using genetic data from multiple genome-wide association studies, we tested genetic markers across the genome in 2,112 lean type 2 diabetes cases (BMI<25 kg/m2), 4,123 obese cases (BMI≥30 kg/m2), and 54,412 healthy controls. We confirmed our results in an additional 2,881 lean cases, 8,702 obese cases, and 18,957 healthy controls. Using these data we found differences in genetic enrichment between lean and obese cases, supporting our original hypothesis. We also searched for genetic variants that may be risk factors only in lean or obese patients and found two novel gene regions not previously reported in European individuals. These findings may influence future study design for type 2 diabetes and provide further insight into the biology of the disease.
| Common diseases such as type 2 diabetes are highly phenotypically heterogeneous. Few studies have performed genome wide association studies in subsets of patients defined by more stringent phenotypic characteristics. It is possible that reducing the heterogeneity of disease cases may increase power to detect associations over and above the loss of power resulting from reduced numbers. To address these questions we hypothesized that the genetic predisposition to Type 2 diabetes may be different in two strata of cases defined by well-accepted cut-offs for body mass index, the strongest known risk factor for type 2 diabetes.
Genome-wide association (GWA) studies have identified ∼50 independent loci robustly associated with type 2 diabetes [1], [2], [3], [4], [5], [6], [7]. These studies have highlighted new candidate pathways involved in the disease [8], [9], identified overlap with monogenic forms of the disease [1], and provided genetic links with correlated phenotypes [10], [11].
The GWA studies of type 2 diabetes have not so far provided a greatly improved understanding of the clinical heterogeneity of the disease. Type 2 diabetes cases vary appreciably in their clinical characteristics, particularly age of diagnosis and body mass index (BMI). There is also a group of patients who may present with evidence of an autoimmune component to their diabetes, but who are not insulin dependent [12]. In contrast, the identification of the genetic component to monogenic forms of diabetes has often explained the clinical heterogeneity observed [13].
Previous studies have provided some evidence of genetic heterogeneity between non-obese and obese type 2 diabetic cases [14], [15], [16], [17]. For example, the variant with the strongest effect on type 2 diabetes risk, in TCF7L2, has a stronger effect in non-obese cases (odds ratio = 1.53 [0.37–1.71] compared to obese cases (OR = 1.21 [1.09–1.35]) [14]. The effect of FTO variation on type 2 diabetes risk depends on how cases and controls are ascertained by BMI status, but this was expected given FTO's known primary effect on BMI. In the most recent GWA studies of type 2 diabetes [1], risk variants tended to have stronger effects in non-obese compared to obese individuals – of 30 loci examined, 23 showed stronger associations in non-obese compared to obese individuals.
We designed the present study in an attempt to understand better the genetic heterogeneity of type 2 diabetes. Type 2 diabetes GWA studies tend to be enriched with cases with stronger family histories and lower average BMIs compared to community based studies. Nevertheless, there is a wide spectrum of BMI amongst type 2 diabetes cases used in GWA studies, with more cases being obese than lean. In this study we tested the hypothesis that we would identify new genetic variants by limiting the clinical heterogeneity of type 2 diabetes. By stratifying cases by their BMI status and performing separate GWA studies for each strata of BMI we identified two signals of association not previously reported in the largest GWA studies in Europeans [1], although one signal has been identified in a South Asian study [7]. In addition we confirmed with additional data that the majority of known type 2 diabetes genetic associations have stronger effects in lean type 2 diabetic cases compared to obese cases.
Descriptions of all cases are available in Table 1, and combined with control details in Tables S1 and S2. Our study was designed to limit the clinical heterogeneity of type 2 diabetes by stratification on BMI, whilst also using the largest sample sizes available:
To test the hypothesis that we would identify new variants associated with type 2 diabetes in different BMI strata, we used the following study design. We used two separate strata of type 2 diabetes cases defined by the two arbitrary, but well established, cut-offs for classifying people as overweight or obese. The first stratum consisted of non-overweight cases, here defined as “lean” (BMI<25 kg/m2). The second strata consisted of obese cases (BMI≥30 kg/m2). For each stratum we used all controls, not selected on BMI to increase statistical power and provide a more robust estimate of the population allele frequency. We did not correct for BMI as BMI was not available in all controls. To check whether or not associations were being driven primarily by effects on BMI we assessed novel variants in an existing GWA studies of BMI using 123,865 individuals from the GIANT consortium [18]. Finally, we performed sensitivity analyses, confirming our findings by stratifying controls by BMI as well as cases.
We chose to include the largest set of studies available. These studies differed in the proportion of total cases defined as lean (8.4–30.4%), the proportion of total cases defined as obese (21.2–77.8%, plus one GWA study, DGDG, that only selected non-obese cases). Some studies were specifically designed as case control studies and some as case-cohort studies, and we note that the extent of phenotyping performed to exclude autoimmune processes was different across studies, ranging from not requiring insulin treatment in the first year of diagnosis and GAD autoantibody negative, to general practitioner diagnosis of type 2 diabetes.
Descriptions of the participating studies are available in the most recent DIAGRAM manuscript [1], with summary statistics also presented in Table 1 and in Tables S1 and S2. The two discovery GWA study meta-analyses comprised 2112 lean type 2 diabetes cases or 4123 obese type 2 diabetes cases, compared against up to 54,412 controls. For a subset of SNPs available on the Metabochip (a custom Illumina iSelect SNP array that included the SNPs identified by GWA studies for several diseases and traits including type 2 diabetes loci) we included data from an additional 263 lean type 2 diabetes cases, 1735 obese type 2 diabetes cases, and 3691 controls from the GoDARTs study [19].
With the exception of the BMI-stratification of cases, the meta-analyses, individual study quality control, and analytical methods were the same as those recently reported [1]. A genomic control inflation factor was calculated for each study for each analysis, and their test statistics were adjusted accordingly. Inverse-variance fixed effect meta-analyses were performed on imputed SNP datasets, testing for an additive genetic effect. All single point effect estimates are given with their [95% confidence intervals (CI)]. Only autosomal SNPs with imputation quality scores >0.5 and a minor allele frequency >1% were included from each study. A SNP was excluded from the meta-analysed dataset if it was present in less than half of the studies. Given the use of two strata, we used a p-value threshold of 2.5×10−8 as the criterion for genome-wide significance.
An additional 4 studies, totalling 2881 lean cases, 8702 obese cases, and 18957 controls were available for de novo genotyping of SNPs (Table S2). For the DGDG replication, all polymorphisms were genotyped using the KASPar system (KBiosciences). For Malmo CC, ADDITION-Ely, and Norfolk Diabetes Case Control Study (NDCCS), Taqman assay genotyping was performed. For all four studies genotyping success rate was >95%, the genotyping error rate was 0% based on re-genotyping of 384 individuals, and all SNPs were in Hardy-Weinberg equilibrium (P>0.05). We re-performed the inverse-variance weighted meta-analysis for the replication SNPs using data from all the discovery and replication datasets.
To test whether or not type 2 diabetes associations could be primarily driven by effects on BMI, we assessed the association of novel SNPs with BMI using data from the GIANT consortium consisting of 123,865 individuals.
There are two possible reasons why a variant may be associated with type 2 diabetes in a stratified sample compared to using all data. First, the variant may have a genuinely larger effect in that stratum compared to the overall sample. Second, chance will influence which SNPs are most strongly associated in different subsets of data. To distinguish between these two possibilities we performed a case only analysis in which we tested whether variants associated with lean or obese type 2 diabetes were also associated with BMI within type 2 diabetes cases. We analysed BMI as a quantitative trait in cases from the GWA studies and meta-analysed the summary statistics. If a variant is genuinely associated with type 2 diabetes with stronger effects in the lean stratum, for example, we would expect the risk allele to be associated with lower BMI within cases. This phenomenon was previously reported for the variant in TCF7L2 [14].
SNP association statistics on glyacemic traits in healthy individuals were provided by the Meta-Analyses of Glucose and Insulin-related traits Consortium (MAGIC). Phenotypes available were fasting insulin (N = 38,238, fasting glucose (N = 46,186), beta-cell function (HOMA-B, N = 36,466), insulin resistance (HOMA-IR, N = 37,037), HbA1C (N = 46,368) and 2 hour glucose (N = 15,234) after an oral glucose challenge. All traits are naturally log transformed, besides fasting glucose, 2 hour glucose and HbA1c. The studies and methodology for these GWA study data are described in their recent publications [2], [20], [21] and available online at www.magicinvestigators.org. We also had access to data from joint meta-analyses of SNP and SNPxBMI interaction on fasting glucose (N = 58,074), insulin (N = 51,570), and 2-hr glucose (N = 15,141), also provided by MAGIC (Manning et al, in press).
Identified SNPs were searched against a collected database of expression SNP (eQTL) results including a range of tissues [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35], [36], [37], [38].
In addition to identifying new loci, we tested the impact of BMI stratification on SNPs previously identified as associated with type 2 diabetes. We calculated the individual SNP association statistics using the lean and obese meta-analyses described above.
To assess the effects of combining information from all known type 2 diabetes SNPs, we next used a single study, the GoDARTs [19] study, independent from the discovery GWA studies. In GoDARTs there were a total of 263 lean type 2 diabetes cases, 1735 obese type 2 diabetes cases, and 3691 controls. Known SNPs (N = 36 on the metabochip) were defined as those reaching genome-wide significance in studies using samples of European descent (excluding FTO due to primary effect on BMI, and DUSP9 not present on the chip) [1], [2], [3]. We also combined the 36 SNPs into a single allele count model. This analysis consisted of a logistic regression model comparing the count of an individual's type 2 diabetes risk alleles, against case-control status. Each risk allele count was weighted by the point estimate effect size of that SNP from the DIAGRAM meta-analysis [1]. We repeated this analysis using stratified controls (BMI<25 kg/m2 versus lean cases and BMI≥30 kg/m2 versus obese cases) instead of all controls. Finally, individuals were binned into quintiles based on their weighted allele score and per-quintile odds ratios calculated.
Three independent association signals reached P<2.5×10−8 in the lean case genome wide meta-analysis (Table 2). Two represented previously reported loci - TCF7L2 (OR = 1.58 [1.47–1.68], P = 2×10−40) and CDKAL1 (OR = 1.26 [1.17–1.35], P = 7×10−10). One novel locus reached genome-wide significance, lead SNP positioned ∼25 kb from the HLA-DQA2 gene (OR = 1.3 [1.19–1.42], P = 1×10−8). Three further independent signals reached P<5×10−7, two of which were previously identified (SNPs in or near ADCY5, OR = 1.25 [1.15–1.35] P = 6×10−8, and SLC30A8, OR = 1.23 [1.15–1.33] P = 4×10−8) and one of which was novel (SNPs in LAMA1, OR = 1.22 [1.12–1.30] P = 1×10−7). Rs numbers are given in Table 2.
In the obese case genome wide meta-analysis, five signals reached genome-wide significance (Table 2), all in or near known loci TCF7L2, FTO, CDKAL1, HHEX, and IGF2BP2. A further three signals reached P<5×10−7; SNPs in or near the MC4R gene (previously associated with BMI), and two other signals; in HMG20A (previously reported in South Asians -OR = 1.14 [1.09–1.19] P = 2×10−7) and in ANKS1A (OR = 1.3 [1.18–1.43] P = 5×10−7).
We sought to replicate the signals reaching P<5×10−7 not previously reported in Europeans. SNPs representing the LAMA1 (rs8090011), HLA-DQA2 (rs3916765), HMG20A (rs7178572), and ANKS1A (rs16896390) signals were genotyped in up to 2,881 lean cases, 8,702 obese cases and 18,957 control individuals. Combined discovery and follow-up association statistics for these SNPs are shown in Table 2. In the lean case analysis, the LAMA1 variant was associated with type 2 diabetes (combined P = 8.4×10−9, OR = 1.13 [1.09–1.18], total lean cases N = 4,993, controls = 70,515) compared to an OR = 1.03 [1.00–1.06] in the obese case analysis (Figure 1 and Figure 2). In the obese case analysis, the HMG20A signal was associated with type 2 diabetes (combined P = 1.3×10−8, OR = 1.11 [1.07–1.15], total obese cases N = 8,583, controls = 62,063) compared to an OR = 1.09 [1.02–1.17], P = 0.015, in the lean analysis (Figure 3 and Figure 4). In previously published studies including 8,130 cases not stratified by BMI [1], the LAMA1 and HMG20A variants reached only nominal levels of significance of P = 0.002 (OR = 1.07 [1.03–1.12]) and P = 0.003, OR = 1.07 [1.02–1.12] respectively (both in the same directions as reported here).
Considering a random-effects model [39] for both LAMA1 and HMG20A signals gave similar evidence for association (LAMA1 lean analysis: P = 5×10−10, obese analysis: P = 0.02; HMG20A lean analysis: P = 0.04, obese analysis: P = 2.7×10−8). Evidence for association at the HLA-DQA2 and ANKS1A signals was reduced when follow-up data were included.
We next attempted to understand further the associations between SNPs in the LAMA1 and HMG20A loci and lean and obese type 2 diabetes cases respectively. Our study design, together with the associations between the FTO and MC4R variants in the obese strata, suggested that variants that primarily operate through BMI could drive our newly identified associations. We therefore assessed the two signals in the existing GWA studies of BMI performed by the GIANT study and consisting of 123,865 individuals [18]. The LAMA1 SNP was not associated with BMI (P = 0.19) whilst the type 2 diabetes risk allele at the HMG20A SNP was nominally associated with increased BMI (P = 0.02).
If the associations at the LAMA1 and HMG20A loci are genuinely stronger in one strata of diabetic cases compared to the other, we should observe an association of those variants with BMI within cases only. This phenomenon has previously been reported for the variants in TCF7L2[14]. The LAMA1 type 2 diabetes risk allele was associated with lower BMI within cases alone (P = 2×10−6 when analysing BMI as a quantitative trait in 26,366 cases), a result consistent with its association being stronger in the lean case analysis. The HMG20A risk allele showed no evidence of association (P>0.05).
Next we used data from MAGIC to assess potential roles of variants in normal glycaemia. The SNP representing the novel LAMA1 association showed no association with fasting glucose (P = 0.48, beta(se) = 0.0027(0.004) N = 46,186), fasting insulin (P = 0.87, beta(se) = 0.0006(0.004) N = 38,238), HbA1C (P = 0.19, beta(se) = 0.005(0.004) N = 46,368), 2-hour glucose response (P = 0.43, beta(se) = −0.016(0.02), N = 15,234), or any of the SNP×BMI-interaction models. However, LAMA1 isn't unique amongst type 2 diabetes loci in showing no effect on glycemic traits in the MAGIC study.
The HMG20A diabetes risk allele was associated with higher fasting glucose (P = 0.04, beta(se) = 0.008(0.004), N = 46,186), higher HbA1C (P = 0.002, beta(se) = 0.01(0.004), N = 46,368) and higher fasting glucose after accounting for BMI and SNPxBMI interaction (P = 0.008, N = 58,074).
In an attempt to gain further insight into likely functional genes in the LAMA1 and HMG20A loci, we tested the lead SNPs at for association in a number of eQTL datasets. Tissues tested included various blood, brain, liver and fat samples (see Methods). Only ‘cis’ associations were considered (eQTL effects on a transcript within 1 Mb of the signal SNPs). The rs7178572 SNP in the HMG20A region was significantly associated with mRNA expression levels of HMG20A in the liver (P = 4×10−5), supported by two separate expression probes, and was the strongest known regional SNP for both the liver eQTL and type 2 diabetes. No other study-wide significant results were observed (N = 14 tissues, 24 datasets/analyses).
For each of 36 published type 2 diabetes loci (identified in European studies and available on the metabochip) we compared the effect sizes between the lean and obese GWA study meta-analyses (Table 3). Among the 36 independent variants, 29 had a larger point estimate odds ratio in the lean analysis compared to the obese analysis (binomial test of 29/36 versus 50% under the null hypothesis of no difference, P = 0.0002). We next assessed the combined effect of these SNPs in a case control study independent of the GWA studies - GoDARTs (Figure 5). In the lean stratum, we observed a weighted per-risk allele OR = 1.13 [1.10–1.17], P = 3.2×10−14. This was larger than the same model fitted in the obese strata where the OR = 1.06 [1.05–1.08], P = 2.2×10−16. Results were very similar when stratifying the controls as well as the cases by BMI: lean weighted per risk-allele OR = 1.13 [1.09–1.17]; obese weighted per risk-allele OR = 1.08 [1.05–1.10] (heterogeneity of odds ratios P = 0.036). We also observed a difference between lean and obese cases when removing controls and fitting a regression model of lean cases vs obese cases (P = 0.0001). None of these 36 variants were associated with BMI in 28,000–32,000 individuals from GIANT [1], [2].
We next divided the case/control samples into risk quintiles, based on the number of risk alleles they carry, weighted by the relative effect sizes of those alleles from the larger DIAGRAM meta-analysis. The risk of being in each quintile relative to the median quintile is shown in Figure 6. For the lean group, we observed an OR = 2.1 [1.47–3.01] for the quintile of individuals carrying the most risk alleles compared to the middle quintile. This effect was larger than that in the obese group where the equivalent OR = 1.37 [1.15–1.64].
We have confirmed our hypothesis that it is possible to identify genetic associations in previously tested samples by constraining the phenotypic heterogeneity of disease cases. By stratifying type 2 diabetes into two well accepted definitions of lean and obese cases, we identified and replicated one locus in each BMI stratum, each previously unreported in European studies: a signal in the LAMA1 gene in the lean stratum and a signal in the HMG20A gene in the obese stratum. Lack of evidence for association with BMI for these two signals in 123,000 individuals [18] argues that these associations are not driven by a primary association with BMI.
There are two reasons why previously undetected genetic associations may be observed in stratified data. First chance, in this context “sampling error”, may occur – new signals may reach statistical thresholds in subsets of data due to a combination of real association and chance. Second, the signal may represent genuine heterogeneity. The enrichment of the LAMA1 signal in lean type 2 diabetes cases compared to obese cases is likely to be a real effect but the enrichment of the HMG20A signal in obese cases is more likely to be due to chance. Whilst we observed some regression to the mean (or “winner's curse”) for the LAMA1 signal, the effects remained different in lean compared to obese cases in the replication samples alone (Figure 1). In addition, the LAMA1 type 2 diabetes risk allele was associated with lower BMI within cases alone (P = 2×10−6 when testing BMI as a quantitative trait in cases) – a similar result was previously reported for the TCF7L2 risk allele [16]. In contrast there is no evidence that the HMG20A signal is stronger in obese replication strata compared to lean replication strata (Figure 3) and there was no association with increased BMI within cases alone (P>0.05 when testing BMI as a quantitative trait in cases).
The LAMA1 signal falls in a recombination block within the LAMA1 gene (Figure 2), with the lead SNP positioned within intron 61. Searching for correlated SNPs (r2>0.5) using 1000 Genomes Project data identified only additional intronic SNPs. Previous cell biology studies support a role for LAMA1, encoding laminin-1, in diabetes etiology - inhibition of LAMA1 expression reduced glucose-stimulated secretion in INS1E cells [40]. Several studies observed the beneficial effects of laminin-1, and extracellular matrix (highly enriched with laminin-1) preparations on pancreatic islet development and function [41], [42], [43], [44], [45], [46]. Laminin-1 is expressed in intra-islet capillaries [47] and a role for laminin receptor 1 was proposed in angiogenesis [48].
The confidence in the HMG20A association is enhanced by several lines of evidence from other studies. The HMG20A signal was previously identified in a GWA study of South Asian individuals [7] and was nominally associated with fasting glucose (P = 0.04, N = 46,186) and HbA1C (P = 0.002, N = 46,368) in non-diabetic individuals analysed by the MAGIC consortium. The association with fasting glucose became stronger when adjusting for BMI in an interaction model (P = 0.008).
We initially discovered a genome-wide significant signal near the HLA-DQA2 locus, which subsequently failed to replicate (rs3916765, P = 1×10−6). This variant is not in the same gene or in linkage disequilibrium with previously reported associations between HLA loci and type 2 diabetes [1], [49]. Concerned with the prospect of this association being due to auto-immune diabetes case admixture, we assessed the association of the strongest known type 1 diabetes signals in our lean meta-analysis. None of these showed any significant evidence of association – including the lead signals from the WTCCC type 1 diabetes study in the HLA region (rs3129941, P = 0.08), or near the INS (rs3842748, P = 0.64) or PTPN22 (rs2476601, P = 0.38) genes.
This study has provided the most robust evidence to date that lean type 2 diabetic cases are likely to carry a disproportionately high load of known type 2 diabetes risk alleles. More than 80% (29/36) of type 2 diabetes variants established in Europeans had stronger effects in lean compared to obese cases and the odds ratio for the 20% of lean cases carrying the most risk alleles was more than twice that of the 20% of obese cases carrying the most risk alleles. The corollary of these findings is that obese cases on average carry a disproportionately low load of confirmed type 2 diabetes risk variants, but their diabetes risk will likely be more heavily influenced by their genetic and environmental predisposition to gaining weight in adulthood.
Despite this enrichment of stronger effects in lean versus obese cases, analyses focused only on lean cases is not a more powerful study design compared to using all cases. For each of the known loci tested, the power gained by increased effect sizes is easily offset by the reduced power of having a case sample size of ∼25%. Nevertheless our data indicate that, given limited resources, recruitment strategies that target leaner type 2 diabetes cases will have more power than those that target a similar number of cases but without enrichment for lower BMI.
There are several limitations to our study. First, the use of an unstratified control group made testing the significance of differences between lean and obese cases difficult in the context of a genome wide meta-analysis. However, several lines of evidence support our conclusions that lean individuals are enriched for known type 2 diabetes genetic effects. This evidence includes: the very large differences between the upper and lower 95% confidence intervals of the weighted per allele effects in lean and obese, the consistency of the weighted per allele results when stratifying controls as well as cases, and the 80/20 proportion of SNPs showing stronger effects in lean compared to obese individuals respectively. Second, after stratifying by BMI, we did not use other criteria to reduce the clinical heterogeneity of type 2 diabetes. Of note, cases within the BMI strata differed appreciably in their age at diagnosis and the degree to which autoimmune or monogenic diabetes had been excluded. Instead, having stratified by BMI, we opted to use the largest available sample sizes. It is possible that a small number of monogenic or autoimmune forms of diabetes amongst our cases could have reduced our power to detect novel variants. Further studies may help refine how known and novel diabetes signals operate in more clinically homogenous settings. Finally, known type 2 diabetes signals are likely to account for only a small fraction of all risk variants that exist in the genome and any inferences we make are limited to the known signals.
In conclusion, we report associations with the LAMA1 and HMG20A (not previously associated at genome-wide significance in Europeans) gene regions with type 2 diabetes risk. We have demonstrated that lean diabetic cases are enriched for known type 2 diabetes risk alleles compared to obese cases. This enrichment is consistent with the observation that many of the variants with the strongest effects on diabetes are associated with reduced beta cell function [1]. At the opposite end of the spectrum, obese cases presumably need fewer diabetes risk variants to push them towards diabetes, as they are already under strain from the physiological impact of obesity and insulin resistance. These data suggest a disease model where type 2 diabetes cases lie across a continuous distribution with regards to genetic/environmental risk, and beta-cell dysfunction versus insulin resistance aetiologies.
|
10.1371/journal.ppat.1001235 | Early Myeloid Dendritic Cell Dysregulation is Predictive of Disease Progression in Simian Immunodeficiency Virus Infection | Myeloid dendritic cells (mDC) are lost from blood in individuals with human immunodeficiency virus (HIV) infection but the mechanism for this loss and its relationship to disease progression are not known. We studied the mDC response in blood and lymph nodes of simian immunodeficiency virus (SIV)-infected rhesus macaques with different disease outcomes. Early changes in blood mDC number were inversely correlated with virus load and reflective of eventual disease outcome, as animals with stable infection that remained disease-free for more than one year had average increases in blood mDC of 200% over preinfection levels at virus set-point, whereas animals that progressed rapidly to AIDS had significant loss of mDC at this time. Short term antiretroviral therapy (ART) transiently reversed mDC loss in progressor animals, whereas discontinuation of ART resulted in a 3.5-fold increase in mDC over preinfection levels only in stable animals, approaching 10-fold in some cases. Progressive SIV infection was associated with increased CCR7 expression on blood mDC and an 8-fold increase in expression of CCL19 mRNA in lymph nodes, consistent with increased mDC recruitment. Paradoxically, lymph node mDC did not accumulate in progressive infection but rather died from caspase-8-dependent apoptosis that was reduced by ART, indicating that increased recruitment is offset by increased death. Lymph node mDC from both stable and progressor animals remained responsive to exogenous stimulation with a TLR7/8 agonist. These data suggest that mDC are mobilized in SIV infection but that an increase in the CCR7-CCL19 chemokine axis associated with high virus burden in progressive infection promotes exodus of activated mDC from blood into lymph nodes where they die from apoptosis. We suggest that inflamed lymph nodes serve as a sink for mDC through recruitment, activation and death that contributes to AIDS pathogenesis.
| Myeloid dendritic cells (mDC) are essential innate immune system cells that are lost from blood in human immunodeficiency virus infection through an ill-defined mechanism. We studied the kinetics of the mDC response in blood and lymph nodes of rhesus macaques infected with the closely related simian immunodeficiency virus. We found that differences in the number of blood mDC correlated with eventual disease outcome, as at virus set-point mDC were increased in blood in animals remaining disease free but lost from blood in animals that progressed rapidly to AIDS. mDC loss was linked to an increase in the chemokine axis responsible for mDC recruitment to lymph nodes; however, mDC did not accumulate in tissues but rather died from apoptosis. Lymph node mDC remained responsive to stimulation with a TLR7/8 agonist during infection. Importantly, mDC dysregulation was partially reversed by antiretroviral therapy. These data indicate that chronic mDC recruitment, activation and death within lymph nodes precede development of disease in SIV infected monkeys and may play a role in AIDS pathogenesis.
| Myeloid dendritic cells (mDC) are professional antigen-presenting cells that are critical for the induction of acquired immune responses to pathogens [1]. Depletion of mDC from blood in human immunodeficiency virus (HIV) infection has been well described and shown to be inversely correlated with virus load and absent from long-term non-progressors, suggesting a relationship between mDC and disease control [2]–[8]. A proposed mechanism to account for mDC loss from blood is their activation and subsequent recruitment to inflamed lymph nodes [9]. Increased expression of costimulatory molecules on blood mDC indicative of activation has been reported in HIV-infected individuals [3], [5], [10], as has accumulation of mDC in peripheral lymph nodes during acute infection [11]. However, findings relating to mDC in lymph nodes during chronic HIV infection are inconsistent, with both accumulation [12], [13] and substantial loss of mDC [14] being reported. mDC are depleted from both blood and lymph nodes of simian immunodeficiency virus (SIV)-infected rhesus macaques during AIDS [15] but data are lacking from earlier stages of infection. Few studies have evaluated mDC dynamics in both blood and lymph node in the same individuals [12], [15] and no longitudinal studies of mDC kinetics in both compartments have been reported. As such, the relationship between mDC loss and recruitment in infection remains ill-defined, and whether differences in mDC dynamics predict disease outcome is not known.
The impact of antiretroviral therapy (ART) on mDC loss and recovery in HIV infection is also unclear, as several studies indicate that ART is not effective at increasing blood mDC [6], [7], [16] while others suggest that ART significantly restores blood mDC numbers [3], [8], [17], [18]. ART rapidly resolves immune activation in lymphoid tissues [19] and may have beneficial effects on lymph node mDC activation and function [13], although this has not been well characterized.
In the present study we followed the mDC response in blood and lymph nodes over time in two cohorts of SIV-infected animals that received ART and adenovirus (Ad)-based immunotherapy with different disease outcomes. We find that loss of blood mDC at virus set-point is predictive of disease progression, whereas an increase in blood mDC is predictive of long-term absence of disease, and that even relatively short periods of ART are beneficial to mDC homeostasis. In animals that progress to AIDS the early loss of mDC from blood is associated with evidence of increased CCR7-CCL19-mediated recruitment to lymph nodes and increased apoptosis within these tissues.
Animals in this study were enrolled in an immunotherapy protocol using Ad-based vectors the majority of which has been previously described [20]. Animals were infected with the pathogenic isolate SIVmac251 by intravenous inoculation and received ART consisting of a combination of two reverse transcriptase inhibitors, 9-[2-(phosphonyl-methoxy)propyl]adenine (PMPA) and 2′-deoxy-5-fluoro-3′-thia-cytidine (FTC), from weeks 12 to 24 and weeks 32 to 44, depending on survival. Immunotherapy consisted of priming with Ad serotype 5 (Ad5)-based vectors expressing SIV Gag, Env and Nef with or without IL-15 at weeks 16 and 22 followed by boosting with Ad35-based vectors expressing the same transgenes at weeks 36 and 42. Control-treated animals were given the same regimen of Ad5-ψ5 and Ad35-ψ5 vectors that lacked transgenes [20]. Ad-based immunotherapy boosted T cell responses to SIV but had no effect on virus load, progression to disease or survival [20](and data not shown). However, when analyzed independent of immunotherapy, animals in the cohort could be readily separated into two groups based on disease progression, with one group remaining healthy until elective sacrifice at a mean of 60 weeks post infection (n = 11, ‘stable’ group), and the other succumbing to AIDS with a mean survival time of 32 weeks (n = 10, ‘progressor’ group, Table 1). AIDS was defined clinically by lymphadenopathy, persistent weight loss and anorexia, with or without opportunistic infections [15]. Equal numbers of animals in the stable and progressor groups received Ad-based immunotherapy with the remainder receiving control vectors or no treatment, confirming the lack of association between immunotherapy and disease outcome (Table 1). The MHC class I molecule Mamu-A*01 was expressed by 3/11 and 0/10 animals in the stable and progressor groups, respectively, consistent with an association of this molecule with control of SIV infection (Table 1) [21].
Peak plasma virus loads in stable and progressor animals at 2 weeks post infection were similar at around 2×107 RNA copies/ml plasma; however virus loads began to diverge by week 4 and at virus set point virus loads differed by ∼1 log between groups (Table 1 and Figure 1). Survival time was inversely correlated with virus load at set-point (Table 1 and Figure 1B), consistent with previous reports [22], [23]. ART beginning at week 12 had parallel although modest effects on virus load in both groups, with an immediate decrease of ∼1.5 logs that fluctuated over the course of therapy (Figure 1). Virus load persisted at ∼0.5 logs below set-point levels after discontinuation of ART and then decreased by ∼1.5 logs with initiation of the second cycle of ART at week 32, again with fluctuations over the course of therapy (Figure 1). Only animals in the stable group survived beyond the second cycle of ART and in these animals virus load remained at ∼1.5 logs below set-point until sacrifice (Table 1 and Figure 1). These data show that disease progression and survival in this cohort of animals correlated with virus load at set-point prior to initiation of ART and not with Ad-based immunotherapy, and that ART was effective at inducing modest but similar decreases in virus load in both groups.
The characteristics of this cohort allowed us to ask whether differences in eventual disease outcome were reflected in earlier changes in the mDC response and whether short-term exposure to ART was beneficial to this response. Blood mDC were identified in peripheral blood mononuclear cells (PBMC) as CD45+ lineage− HLA-DR+ CD11c+ cells (Figure 2A) and were enumerated based on the ratio of mDC to CD4+ T cells [20], [24]. Staining of blood cells with antibody to CD11c was inconsistent in animals R487 (stable group) and M5406 (progressor group) making it difficult to delineate mDC at all time points (data not shown), and as a result these animals were not studied further. The median number of blood mDC in the remaining 19 animals prior to infection was 51 cells/ul with a relatively large range from 16 to 202 cells/ul, consistent with our previous findings (Figure 2C, D) [24]. Blood mDC were reduced at 2 weeks post infection relative to baseline levels when all animals were analyzed together (P = .03), although when each group was analyzed separately this decrease was not significant. However, in the post-acute period the mDC response diverged, as mDC in progressor animals continued to decline to week 12 when they were significantly reduced in number relative to preinfection time points. In contrast, mDC in stable animals significantly increased from weeks 2 to 12 (Figure 2B–D). The relative change in the number of blood mDC in individual animals over the first 12 weeks of infection was significant, as mDC dropped to around 30% of preinfection levels in some progressor animals (mean for the group 60%) but increased to nearly 500% in some stable animals (mean for the group 206%) (Figure 2E). This change was inversely correlated with virus load at week 12 post infection, revealing a relationship between viral burden and mDC homeostasis (Figure 2E). Exposure to the first round of ART in progressor animals resulted in an increase in mDC number from weeks 12 to 20, when virus load was near its lowest point, and appeared to stabilize the number of blood mDC in the stable group (Figure 2B–D). However, after ART was discontinued at week 24 the number of mDC in stable but not progressor animals rose markedly reaching a mean increase of 3.5 fold over baseline at week 32, with individual increases approaching 10-fold in some animals (Figure 2B–D). The magnitude of the mDC response after ART was not influenced by the vaccination regimen received during ART, as a comparison of mean mDC counts for all animals from weeks 28 to 32 (the period of greatest response) based on the type of immunotherapy received revealed no statistically significant differences (data not shown, Kruskal-Wallis test, P = 0.3). Initiation of the second round of ART again reduced mDC number in the majority of stable animals, concurrent with the reduction in virus load, after which the number of mDC remained relatively constant (Figure 2B, D). The divergent mDC response contrasted with changes in CD4+ T cells, which did not statistically differ between groups at any time before or after infection (Figure 2F). These data indicate that differences in eventual disease outcome in SIV infection are reflected by differences in the blood mDC response that are apparent relatively early in infection. They also indicate that short-term ART may be effective at transiently restoring blood mDC in animals with the most severe disease.
We next asked whether differences in disease progression were reflected in earlier differences in activation of circulating mDC in SIV-infected macaques. For these and subsequent analyses we focused on the first 32 weeks of infection incorporating one 12-week cycle of ART and one 8-week period of treatment interruption, as after this time the number of animals surviving in the progressor group rapidly diminished (Table 1). Expression of the costimulatory molecules CD80 and CD86 was markedly increased in all animals at 2 weeks post infection indicative of rapid mDC activation (Figure 3A–D). However, by 12 weeks post infection differences in mDC activation were evident between groups particularly with respect to the chemokine receptor CCR7, which was expressed by a significantly greater proportion of mDC in animals that progressed to AIDS relative to animals with stable infection (Figure 3A). A majority of CCR7+ mDC in progressor animals also expressed CD86 with a smaller proportion expressing CD80, consistent with activation (Figure 3E). The 12-week course ART was effective at reducing blood mDC activation, particularly with respect to CD80, and in animals in the stable group expression of all markers of activation returned to preinfection levels during ART (Figure 3A–C). The increase in CD80 at week 32 suggested mDC were again activated during the period of ART discontinuation, although no increase in CCR7 or CD86 expression was noted at this time (Figure 3A–C). These findings indicate that during chronic infection animals that progress to AIDS have increased blood mDC activation relative to animals with stable infection. They also confirm that ART has the beneficial effect of reducing mDC activation, consistent with findings in HIV-infected humans [3], [12].
The finding that loss of blood mDC in progressor animals occurs as the proportion of mDC expressing CCR7 increases is consistent with excessive mDC recruitment to lymph nodes via the CCR7/CCL19/CCL21 pathway, as has been suggested by in vitro studies [9]. To examine this potential in vivo, we used flow cytometry to identify mDC in lymph node cell suspensions taken prior to infection and at intervals after infection in our two groups of animals. mDC were defined as lineage− HLA-DR+ CD11c+ cells (Figure 4A) and enumerated as a proportion of all cells in the lineage− HLA-DR+ gate, which we have previously shown to be an accurate indicator of the absolute number of mDC per unit of weight [15]. Surprisingly, we found no significant difference in the number of lymph node mDC as a result of SIV infection regardless of disease progression, indicating a lack of mDC accumulation (Figure 4B). However, the phenotype of mDC within lymph nodes was significantly different as a function of disease, as animals with stable but not progressive infection had a lower percentage of mDC expressing CCR7, CD40 and CD86 and reduced mDC expression of MHC class II at 12 weeks relative to preinfection time points, reflecting reduced mDC activation (Figure 4C, D). To address the issue of mDC recruitment further, we next used real time PCR to determine the relative expression of CCR7 ligands in lymph node tissues. We found that CCL19 but not CCL21 mRNA was increased 8-fold in lymph nodes at 12 weeks post infection, but only in animals that progressed to AIDS (Figure 4E). Together with our findings in blood, these data suggest that mDC are recruited to lymph nodes in progressive disease via an enhanced CCR7/CCL19 pathway, but that expanded mDC recruitment fails to result in mDC accumulation.
The lack of mDC accumulation in lymph nodes despite evidence for enhanced CCR7/CCL19-mediated recruitment in progressive infection led us to suspect that lymph node mDC were dying at an increased rate in these tissues. To examine this possibility we identified live mDC in lymph node cell suspensions as being lineage− HLA-DR+ CD11c+ cells that lacked staining with a fixable dead-cell dye, and then identified cells undergoing early apoptosis within this gate using an antibody to active caspase-3 (Figure 5A). At week 12 post infection, 15% of lymph node mDC in animals that eventually progressed to AIDS were entering apoptosis, representing a 3-fold increase from preinfection levels, whereas lymph node mDC from animals with stable infection had no significant change in apoptosis (Figure 5B). ART given from week 12 to 24 post infection decreased the frequency of apoptotic mDC in progressor animals, although this did not reach statistical significance (Figure 5B). To determine whether apoptosis was mediated by extrinsic or intrinsic pathways we exposed lymph node cells from progressor animals taken at week 12 post infection to small molecule inhibitors of caspase-8 or caspase-9, respectively. The presence of caspase-8 inhibitor Z-IETD-FMK reduced apoptosis by more than 50% relative to a control peptide whereas the caspase-9 inhibitor had minimal effect (Figure 5C), suggesting that cell-extrinsic mediators of apoptosis were predominant. Consistent with this finding, lymph node mDC taken from progressor but not stable animals at week 12 post infection showed a significant increase in the proportion of mDC expressing CD95 relative to preinfection samples (Figure 5D). Together, these data suggest that increased mDC apoptosis in lymph nodes during chronic infection in animals that progress to AIDS offsets the increase in mDC recruitment from blood, resulting in no net accumulation of mDC.
Changes in mDC activation and apoptosis within lymph nodes during SIV infection could impact the capacity of these cells to respond to microbial stimuli and subsequently induce adaptive T cell immune responses. To investigate the functional capacity of mDC following SIV infection in our two groups of animals we stimulated lymph node cell suspensions taken at intervals before and after infection with 3M-007, a small molecule synthetic agonist of TLR7/8, which, like HIV and SIV RNA, activates mDC through their engagement of TLR8 [25]–[27]. We analyzed mDC for expression of two key immunoregulatory cytokines, TNF-α and IL-12 (p40/p70). Interestingly, lymph node mDC taken prior to infection responded relatively poorly to short-term stimulation with a small proportion of cells producing TNF-α and IL-12 (Figure 6). In contrast, stimulation of mDC taken at 12 weeks post infection resulted in 20 to 30% of cells producing TNF-α and a smaller but significant percentage producing IL-12, representing a 4- to 5-fold increase above preinfection levels regardless of disease outcome (Figure 6). ART reduced mDC responsiveness to TLR8 stimulation although this did not reach statistical significance (Figure 6B). These data indicate that mDC resident in lymph nodes of SIV-infected rhesus macaques are functional capable of responding to stimulation, and may even be hyperresponsive as a consequence of SIV infection.
In this study we examined the relationship between mDC dynamics and disease progression over time in pathogenic SIV infection of rhesus macaques. We show for the first time that mDC are preferentially lost from blood in animals that progress to AIDS but are increased in blood of animals with long-term stable infection. This divergent mDC response was apparent at virus set-point, indicating that changes in blood mDC number over the first 3 months of infection are predictive of eventual disease progression.
mDC are recruited from blood to lymphoid tissues through upregulation of CCR7, the ligand for chemokines CCL19 and CCL21 that are expressed in the lymph node paracortex [28]. In animals with progressive infection, mDC loss from blood was associated with an increase in the frequency of blood mDC expressing CCR7 and an increase in expression of CCL19 in lymph nodes, consistent with increased extravasation to lymph nodes that exceeded the rate of mDC production from bone marrow. Expression of CCL19 has been shown previously to be markedly increased in lymph nodes during the acute phase of SIV infection [29], and our data suggest that expression in lymph nodes remains high into chronic infection as a function of virus load. Indeed, recent studies have shown that increased levels of CCL19 and CCL21 in blood correlate with higher virus loads and disease progression in HIV infected humans [30]. In vitro exposure to CCL19 and CCL21 also promotes an inflammatory response in PBMC from HIV-infected individuals with high virus loads [31]. We now provide evidence of a functional link between CCL19 upregulation in lymph nodes and increased expression of CCR7 on circulating mDC that promotes mDC recruitment to lymph nodes in progressive SIV infection. While not examined in this study, the potential exists for proinflammatory factors to promote differential emigration of mDC to lymph nodes in progressive relative to stable SIV infection. In particular, lipopolysaccharide is increased in the circulation during chronic HIV and pathogenic SIV infection as a consequence of microbial translocation through increased gut permeability [32]. Lipopolysaccharide activates mDC via engagement of TLR4 [33] and is a potent activator of DC migration in vivo [34], [35].
In contrast to progressive infection, we found that mDC in animals that controlled SIV infection had significant increases in blood mDC over time, with increases of up to 5-fold by virus set-point and nearly 10-fold in some cases at 32 weeks of infection. Studies in HIV infected individuals have indicated that mDC loss is inversely proportional to virus load, as we have shown, and is not observed in long-term non-progressors [2], [5], [8], but such cross sectional studies have by design not revealed changes over time. Increased blood mDC may arise from increased hematopoiesis in bone marrow in response to inflammatory cytokines such as TNF-α and IL-1 that are elevated during HIV infection and promote DC generation [36], [37]. The lack of an upregulated CCR7-CCL19 axis in this group would exacerbate the impact of enhanced DC production and mobilization into blood by limiting mDC exodus into tissues.
Paradoxically, there was no net increase in mDC within lymph nodes in monkeys with progressive SIV infection, associated with an increase in mDC apoptosis, suggesting that increased recruitment to lymph nodes is offset by increased cell death in severe infection. mDC apoptosis was caspase-8-dependent and associated with increased CD95 expression, similar to the findings for plasmacytoid DC in HIV and SIV infection [38], [39], consistent with a cell-extrinsic mechanism of apoptosis involving CD95 ligation. Apoptosis through virus infection of mDC is unlikely to be a significant factor, as previous studies indicate that only a minor fraction of lymph node mDC contain incorporated viral DNA during peak viremia [38]. HIV and SIV clearly affect mDC in the absence of productive infection, in particular through interactions of viral RNA with endosomal TLR8; however this interaction tends to promote cell survival rather than apoptosis [9], [27]. While the increase in mDC recruitment appears to keep pace with apoptosis in tissues during the chronic stages of SIV infection studied here it is clear that mDC are ultimately lost from lymph nodes as AIDS is established, as previously reported [15]. This eventual decline may be associated with a similar decline in lymph node expression of CCL19 in the final stages of disease [29].
Several reports have described the presence of semimature mDC with reduced expression of costimulatory molecules and/or CD83 in lymph node and spleen of HIV-infected humans [11], [13], [40] and SIV-infected macaques [41], [42]. Our data now suggest that these cells may have a beneficial function in vivo, as lymph node mDC with significantly lower expression of CCR7 and costimulatory molecules consistent with a semimature state were present only in animals with long-term stable infection. In vitro, semimature DC with tolerogenic function are derived from exposure to immunoregulatory cytokines including IL-10 and transforming growth factor-β [43], however whether these factors modulate DC maturation and function in progressive versus stable SIV and HIV infection is not known. Semimature mDC from HIV-infected lymph nodes have been shown to promote regulatory T cell function [13]. While we were not able to examine the effect of these cells on regulatory T cells in this study, the prevalence of semimature mDC in stable but not progressive infection might suggest a role for enhanced regulatory T cell responses in disease control. The role of regulatory T cells in pathogenic and nonpathogenic SIV infection is currently controversial [44]–[46], and the interplay between mDC and regulatory T cells in control and progression to disease deserves attention. In contrast to stable infection, mDC in lymph nodes of animals with progressive infection showed essentially no difference in expression of CCR7 and activation markers relative to naïve animals, although the proportion of cells expressing these markers was substantially greater than in blood. It is possible that activated mDC undergo apoptosis immediately upon entering lymph nodes, or alternatively that other newly identified costimulatory molecules from the CD28 and TNFR families not examined here may be differentially expressed in progressively infected lymph nodes [47].
In our study the two short courses of ART had only modest although similar effects on virus load in both groups of animals, reducing virus levels in plasma by ∼1.5 logs. This may be due to the fact that ART was initiated in chronic as opposed to acute infection and that therapy was limited to PMPA and FTC which both target the same viral protein, reverse transcriptase. Similarly limited effects of ART on virus load in SIV infection have been reported by others [48], [49]. Despite this, ART had noticeably beneficial effects on mDC homeostasis. In blood, ART reduced mDC activation and transiently restored mDC numbers in monkeys with progressive infection, consistent with reports in HIV-infected individuals [8], [17], [18]. Most strikingly, discontinuation of ART in stable animals led to a marked increase in the number of mDC in blood. In lymph nodes, ART resulted in a decrease in mDC apoptosis in animals with progressive infection and a reduction in mDC responsiveness overall. Consistent with this finding, expression of proinflammatory factors and CD95L that likely induce functional activation and apoptosis of mDC are substantially reduced in SIV- and HIV-infected lymph nodes in response to ART [19], [50], [51]. Animals in the progressive infection group died at a median time of 34 weeks post infection and did not receive the full second course of ART initiated at week 32. We do not believe this difference in treatment interval was a determining factor in survival, as disease status and time to sacrifice were correlated with virus load at set-point, before initiation of any therapy, and thus were independent of ART. HIV-infected individuals with higher baseline virus loads and immune activation have poorer reconstitution of innate immune cells in response to ART [16], [52]. In our study, differences in baseline virus load influenced the response to ART, as animals with stable and progressive infection had transient increases and decreases, respectively, in the number of blood mDC, although this could clearly be influenced by the differences in virus load in the two groups while on ART. It will be important to determine the impact of improved antiretroviral drug regimens on mDC dynamics in SIV infection, including the orally available integrase inhibitors that are highly active in monkeys [53].
Our data indicate that mDC present in lymph nodes in SIV infected monkeys remain functionally responsive to exogenous stimulation regardless of disease outcome. The finding that ex vivo stimulation through TLR8 induced a five-fold increase in expression of TNF-α relative to naïve animals suggests that these cells may in fact be hyperresponsive, although testing with a more extensive panel of agonists targeting different TLR ligands is needed to confirm this. CCL19 induces terminal activation of DC and promotes DC production of proinflammatory cytokines within lymph nodes [54], although this effect would not explain the finding of increased responsiveness of mDC in animals with stable infection that had normal levels of CCL19 in our study. It is possible that other proinflammatory factors such as IFN-γ that induce DC activation [55] and are markedly increased in lymph nodes during pathogenic SIV infection [56] are responsible for mDC increased responsiveness in SIV infection.
An increasing emphasis in HIV and SIV pathogenesis is now placed on the role of gut mucosa in disease, as this is a major site of virus replication and CD4+ T cell depletion [57]–[59]. mDC are recruited to inflamed respiratory mucosal surfaces in children with respiratory viral infections [60], and it is likely mDC and other DC subsets are similarly recruited to gut and vaginal mucosa in SIV infection [61]. It will be important to evaluate the mDC response in gut mucosa and its relationship to disease progression in SIV infection. However, such quantitative studies are technically difficult to perform as the DC is a relatively rare cell that can only be isolated in sufficient numbers through gut resection surgeries as opposed to the more commonly performed endoscopic biopsies.
Collectively, these data suggest that the inflammatory response associated with increased virus load during progressive SIV infection leads to an increase in the CCR7-CCL19 chemokine axis that serves to accelerate mDC recruitment to lymph nodes. Apoptosis of mDC within tissues during this chronic phase, which was found only in animals with progressive infection, would compromise the innate and adaptive immune response to opportunistic pathogens promoting disease progression. It is currently not clear whether recently recruited and activated mDC produce increased levels of proinflammatory cytokines in vivo that may mediate immune activation characteristic of HIV and pathogenic SIV infection [62]. Interestingly, increased turnover of blood monocytes associated with apoptosis of tissue macrophages has been shown to correlate with progression to disease in SIV-infected macaques and is a better predictive marker than viral load or lymphocyte activation [63], [64]. This response is not likely limited to lymph nodes, as evidenced by the fact that increased monocyte turnover and recruitment to brain correlates with the severity of SIV encephalitis [65]. These findings point to a broad-based dysregulation of mDC and monocytes in blood and tissues as a significant factor in the pathogenesis of AIDS.
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. The protocol was approved by the Institutional Animal Care and Use Committee at the University of Pittsburgh (Assurance Number A3187-01). Surgeries were performed under anesthesia induced and maintained with ketamine hydrochloride and medetomidine hydrochloride, and all efforts were made to minimize suffering.
Twenty one Indian-origin rhesus macaques (Macaca mulatta) used in this study were housed at the University of Pittsburgh Primate Facility for Infectious Disease Research. All animals were infected by intravenous inoculation with 1,000 TCID50 of uncloned, pathogenic SIVmac251 (provided by Christopher J. Miller, California National Primate Research Center). Virus load in plasma was determined as described previously [66]. ART consisted of PMPA (20 mg/kg/d, subcutaneous injection) and FTC (30 mg/kg/d, subcutaneous injection; both provided by Michael Miller, Gilead Sciences) and was given from weeks 12–24 and from weeks 32–44 as described [20]. All animals except R189 (sacrificed at week 11 post infection) received one or more administrations of Ad-based vectors during the study depending on survival. Priming injections of separate Ad5-based vectors expressing codon-optimized SIVmac239 gag, env and nef with and without rhesus IL-15.FLAG or empty Ad-ψ5 were given by intramuscular injection at week 16 (5×1010 total viral particles) and week 22 (1×1011 total viral particles), and boosting injections of the same quantity of Ad35-based vectors expressing the same transgenes were given at week 36 and 42. All Ad vectors were E1/E3-deleted with the exception of Ad35 containing the env gene which was E3 deleted. Lymph nodes were taken from the axillary or inguinal regions prior to infection and at weeks 12, 24 and 32 post infection and single cell suspensions were generated by disruption and digestion with collagenase D, as described [67].
Identification of mDC was performed as previously described with some modifications [15], [24]. Briefly, PBMC or lymph node cell suspensions were stained with fluorescently-labeled antibodies to Lineage markers [CD3 (clone SP34-2; all antibodies from BD Bioscience unless otherwise noted), CD14 (M5E2), and CD20 (2H7)], HLA-DR (G46-6) and CD11c (S-HCL-3), with and without antibodies to CD45 (D058-1283), CD80 (L307.4), CD86 (FUN-1), CCR7 (150503, R&D Systems), CD40 (5C3) and CD95 (DX2). An amine-reactive fixable dead-cell dye (Invitrogen) was used to discriminate live from dead cells. mDC were defined as Lineage− HLA-DR+ cells expressing CD11c. In lymph nodes a broad Lineage− HLA-DR+/++ gate was used to include all mDC as described previously The number of blood CD4+ T cells was quantified using a precise volume of blood stained with antibodies in the absence of any wash step in TruCOUNT tubes (BD Biosciences) that contained a known number of fluorescent beads to provide internal calibration, as previously reported [20]. The number of blood mDC was then calculated based on the ratio of mDC to CD4+ T cells in PBMCs at the same time point [24]. All analyses were done on an LSR II flow cytometer with FACSDiva software (BD Bioscience).
Intracellular cytokine production by lymph node mDC was measured as described previously for plasmacytoid DC with minor modifications [38]. Briefly, cell suspensions were cultured for 5 hours with 10 µM of the TLR7/8 agonist 3M-007 (3 M Pharmaceuticals) with and without the addition of 10 µg/mL brefeldin A (Sigma) after 1 hour. Cells were stained with surface-labeling antibodies as above and fixed and permeabilized prior to incubation with antibody to TNF-α (MAb11) and IL-12 (8.6, Mitenyi Biotec) and analysis by flow cytometry. To detect apoptosis in mDC, lymph node cell suspensions were cultured in media for 3 hours with and without caspase-8 inhibitor Z-IETD-FMK, caspase-9 inhibitor Z-LEHD-FMK or irrelevant peptide Z-Fa-FMK (BD Biosciences). Cells were stained with surface-labeling antibodies as above and fixed and permeabilized prior to incubation with antibody to active caspase-3 (C92-605) and analysis by flow cytometry.
Total lymph node RNA was extracted and purified from cell suspensions generated from biopsies taken prior to or 12 weeks after infection using the RNAeasy kit (Qiagen) after treatment with DNAse I (Invitrogen). cDNA was synthesized using random primers and Superscript II reverse transcriptase (Invitrogen). Primers and probes from Taqman human gene expression arrays (Applied Biosystems, Foster City, CA) were utilized for real time PCR analysis of CCL19, CCL21 and β-glucuronidase expression as previously described [68]. mRNA expression levels for each gene were calculated with the 2−ΔCT method using β-glucuronidase as the endogenous control [69].
Comparisons between two groups were carried out using the Mann-Whitney U test. Comparison of DC numbers across different time points was carried out using the Wilcoxon signed-rank test. Correlations were determined using the non-parametric Spearman rank test. Graphpad Prism 5 (Graphpad Software) was used for statistical analysis. All P values are two-sided with significance considered to be P<0.05.
The identification of genes analyzed in this paper as defined by Entrez-Gene are 574386 (CCL19), 574183 (CCL21) and 677692 (β-glucuronidase).
|
10.1371/journal.pntd.0002042 | Multiplex Real-Time PCR Diagnostic of Relapsing Fevers in Africa | In Africa, relapsing fever borreliae are neglected arthropod-borne pathogens causing mild to deadly septicemia and miscarriage. The closely related Borrelia crocidurae, Borrelia duttonii, Borrelia recurrentis and Borrelia hispanica are rarely diagnosed at the species level, hampering refined epidemiological and clinical knowledge of the relapsing fevers. It would be hugely beneficial to have simultaneous detection and identification of Borrelia to species level directly from clinical samples.
We designed a multiplex real-time PCR protocol targeting the 16S rRNA gene detecting all four Borrelia, the glpQ gene specifically detecting B. crocidurae, the recN gene specifically detecting B. duttonii/B. recurrentis and the recC gene specifically detecting B. hispanica. Compared to combined 16S rRNA gene and flaB gene sequencing as the gold standard, multiplex real-time PCR analyses of 171 Borrelia-positive and 101 Borrelia-negative control blood specimens yielded 100% sensitivity and specificity for B. duttonii/B. recurrentis and B. hispanica and 99% sensitivity and specificity for B. crocidurae.
The multiplex real-time PCR developed in this study is a rapid technique for both molecular detection and speciation of relapsing fever borreliae from blood in Africa. It could be incorporated in point-of-care laboratory to confirm diagnosis and provide evidence of the burden of infection attributed to different species of known or potentially novel relapsing fever borreliae.
| Four cultured Borrelia species are responsible for relapsing fever in Africa. Three species (Borrelia crocidurae, Borrelia duttonii, Borrelia hispanica) are transmitted by ticks, whereas Borrelia recurrentis is transmitted by the body lice. These Borrelia species result in febrile infection mimicking malaria with varying severity, but particularly devastating during pregnancy where infection will often cause miscarriage. The lack of comprehensive laboratory tools for detection and speciation of these borreliae limits both medical management of patients and knowledge of the epidemiology of relapsing fevers in Africa. Based on genome analysis, we develop herein a multiplex real-time PCR assay targeting the 16S rRNA gene detecting all four borreliae, the glpQ gene detecting B. croidurae, the recN gene detecting B. duttonii/B. recurrentis and recC gene detecting B. hispanica. Compared to gold standard, this multiplex real-time PCR assay yielded 100% sensitivity and specificity for B. duttonii/B. reccurentis and B. hispanica and 99% sensitivity and specificity for B. crocidurae, when applied to 398 blood specimens. These findings provide the proof-of-concept that multiplex real-time PCR is a new tool for diagnosis of relapsing fever borreliae in Africa.
| In Africa, relapsing fevers are neglected febrile infections caused by Ornithodoros spp. tick-borne borreliae (Borrelia crocidurae, Borrelia hispanica and Borrelia duttonii) and the Pediculus humanus louse-borne Borrelia recurrentis [1], [2]. Also, poorly characterized, yet uncultured new “Borrelia mvumi” species has been reported in acute patient's blood and Ornithodoros porcinus argasid ticks in Tanzania [3]. The relative geographic specificity of each Borrelia species has been challenged by coexistence of two species in the same region [4]. B. hispanica prevalence was reported to be 20.5% among febrile patients in Northwestern Morocco [5]; the prevalence of cases attributed to B. crocidurae among febrile patients is of 11 per 100 person-years in Senegal [1]. B. duttonii has been documented in Tanzania and B. recurrentis in Ethiopia [5], [6], [7], [8]. Relapsing fevers are of further concern in travelers returning from Africa in Europe [9], [10], [11].
Relapsing fevers are treatable infections but the severity of the disease ranges from asymptomatic to fatal if left untreated [12]. In Rwanda and in Tanzania the investigators found a 30% risk for pregnancy loss and a perinatal mortality rate of 15% [8], [13]. The prognosis depends in part on the causative species with the case-fatality ratio being higher for B. recurrentis infection than for the other infections [12]. However, the vast majority of patients are diagnosed on the basis of non-specific clinical features that overlap with those of malaria [4] and the poorly-sensitive, non-species specific microscopic observation of blood-borne Borrelia [1].
PCR-based tests have been therefore developed to improve the laboratory-based diagnosis of relapsing fevers in Africa [4]. In particular, real-time PCR targeting the 16S rRNA gene or the glpQ gene improved the sensitivity of the diagnosis when compared to microscopy [14], [15]. Also, we previously showed that PCR-sequencing intergenic spacers could be used for genotyping B. crocidurae, B. duttonii and B. recurrentis [16].
Here, we present the development and evaluation of a multiplex, quantitative real-time PCR detecting any relapsing fever Borrelia [14] and specifically B. crocidurae, B. hispanica and B. duttonii/B. recurrentis based on post-genomic analyses [16], [17]
B. crocidurae Achema strain, B. recurrentis A1 strain and B. duttonii Ly strain were grown in BSK-H medium (Sigma-Aldrich, Saint Quentin Fallavier, France) supplemented with heat-inactivated 10% rabbit serum (Eurobio, Courtaboeuf, France) before DNA extraction. B. hispanica DNA was directly extracted from two argasid ticks Ornithodoros erraticus sensu lato collected from Morocco. DNA was extracted from all specimens using QIAamp DNA Blood mini kit (QIAGEN, Hilden, Germany) according to the manufacturer's instructions. Reference identification of borreliae was made by combining the 16S rRNA gene and flaB gene sequencing [6], [18].
Total blood DNA was extracted from 21 blood specimens found positive by microscopy collected in 1994 from patients with relapsing fever in Addis Ababa, Ethiopia, 18 specimens collected in Mvumi, Tanzania [19], 9 specimens collected in 2011 in Bahir Dah, Highlands of Ethiopia [19] and 224 blood specimens collected from febrile patients from Ndiop and Dielmo villages in Senegal between 2008 and 2012 where B. crocidurae is endemic.
Genome sequence of B. duttonii (GenBank accession number CP000976), B. recurrentis (GenBank accession number CP000993) and B. crocidurae (GenBank accession number CP003426.1) were downloaded from GenBank. Comparative genomic analyses were performed on the chromosomes in order to identify species-specific sequences. In addition, a 16SrRNA gene sequence-based system previously developed in our laboratory was used for Borrelia genus detection as previously described (14). Sequence alignments were performed using MULTALIN software for selection of each target sequence [20]. The primers and probes were constructed by using primer3 program at.http://frodo.wi.mit.edu/. Specificity of primers and probes were determined in-silico. Two different fluorescent dyes, VIC and FAM were used for labeling the probes.
The single real-time PCR experiment was performed on Roche Lightcycler (RocheDiagnostic, Maylan, France). The amplification program included two initial holds at 50°C for 2 min and 95°C for 15 min, followed by 40 cycles consisting of 95°C for 30 seconds and 60°C for 1 minute. Five µL of extracted DNA, 0.5 µL of each primer (10 pmol) and 0.5 µL of probe (10 pmol) were added to the 10 µL Quantitative PCR Master mix (Quantitec, Qiagen) and the volume was adjusted to 15 µL by adding distilled water. The multiplex real-time PCR was performed using a Stratagene Mx3000P real-time thermocycler (Agilent, Courbevoie, France) by adding five microliters of extracted DNA, 0.5 µL of each primer (10 pmol) and 0.5 µL of each probe (10 pmol) [1 labeled with FAM and 1 labeled with VIC] to 12.5 µL of Quantitative PCR Master mix 2X (Quantitec) and the final volume was adjusted to 20 µL by adding distilled water. Negative control consisting of DNA-free water was included every 10 tested specimens.
To assess the specificity of the real-time PCR systems developed herein, DNA extracted from Borrelia burgdorferi, Borrelia hermsii, Borrelia parkeri, Coxiella burnetii, Bartonella henselae, Rickettsia africae, Rickettsia felis and Tropheryma whipplei were incorporated into real-time PCR using the experimental conditions described above. In order to determine sensitivity, a puc 57 plasmid was constructed containing the human albumin gene, a 129-bp recN gene fragment from B. duttonii, a 122-bp recC gene fragment from B. hispanica and a 110-bp glpQ gene fragment from B. crocidurae (Invitrogen, Saint Aubin, France) (Figure 1). Tenfold serial dilutions of this constructed puc 57 plasmid were prepared equivalent to 107 to 101 Borrelia organisms.
This study was approved by the IFR48 Ethic Committee. All patients provided informed written consent.
As a specific target for B. crocidurae, we selected glpQ encoding glycerophosphodiester phosphodiesterase that is conserved among relapsing fever borreliae but absent from Lyme disease borreliae and additionally possesses a B. crocidurae specific 4-bp single nucleotide polymorphisms (SNPs). We selected the chromosomal recN gene encoding DNA repair ATPase for B. duttonii/B. recurrentis, that is conserved among the relapsing fever group and absent in the Lyme disease group borreliae, and furthermore exhibits a 5-bp specific SNP. For B. hispanica, we selected recC gene encoding exodeoxyribonuclease V, present in both relapsing fever and Lyme disease group borreliae and exhibiting 4-bp species-specific SNP. One probe specific for each of the three targeted regions was designed to span the region containing the SNPs. Sequences of the primers and probes are given in Table 1. Each set of species-specific primers and probe was first evaluated alone before being incorporated into a multiplex format. There was no difference in the amplification curves when comparing the single-target real-time PCR with multi-target real-time PCR assays. Figure 2 illustrates the results of these two experimental steps and shows that 16SrRNA gene probe labeled with FAM and glpQ gene probe labeled with VIC fluorescent dyes could both be detected in one PCR reaction performed simultanoiusly. Similar results were obtained with recN and recC labeled with VIC and 16SrRNA probe labeled with FAM.
In all experiments, negative controls remained negative. The cycle threshold (Ct) values for the constructed plasmid ranged from 18 (107 copies) to 36 (100 copies) per 5 µL of plasmid dilution for recN, recC and glpQ. Based on these results, we used a Ct cutoff value of 36 for interpretation a clinical blood specimens as positive. The 16S rRNA probe detected all the Borrelia-positive specimens regardless of the species with Ct values ranging from 18 to 35. The glpQ assay designed to be specific for B. crocidurae did not amplify B. duttonii or B. recurrentis reference strains, 18 B. duttonii-positive blood samples, 30 B. recurrentis-positive blood samples, two B. hispanica-positive ticks, and other strains mentioned above. The recN assay for specific detection of B. duttonii/B. recurrentis did not detect B. crocidurae, B. hispanica, B. burgdorferi and other strains mentioned above. Likewise, the recC system specific for B. hispanica did not detect B. crocidurae, B. duttonii/B. recurrentis, B. burgdorferi and other strains mentioned above.
Human albumin used as a positive control was detected by real-time PCR in all tested human blood specimens, indicating lack of PCR inhibition. When applied to 101 specimens negative for borreliae and 123 specimens found positive for B. crocidurae using combined 16S rRNA/flaB-gene PCR gold standard the observed Ct value for the clinical samples varied between 18 to 35. The multiplex real-time PCR yielded 100% specificity and 99% sensitivity (one positive specimen remained negative). No DNA remained from this false-negative specimen to enable targeted study of glpQ for mutations in the probe region. When applied to 101 specimens negative for borreliae and 30 specimens found positive for B. recurrentis and 18 specimens found positive for B. duttonii using gold standard, the multiplex real-time PCR yielded 100% specificity and sensitivity. As for B. hispanica DNAs samples, the two tick extracts were detected by recC probe.
In this study, all the negative controls remained negative in every real-time PCR experiment. Also, no evidence of PCR inhibition was detected using human blood as confirmed by amplification of the human albumin internal control in every PCR run. Specificity of primers and probes was confirmed by in-silico analyses and reinforced by experimental demonstrations that these assays failed to amplifyother microorganisms responsible for septicemia, including R. felis [21] and T. whipplei [22], all demonstrated to be emerging, highly prevalent pathogens in Africa and in Senegal in particular. Therefore, results reported herein were interpreted as authentic.
In this study, species-specific primers and probes based on the glpQ, recC and recN gene sequences were selected from the alignment of the B. crocidurae, B. duttonii, B. recurrentis and B. burgdorferi reference chromosome genomes [16], [17]. Plasmid sequences were avoided, because of their instability among different strains of the same species, and during replication of the same isolate, with a risk of resulting in false-negative results. This approach proved successful for the differentiation between B. crocidurae, B. duttonii/B. recurrentis and B. hispanica. Despite evident interest in distinguishing B. duttonii and B. recurrentis for accurate epidemiological purposes, discrimination between B. duttonii and B. recurrentis was not possible here in agreement with previously reported very close genetic and genomic proximity of both species [16], [17]. Indeed, genetic and genomic data suggested that B. duttonii and B. recurrentis could be regarded as a single Borrelia species [17]. This limitation may not be problematic as for the routine diagnosis since these two species are respectively transmitted by tick and lice in very different epidemiological contexts [23]. Also, the multiplex real-time PCR proved highly sensitive, detecting 100 copies, that is more sensitive than the 103–105 borreliae/µL reported for microscopy [14], [24]. Previously, borreliae not detectable by microscopy, were detected by using real-time PCR targeting the fla and the glpQ genes [3], [15]. These assays however could not identify borreliae at the species level [3], [15]. Another real-time PCR assay was devoted to the specific detection of B. recurrentis and Rickettsia prowazekii, as these two pathogens are both transmitted by body lice. This targeted the flagellin gene of B. recurrentis with a sensitivity of 101 borreliae [25].
The developed real-time PCR was validated against large number of samples from areas endemic for diverse borreliae causing human infection in Africa, showing 100% sensitivity and specificity except for glpQ gene which had 99% sensitivity due to failure to identify B. crocidurae in one blood specimen. Unfortunately, we could not further analyze this specimen to assess whether this false negative result arose from glpQ mutations or was indicative of a different species or subspecies related to B. crocidurae. Indeed, several recent reports indicate that new relapsing fever Borrelia species are present in Tanzania and South Africa [26], [27]. Therefore careful optimization is required to ensure that the multiplex real-time PCR technique employed will not miss these new species or other Borrelia species. For this, we incorporated the 16S rRNA gene probe in the system to serve as an indicator of the detection of any relapsing fever Borrelia in the specimen. A specimen detected positive by the 16S rRNA gene probe and negative by the species-specific probes would indicate a new Borrelia species. Such samples could be further subjected to in situ typing such as the recently described multiple spacer sequence typing [16].
In conclusion, as detection and identification of these genetically closely related relapsing fever borreliae in Africa remains challenging, the multiplex real-time PCR assay reported herein offers significant improvement over existing procedures for the diagnosis of relapsing fevers in Africa. Importantly, it permits rapid differentiation of relapsing fevers from the clinically similar malaria, that requires drastically different therapeutic management. It is a sensitive and specific technique capable of detection of major Borrelia pathogens in humans, yet will not overlook detection of potentially new species. This multiplex real-time assay being amenable to point-of-care laboratories in Africa [28], it provides an effective solution for enhanced characterization of relapsing fever borreliae in Africa, improving medical management of patients and facilitating epidemiological studies.
|
10.1371/journal.pntd.0002363 | Induction of Mouse Melioidosis with Meningitis by CD11b+ Phagocytic Cells Harboring Intracellular B. pseudomallei as a Trojan Horse | Approximately 3–5% of patients with melioidosis manifest CNS symptoms; however, the clinical data regarding neurological melioidosis are limited.
We established a mouse model of melioidosis with meningitis characterized by neutrophil infiltration into the meninges histologically and B. pseudomallei in the cerebrospinal fluid (CSF) by bacteriological culturing methods. As the disease progresses, the bacteria successively colonize the spleen, liver, bone marrow (BM) and brain and invade splenic and BM cells by days 2 and 6 post-infection, respectively. The predominant cell types intracellularly infected with B. pseudomallei were splenic and BM CD11b+ populations. The CD11b+Ly6Chigh inflamed monocytes, CD11b+Ly6Clow resident monocytes, CD11b+Ly6G+ neutrophils, CD11b+F4/80+ macrophages and CD11b+CD19+ B cells were expanded in the spleen and BM during the progression of melioidosis. After adoptive transfer of CD11b populations harboring B. pseudomallei, the infected CD11b+ cells induced bacterial colonization in the brain, whereas CD11b− cells only partially induced colonization; extracellular (free) B. pseudomallei were unable to colonize the brain. CD62L (selectin) was absent on splenic CD11b+ cells on day 4 but was expressed on day 10 post-infection. Adoptive transfer of CD11b+ cells expressing CD62L (harvested on day 10 post-infection) resulted in meningitis in the recipients, but transfer of CD11b+ CD62L-negative cells did not.
We suggest that B. pseudomallei-infected CD11b+ selectin-expressing cells act as a Trojan horse and are able to transmigrate across endothelial cells, resulting in melioidosis with meningitis.
| A mouse model of melioidosis with meningitis was established in this study by injecting mice with B. pseudomallei intravenously. Bacteria successively colonized the spleen, liver, bone marrow (BM) and brain. After infection, the splenic and BM CD11b+ populations carrying intracellular B. pseudomallei selectively expanded and became predominant. After adoptive transfer, melioidosis with meningitis was induced by the infected BM CD11b+ cells, partially induced by BM CD11b− cells and was not induced by splenic CD11b− cells or extracellular bacteria. The induction of melioidosis with meningitis was correlated with an increase in splenic CD11b+ selectin (CD62L)-expressing cells.
| The saprophytic rod Burkholderia pseudomallei is a causative agent of melioidosis and is endemic to tropical areas such as Southeast Asia and northern Australia [1]. The main modes of transmission of melioidosis are inhalation and subcutaneous inoculation [2]. Ingestion can cause a systemic infection, and consequently, the gastrointestinal tract can serve as a reservoir for the dissemination of melioidosis [3], [4]. Acute melioidosis with septicemia, which is transmitted through various routes of infection, is the most severe for humans [5] and animals [3], [6]–[10]. However, the clinical spectrum of melioidosis varies; approximately 3–5% of patients develop neurological symptoms, including macroscopic brain abscess, brainstem encephalitis or flaccid paraparesis [11]–[15]. Although melioidosis with primary meningitis is rarely seen, meningitis could arise due to the spread of B. pseudomallei from a remote infected site via the blood-stream or from ruptured cerebral abscesses into adjacent foci [15]. Fatalities due to melioidosis with meningitis have been reported in neonates, patients receiving inappropriate antibiotic treatment and patients with long-term infections [14], [16]–[18].
During mouse bacteremic melioidosis, the spleen and liver are the primary infected foci; both contain a large amount of B. pseudomallei, even during chronic melioidosis [7], [10], [19]. It has been reported that the bacteria colonizing the lungs, lymph nodes and brain disseminate locally from infected foci [7], [20]. The induction of interferon-γ (IFN-γ) and the activation of phagocytic cells are essential for the early control of B. pseudomallei in mice. IFN-γ depletion in the blood results in a rapid increase in bacterial burdens in the organs [7], [21]. The replication of invasive B. pseudomallei in infected foci can be controlled by host immunological events that recruit a large number of activated neutrophils and monocytes [19], [22], [23]. However, it is very difficult for the host to clear B. pseudomallei because B. pseudomallei invades macrophages, monocytes and hepatocytes and grows intracellularly [24]–[26]. B. pseudomallei induces cellular actin polymerization and rearrangement, resulting in cell-cell fusion and the formation of multinucleate giant cells, thus facilitating cell-to-cell spread [27]–[29]. It is believed that the intracellular bacteria grow steadily when host cytokines are depleted or when macrophage activity is attenuated [30], [31].
Meningeal neutrophil infiltration is a hallmark of bacterial meningitis. Leukocytes do not normally adhere to endothelial cells except during activation. Endothelial cells and leukocytes express complementary adhesion molecules (selectins and integrins) that are responsible for rolling, adhesion and transendothelial migration (of leukocytes) into the meninges [32]–[34]. Mouse bacteremic melioidosis induces a robust inflammatory response marked by the upregulation of the cytokine-induced neutrophil chemoattractant (KC), macrophage inflammatory protein-2 (MIP-2), monocyte chemoattractant protein-1 (MCP-1), granulocyte-macrophage colony-stimulating factor (GM-CSF) and macrophage CSF (M-CSF) [19]. Circulating activated phagocytes that are intracellularly infected with B. pseudomallei can cross the endothelial cells into the brain, and consequently, melioidosis-associated meningitis can occur. In this study, we addressed whether an activated phagocytic population harboring B. pseudomallei plays a role in inducing mouse melioidosis with meningitis.
In this study, animal experiments were conducted following the Guide for the Care and Use of Laboratory Animals (National Animal Laboratory Center, Taiwan) and were approved by the Institutional Animal Care and Use Committee at the National Kaohsiung Normal University, Taiwan (approval ID: 9901). Linking B. pseudomallei data and private information of melioidosis patients is legally prohibited by the Personal Information Protection Act (Taiwan). All experiments using viable B. pseudomallei were performed in an air flow-controlled lab (BSL III level), and the procedures were approved by the Institutional Biosafety Committee (NKNU, Taiwan).
B. pseudomallei vgh19 (id, 3052; http://bpseudomallei.mlst.net) was obtained from the blood of a melioidosis patient with septicemia in Kaohsiung Veterans General Hospital, Taiwan. B. pseudomallei-GFP (expressing green fluorescence protein [GFP]) was made using a tri-parental mating system. The plasmid pKNOCK (wbiI-pfliC-gfp-cat) was constructed using specific PCR products: the truncated wbiI gene (ID: 3689613) and the fliC promoter [pfliC, 517 bp upstream of the coding region of fliC gene (ID:3688602)] from B. pseudomallei vgh19. The gfp gene was excised from the pUT-miniTn5-gfp-tet (AY364166) plasmid using appropriate restriction sites [35]. For conjugation, E. coli DH5α pir pKNOCK (wbiI-pfliC-gfp-cat, Cmr,Aps; 109 CFU) was used as a donor strain, E. coli pRK2013 (109 CFU; Cms, Aps) was used as a helper bacteria and B. pseudomallei vgh19 was the recipient (109 CFU; Cms, Apr); the three strains were mixed and filtered onto cellulose paper (<0.45 µm). B. pseudomallei-GFP was selected for using the chloramphenicol resistance and screened for using green fluorescence observed by fluorescent microscopy (Eclipse 50i; Nikon, Shinjuku, Tokyo, Japan). The specific insertion site was confirmed by PCR, sequencing and Southern hybridization. The genome of the GFP strain contained a single recombination event and generated an insertion of the pfliC-gfp-cat cassette into the wbiI gene, creating a defective wbiI gene and a functional wbiI gene (Figure S1).
BALB/c mice (females, 8 weeks old; n = 20 for each group) were intravenously infected via the tail vein with B. pseudomallei vgh19 or -GFP at the indicated concentration (100 µl; 5–5×105 CFU). Survival rates and disease scores were recorded daily. The clinical scores including neurological signs are described in Table S1.
At days 0, 2, 5 and 10 post-infection, serum was obtained from blood collected from each mouse by cardiac puncture to analyze cytokines and liver enzyme function (n = 6 for each experimental group, performed in duplicate). The liver enzymes glutamic oxaloacetic transaminase (GOT) and glutamic pyruvic transaminase (GPT) were measured using a clinical biochemistry analysis system (COBAS INTEGRATM 800, Roche, Basel, Switzerland).
The mouse serum cytokines TNF-α, IL-6, IL-12, IFN-γ, IL-10 and MCP-1 were measured using a Cytometric Bead Array kit (CBA; BD Biosciences, San Diego, CA, USA) according to the manufacturer's instructions [36]. Two-color cytometric data were collected at emission wavelengths of 423 nm and 578 nm by flow cytometry (BD Biosystems FACSCalibur system, BD Biosciences). Data analysis was performed using FCAP Array Software (version, 3.0; Bender Medsystems, Burlington, CA, USA), and standard curves were generated for each cytokine using the mixed cytokine/chemokine standard provided.
The spleen, liver, bone marrow (BM) and skull were excised, fixed in 4% formaldehyde, de-calcified with 10% trichloroacetic acid if the tissue contained bone and processed for paraffin wax embedding using standard techniques [37].
At the indicated times, the infected mice were sacrificed, and the solid organs (spleen, liver and brain) were excised. Bacterial loads (CFU/g) in the liver (ca. 0.5 g), spleen (ca. 0.02 g) and brain (ca. 0.4 g) were determined using a sequential weighing, homogenization in 500 µl PBS (phosphate buffered saline) and serial dilution protocol [38]. To remove contaminating intravascular blood and bacteria, the mice were perfused with PBS by cardiac injection (ca. 100 ml), and the bacterial burdens in the brain were determined. Blood (100 µl) was collected from the heart by cardiac puncture, and the bacterial concentration (CFU/ml) was measured by the serial dilution method followed by plate counts. To cultivate B. pseudomallei, both blood (100 µl) and CSF (cerebrospinal fluid, 5 µl) were seeded into culture bottles containing Ashdown's broth. CSF was collected from the cisterna magna by capillary (0.05 mm) puncture; the samples were discarded if contaminated with red blood cells. After a 7 d-incubation, the presence or absence of B. pseudomallei was observed. BM was aseptically isolated from the femur by removing both ends of the bone and flushing the interior 1 ml PBS/2% FCS (fetal calf serum) using a syringe. After flushing, the total numbers of bacteria (CFU/ml) in the BM were determined using serial dilution and plate counts. Bacterial burden data were averaged (mean ± standard deviation (SD)) from independent experiments (n = 6) within each condition in duplicate. If the bacterial load in the brain was very low as anticipation, the entire brain homogenate was plated. The limits of detection were 20, 50 and 3 CFU/g for liver, spleen and brain and 10 CFU/ml for blood and BM. The lower limit of CSF detection was 200 CFU/ml in this study.
Single cell suspensions from spleen and BM were prepared from infected mice at the indicated times (n = 3 for each experimental group performed in duplicate). Mononuclear cell layers in the spleen were separated by Ficoll-hypaque density gradient centrifugation (Sigma, St Louis, MO, USA). Red blood cells were lysed by treating with 0.83% NH4Cl for 3 min and then rapidly buffering with PBS containing 2% FCS. BM suspensions were prepared from the liquid obtained from flushing after filtering using nylon membranes (30 µm; Millipore, Billerica, MA, USA). All preparations were treated with kanamycin (400 µg/ml for 2 h) to remove extracellular B. pseudomallei. The number of splenic and BM cells harboring B. pseudomallei (CFU/106 cells, individual groups, n = 3, duplicate) was measured by plate counts after serial dilution.
The single cell spleen and BM preparations were obtained as described above. Peripheral blood mononuclear cells (PBMCs) were prepared using OptiLyse C solution (Beckman Coulter, Inc., Brea, CA, USA). These cells were stained with monoclonal PE (phycoerythrin)- or PE-Cy 7-conjugated anti-mouse specific antibodies (Table S2 contains the clones, isotype controls, species and concentration of antibody used in this study) for 30 min on ice in the dark according to the manufacturer's protocol. Stained cells were evaluated by flow cytometry (Cell Lab Quanta SC, Beckman Coulter, Inc.), and the flow cytometry data were analyzed using CXP software (Beckman Coulter, Inc.).
The CD11b+, Ly6C+ and CD19+ populations were isolated from the spleen or BM of infected mice using an EasySep Mouse Positive Selection Kit (n = 3 for each experimental group, duplicate; STEMCELL Tech. Inc., Vancouver, Canada). For each purification, the purity was confirmed to be >98% by flow cytometry. The CD11b− population was collected from the unbound fraction from cells repeatedly passed through a CD11b+ cell binding column and was >83% pure. The amounts of intracellular B. pseudomallei (CFU) in each population, if necessary, were determined by the serial dilution method as described above.
For adoptive transfer, the donor cells in this study were splenic or BM, CD11b+ or CD11b− and selectin-negative or selectin–expressing cells. CD11b+ or CD11b− splenic or BM cells were isolated from the mice on day 10 post-infection. The selectin-negative cells were prepared from mouse spleens on day 4 post-infection while selectin-expressing cells were prepared on day 10 post-infection. The isolation protocols for single cell suspensions or CD11b subpopulations from spleen and BM were performed as described above. Prior to adoptive transfer, the donor cells (ca. 104–106 cells) were adjusted to contain 50, 500 or 2000 CFU of cultivated B. pseudomallei. This adjustment based on the ratio of cultivatable bacteria to fluorescent cells (refer to Text S1, protocols and comments of adjusting donor cells and Figure S2, Scheme of the adoptive transfer protocols). The data were unavailable if intracellular B. pseudomallei within the donor cells used for adoptive transfer were determined to be >±15% of 50, 500 or 2000 CFU by plate counts. These donor cells were adoptively transferred to recipient mice (uninfected BALB/c mice, female, 8–10 week old; n = 3 or 6 for each experimental condition, duplicate) by intravenous injection. On day 2 post-transfer, the mice were sacrificed and the bacterial burdens in the brain, liver, spleen and BM, the bacterial cultures using blood and CSF, the concentration of serum cytokine (TNF-α, IL-1β, IL-6, IFN-γ, MCP-1) and the levels of liver enzymes (GOT and GPT) in sera were determined by the protocols described above.
Melioidosis with meningitis after adoptive transfer was evaluated by histological and bacteriological examination. On day 2 post-transfer, a total of 8 wax blocks containing brain tissue pieces cut from certain position for each skull were prepared from the mouse groups, including adoptive transfer of infected CD11b+, CD11b−, selectin-expressing and selectin-negative cells, respectively (n = 3 for each group, duplicate). After H&E staining (see above), histological changes were divided into intensive neutrophil infiltration (cellular debris), moderate neutrophil infiltration (no cellular debris) and vascular congestion with or without slight cell infiltration. The histological category of the recipient was determined according to their sections present in the most severe pathological features. Bacteriological examination was performed by collecting mouse CSF (5 µl) and testing for infection using bacterial culture techniques (as described above). After a 7-d incubation, the proportion (%) of samples with B. pseudomallei growth were calculated based on the outcomes of all tested mice.
All data are expressed as the means ± SD and were derived from independent mice (n = 3 or n = 6), performed in two independent experiments. Differences between two experimental groups were analyzed using an unpaired Student's t test (SPSS, version 19.0); p-values<0.05 were considered to be statistically significant.
The nucleotide sequences obtained in this study are available at the National Center for Biotechnology Information (NCBI) under the following accession numbers: wbiI (gene ID: 3689613), fliC promoter (517 bp upstream of the coding region of fliC, gene ID:3688602) and pUT-miniTn5-gfp-tet (GenBank: AY364166). The nucleotide sequence of the plasmid pKNOCK-Cm can be found in reference 35 or at http://gordonlab.wustl.edu/plasmids/pKNOCK-cm%20sequence.txt.
After intravenous injection of B. pseudomallei vgh19 (50 CFU), the survival of mice with melioidosis has two periods where death occurred (decline phases). Approximately 20% of the infected mice died within 4 days; the survivors rapidly declined and died on day 10 post-infection, with the exception of one mouse that died between 5 and 9 days (Figure S3). In this study, the overall progress of melioidosis was divided into three phases and their characteristics are summarized in Table S3.
During phase I, mice manifesting melioidosis developed a slightly hunched back, ruffled fur and showed decreased activity. The average clinical score was 6.56 (95% confidence interval [CI], 5.11–9.6). During phase II, the score decreased slightly to 4.2 (95% CI, 3.32–6.1); however, during phase III, 86.6% (13/15) of the mice developed severe disease with neurological signs (photophobia, ataxia and limb paresis or paralysis) as well as a rapid loss of body weight (up to 20% of the baseline) and a severely diminished activity level (clinical score, 18.5; 95% CI, 17.76–20.47). Cytokine levels (TNF-α, IL-6, IFN-γ, IL-10 and MCP-1, except for IL-12; Figure S4A) and liver enzyme activities (GOP and GPT; Figure S4B) in the infected mice were relatively high during phase I but lower during phase II. The levels of IFN-γ and MCP-1, both of which are involved in the activation and attraction of macrophages, increased during phase III.
Histological examination of infected mice sacrificed on days 2 (phase I), 5 (phase II) and 10 (phase III) post-infection were performed. Multiple microabscesses consisting primarily of degraded neutrophils and cellular debris were observed in the spleen (Figure 1A–C) and liver (Figure 1D–F) during all phases. During the progression of melioidosis, abscesses in both of these organs increased and expanded. Necrotic hepatocytes were present in phase II and expanded during phase III. Although normal hematogenous cells were found in the BM during phase I, cellular debris was present in the femur and the vertebrae during both phase II and III (Figure 1G–I). Immunological activity occurred in the central nervous system (CNS) during phase III as determined by histology. In the mice evaluated during phase III (n = 15), minimal neutrophil migration into the cerebral margins was noted in 86.7% (n = 13) of the infected mice (Figure 1J), and neutrophil infiltration into the meningeal subarachnoid space was observed in 86.7% (n = 13) of the mice (Figure 1K). Severe meningitis was found in 73.3% (n = 11) of mice, as indicated by darkly stained cells (neutrophils, lymphocyte, microglia or unidentified cells) accumulating in the meninges (Figure 1L). Suppurative meningitis (n = 5; Figure 1M), cerebellar microabscesses (n = 2; Figure 1N) and brain stem encephalitis (n = 3; Figure 1O) were occasionally observed.
The kinetics of bacterial burdens was determined by following a time course. These results indicate that the bacteria colonized the spleen, liver, BM and brain successively and plateaued on days 4 (spleen), 6 (liver), 8 (BM) and 12 (brain) post-infection (Figure 2A). Convincing evidence for substantial bacterial burden in the brain tissue during phase III was obtained by removing intravascular blood and bacteria by perfusion, revealing bacterial burdens of >103 CFU/ml in brain tissues on days 10 and 12 post-infection. Additionally, B. pseudomallei was isolated from mouse CSF after 10 d of infection. Bacteremia was not detected by blood cultures on days 3 and 5 post-infection; however, plate counts revealed densities of <103 CFU/ml after 5 d of infection (Figure 2A). Two mice with no neurological signs during phase III simultaneously showed no pathohistological changes in the CNS, negative CSF cultures and negative bacterial loads in the brain (data not shown). We also confirmed that the growth of intracellular bacteria occurred by day 2 in the spleen and by day 6 post-infection in the BM (Figure 2B). Taken together, these data suggest that the primary focus of infection was the spleen (or liver), from which the bacteria subsequently migrated to the BM and later to the brain.
High levels of bacteremia were not capable of inducing the early melioidosis-related meningitis. Neither bacterial brain colonization nor meningeal neutrophil infiltration were observed during day 4 post-infection in mice infected with 5×102, 5×103 or 5×104 CFU of B. pseudomallei. All infected mice rapidly died by 4 d of infection with a concentration of ≥5×103 CFU (n = 10, independent duplicate experiments).
Because bacterial burdens in the brain were observed after increases in the intracellular bacteria persisting in the spleen and BM, we hypothesized that a vector acting as a Trojan horse that was present in both organs induced melioidosis with CNS symptoms. We first determined the subpopulations of the spleen, BM and PBMCs that carried intracellular B. pseudomallei. Prior to this experiment, BALB/c mice were intravenously infected with B. pseudomallei-GFP, which was derived from B. pseudomallei vgh19. Both GFP and vgh19 strains have identical lethal doses and bacterial burden kinetics in the mouse melioidosis model (see Figure S5; Comparison of survival rates and bacterial burdens in organs from mice intravenously infected with B. pseudomallei vgh19 and GFP).
By analyzing the fluorescence of each population of phagocytic (CD11b+, Ly6C+, Ly6G+ and F4/80+) and lymphoid (CD3+, CD4+, CD8+, CD19+ and NK1.1+) cells (Figure 3, representatives on day 10 post-infection), it was found that both CD11b+ and Ly6C+ cells were the predominant fluorescent populations in the spleen, BM and PBMC. Distinctly lower amounts of fluorescence were observed in phagocytic Ly6G+ and F4/80+ cells and lymphoid CD19+ cells present in the spleen, BM and PBMC on days 2 and 10 post-infection. Fluorescent CD3+, CD4+, CD8+ and NK1.1 cells were not observed on days 2 and 10 post-infection (Summary in Table 1).
To confirm that B. pseudomallei persists in phagocytic or lymphoid cells, the amounts of intracellular bacteria were estimated by isolating splenic or BM CD11b+, Ly6C+ and CD19+ populations and performing plate counts of these populations after serial dilution. There were 1.0–2.23×103 CFU/106 cells in CD11b+, 0.43–0.53×103 CFU/106 cells in Ly6C+ and 0.06–0.12×103 CFU/106 cells in CD19+ populations from the spleen and BM.
It is know that the CD11b+ population includes inflamed monocytes (CD11b+Ly6Chigh), residential monocytes (CD11b+Ly6Clow), neutrophils (CD11b+Ly6G+), macrophages (CD11b+F4/80+) and B cells (CD11b+CD19+). Compared with uninfected mice, the levels of both the CD11b+Ly6Chigh and CD11b+Ly6Clow cells in the BM of infected mice were increased significantly, reaching 22.9% and 48.2%, respectively (for a total of 71.1%) on day 10 post-infection. In the spleen, the CD11b+Ly6Chigh cells predominated on day 4; however, this predominance shifted to CD11b+Ly6Clow cells on day 10 post-infection (Figure 4A). In contrast, on day 10 post-infection, macrophage (CD11b+F4/80+) levels were high in the BM and even higher in the spleens of infected mice compared with those of uninfected mice (Figure 4B). Neutrophils (CD11b+Ly6G+) were increased by 3.4 and 13.9-fold in the BM and spleen, respectively (Figure 4C), and CD11b+CD19+ B cells were increased by 5.25 and 3.4-fold in the BM and spleen on day 10 post-infection compared with their levels in uninfected mice (Figure 4D). These results indicate that both phagocytic and lymphoid CD11b+ cells in the BM and spleen were expanded on day 10 post-infection.
To determine whether the induction of bacterial infection in the brain was due to an infected cell used as a “Trojan horse”, the infected BM and splenic cells were isolated and adoptively transferred to healthy individuals. In this study, no adverse responses (deaths, signs of illness such as shivering or decreased activity; or pathohistological changes in the spleen, liver and brain) were observed after adoptive transfer of uninfected donor cells (104–106 cells) (data not shown). However, adoptive transfer of either infected BM or splenic cells increased the amount of brain-colonizing bacteria in a dose-dependent manner (Figure 5A). As in the above studies, we have demonstrated that the proportion of the CD11b+ population in the BM was higher than that in the spleen (Figure 4). Thus, we isolated CD11b+ and CD11b− populations from both organs and adjusted the isolated cells to obtain equal numbers of intracellular bacteria (50, 500 or 2000 CFU). Two days after adoptive transfer with CD11b+ cells from either the BM or spleen, significant bacterial burdens were observed in brains of the recipient mice (Figure 5B).
We demonstrated that extracellular bacteria doses (free B. pseudomallei) of 50 to 5000 CFU prepared from LB cultures were not able to cause bacterial infection in the brains by day 4 post-infection as described above. To rule out the possibility that the increased ability of the bacteria to colonize the brain is due to host adaptation in the donor mice, the infecting B. pseudomallei (50, 500 or 2000 CFU) were harvested from disrupted BM or splenic cells and subsequently injected intravenously. On day 2 post-injection, no colonizing bacteria were found in the brain (Figure 5A). However, the survival rates and the bacterial burdens were identical between mice infected with extracellular B. pseudomallei obtained from disrupted cells and the mice infected with B. pseudomallei from LB cultures (data not shown). The invasive ability of extracellular B. pseudomallei was not improved after adaptation in the host.
Because we found that CD11b+ cells carrying intracellular B. pseudomallei play a role in inducing bacterial colonization of the brain, we next measured the expression of selectin (CD62L) and integrin (CD18 or CD31) on CD11b+ cells during the progression of melioidosis because these molecules are involved in the transmigration of leukocytes. Our results revealed that the CD11b+CD62L+, CD11b+CD31+ and CD11b+CD18+ populations were substantially increased after 10 d of infection (Figure 6A, for representative data). Of the populations that varied between days 4 and 10 post-infection, both CD11b+CD31+ and CD11b+CD18+ populations showed mild increases ranging from 1.15 to 3-fold in the BM (Figure 6B) and spleen (Figure 6C); however, CD11b+CD62L+ population increased as much as 7.1-fold in the spleen (Figure 6C).
On day 4 post-infection, bacteria were not observed in the brain (Figure 2A); however, intracellular B. pseudomallei has persisted in the spleen (Figure 2B). Besides, the splenic CD11b+ population increased on day 4 post-infection (Figure 4A) and CD11b+ cells harboring B. pseudomallei isolated from the spleen (on day 10 post-infection) induced bacterial brain colonization (Figure 5B). Thus, we hypothesized that increased selectin expression on infected splenic cells is related to an increased bacterial burden in brain. We demonstrated that a majority of the splenic CD11b+ population lacked surface CD62L (defined as selectin-negative cells) on day 4 post-infection. However, the proportion of selectin-expressing cells increased to 19.7±3.5% on day 10 post-infection (Figure 7A). The bacterial burdens in the brains of recipients were rapidly increased with the adoptive transfer of CD11b+ selectin-expressing cells but not CD11b+ selectin-negative cells (Figure 7B). This result suggests that surface expressed L-selectin may be an important molecule involved in the development of B. pseudomallei brain infection.
After adoptive transfer of splenic CD11b+ or selecting-expressing cells (harboring 2000 CFU of cultivated B. pseudomallei), mouse melioidosis rapidly progressed (Table S4, Summary of the characteristics of melioidosis progression after adoptive transfer). Both histological and bacteriological analyses were performed to evaluate the development of melioidosis-associated meningitis (n = 3, each analysis was performed in duplicate). For histological sectioning, tissue blocks from each mouse were prepared in the same way (Figure 8A, positioning of the tissue blocks). The degrees of meningeal inflammation were determined based on neutrophil infiltration (i.e., intensive, moderate or rare infiltration) (Figure 8B, for representative histological images). In the mouse groups that received adoptively transferred infected splenic CD11b+ or selectin-expressing cells, the intensive neutrophil infiltration (17%, for the adoptively transferred group of CD11b+ cells; 33%, for the transferred group of selectin-expressing cells) and moderate neutrophil infiltration (67%, for both groups of CD11b+ and selectin-expressing cells) were observed in brain tissue (Figure 8C). The major foci were localized in the regions surrounding cerebral superior sagittal sinus for all infected mice, whereas brain abscesses and brain stem encephalitis were not observed. The observation of meningeal neutrophil infiltration with cellular necrosis agreed that bacterial burdens in the brains were rapidly increased after adoptive transfer of CD11b+ cells (Figure 5B) or selectin-expressing cells (Figures 7B). In contrast, in the mouse groups that received adoptively transferred infected splenic CD11b− cells or selectin-negative cells, the bacterial colonization of the brain was absent or rare (Figure 5B, 7B). The typical neutrophil infiltration into the meninges was also not observed in these mice, although vascular congestion with a few cells surrounding the blood vessels was noted (Figure 8C). The induction of melioidosis with meningitis was reconfirmed based on the growth of B. pseudomallei in the CSF of mice that received adoptively transferred infected CD11b+ or selectin-expressing cells. However, B. pseudomallei was never isolated from the CSF of mice that received adoptively transferred infected CD11b− or selectin-negative cells on day 2 post-transfer (Figure 8D).
The BALB/c mouse, a model that reflects the clinical course of human acute melioidosis [3], [4], [6], [10], [19], [20], was used to obtain reproducible manifestations of neurological melioidosis that includes aspects of leptomeningitis, meningoencephalitis, encephalomyelitis and brain abscesses secondary to remote infected foci. After adoptive transfer of infected cells, the initial meningeal foci were localized in the regions surrounding cerebral superior sagittal sinus because, on day 2 post-transfer, macroscopic brain abscesses, encephalomyelitis, brain stem encephalitis and even neutrophil infiltration in another meningeal place were not found. It has been suggested that B. pseudomallei infection in the brain is caused by bacteria traveling from the olfactory nerve to the brain, resulting in CNS symptoms after the initial intranasal infection [39]. However, inhalation, ingestion and subcutaneous inoculation, which were the possible melioidosis infection routes, subsequently resulted in systemic dissemination. Although bacterial colonization in the brain, neutrophil infiltration in the meninges and bacterial cultures from the CSF were characterized by intravenous infection in this study, the mechanisms of developing a CNS infection most likely differ if the infection was initiated through different means. Meningitis may be frequently overlooked because it is secondary to the primary hematogenous spread of the melioidosis associated with acute sepsis [15]. This study is the first analysis of the manifestations of melioidosis-associated meningitis in a mouse model.
Primary meningitis-causing bacteria such as Escherichia coli, group B Streptococcus and Neisseria meningitides commonly trigger translocation across the vascular endothelium via the binding of a ligand to a specific receptor [40]–[42]. Aside from this mechanism, it has been reported that monocytic Ly6C+ cells shelter intracellular Listeria monocytogenes. These cells, in a so-called Trojan horse mechanism, act as a vessel to allow the bacteria to cross into the brain [43]–[45]. In this study, melioidosis with meningitis exhibited the following characteristics: (1) Severe bacteremia was not necessary; (2) neutrophil infiltration into the meninges and bacterial burdens in the brain occurred during the second (phase III) but not the first (phase I) decline phase; (3) meningitis was induced by infected CD11b+ cells but not by extracellular (free) bacteria; and (4) selectin was involved in the onset of meningitis. These results imply that B. pseudomallei-infected cells likely act as a Trojan horse, an important cause of melioidosis with meningitis.
After primary B. pseudomallei infection, splenitis and hepatitis, including necrotic areas composed of degraded and viable macrophages and neutrophils, are commonly observed [10], [31], [37]. It has been reported that mice with fulminating melioidosis produce high levels of TNF-α, IL-6, IFN-γ and MCP-1, whereas infected but asymptomatic mice do not [6], [19], [46]. Our results revealed that, on day 5 post-infection, these inflammatory cytokines were significantly decreased in the serum; however, the expression of the macrophage activation- and attraction-related cytokines IFN-γ and MCP-1 increased on day 10 post-infection. Simultaneously, neutrophil/monocyte infiltration into the meninges was observed. This result supports previous reports indicating that MCP-1 mediates the recruitment of CCR2+ monocytes to the inflamed CNS [45], [47]. However, TNF-α and IL-6 play essential roles in the control of systemic B. pseudomallei infection [48], [49]. Additionally, IL-12 is essential for the production of IFN-γ. IFN-γ and/or bacterial lipopolysaccharide (LPS) are ligands known to induce the activation of highly microbicidal macrophages (M1) [50]. In this study, the lack of TNF-α, IL-6 and IL-12 induction on day 10 post-infection could result in an incomplete immune response against B. pseudomallei. Consequently, the growth of the intracellular bacteria was not restricted; instead, they burst their carrier cells and colonized the brain.
The timing of bacterial colonization in the spleen and BM was related to the expansion of splenic and BM CD11b+ cells, a predominant population that carried more intracellular B. pseudomallei than BM Ly6C+ and BM CD19+ cells. This result implies that B. pseudomallei exhibits different invasive tropisms toward phagocytic and lymphoid cells. B. pseudomallei is capable of persisting in neutrophils, macrophages and/or monocytes during early infection based on the presence of fluorescent cells in splenic CD11b+, Ly6C+, Ly6G+ and F4/80+ populations on day 2 post-infection. CD11b, a subunit of the adhesion molecule of Mac-1 (CD11b/CD18, αMβ2, CR3), is highly expressed on antigen-presenting cells, monocytes, neutrophils and B cells and is involved in the transmigration of these cells [51]–[53]. Generally, CD11b+ neutrophils are significantly expanded during gram-negative bacterial infection compared with gram-positive bacterial infection [54]. We have demonstrated that neutrophils (CD11b+Ly6G+) and monocytes (CD11b+Ly6Chigh and CD11b+Ly6Clow) are expanded in the BM and spleen during melioidosis. It is not clear how the initial inflammatory events in the brain occur; however, monocytes and neutrophils are increased in inflamed tissues in mice with melioidosis [19], [22], [23]. Thus, we suggest that the B. pseudomallei-loaded CD11b+ cells provided the factor necessary for the systemic dissemination of the infection via the bloodstream and the subsequent breach of the endothelial cells to infect the brain.
The Ly6Chigh population (inflamed monocytes) becomes expanded by infection with L. monocytogenes in the BM. All the L. monocytogenes-infected cells were CD11b+ monocytes (approximately 90% of the CD11b+ population simultaneously expressed Ly6C in the cases of listeriosis) [45]. These infected cells were released from the BM into the circulation and ultimately invaded the CNS [45], [55]–[57]. In this study, by following a bacterial burden time course, it appeared likely that the infected cells that induced melioidosis-associated meningitis originated from the BM. However, we found that B. pseudomallei persisted intracellularly in phagocytic and lymphoid cells. Indeed, adoptive transfer of infected BM CD11b− cells (most likely CD11b−Ly6C+ or CD11b−CD19+ cells) resulted in some degree of bacterial brain colonization. Additionally, lymphocytes were accumulated in the meninges based on histological examination. Moreover, it has been reported that B. pseudomallei enters lymphocytes and replicates intracellularly [58]. Thus, we suggest that all infected cells from the spleen and BM could simultaneously or synergistically contribute to meningitis.
In accordance with the Trojan horse mechanism, B. pseudomallei is shuttled to the brain within immune cells [34], [41], [59]. The cell adhesion molecules selectin (CD62L) and integrin (CD18 and CD31), expressed on CD11b+ cells, facilitate transmigration through the endothelial cells, and increases in the levels of these adhesion molecules facilitate the development of meningitis. In this study, mouse melioidosis with meningitis was induced by a low bacterial dose (50 CFU); however, compared with other organs, the spleen is a more permissive environment for bacterial replication in mice infected with lethal, sublethal or very low doses [10], [31]. We propose that splenic CD11b+ cells act as reservoirs, harboring intracellular B. pseudomallei during early infection. CD11b+Ly6C+CD62L+ cells are inflammatory, whereas CD11b+Ly6C−CD62L− cells are residential monocytes in the mouse spleen [60], [61]. Exacerbated melioidosis could develop in concert with meningitis when splenic inflammatory CD11b+ selectin-expressing cells gradually increase during disease progression.
|
10.1371/journal.ppat.1006851 | Phospholipid scramblase 1 interacts with influenza A virus NP, impairing its nuclear import and thereby suppressing virus replication | Transcription and replication of the influenza A virus (IAV) genome occur in the nucleus of infected cells and are carried out by the viral ribonucleoprotein complex (vRNP). As a major component of the vRNP complex, the viral nucleoprotein (NP) mediates the nuclear import of the vRNP complex via its nuclear localization signals (NLSs). Clearly, an effective way for the host to antagonize IAV infection would be by targeting vRNP nuclear import. Here, we identified phospholipid scramblase 1 (PLSCR1) as a binding partner of NP by using a yeast two-hybrid (Y2H) screen. The interaction between NP and PLSCR1 in mammalian cells was demonstrated by using co-immunoprecipitation and pull-down assays. We found that the stable overexpression of PLSCR1 suppressed the nuclear import of NP, hindered the virus life cycle, and significantly inhibited the replication of various influenza subtypes. In contrast, siRNA knockdown or CRISPR/Cas9 knockout of PLSCR1 increased virus propagation. Further analysis indicated that the inhibitory effect of PLSCR1 on the nuclear import of NP was not caused by affecting the phosphorylation status of NP or by stimulating the interferon (IFN) pathways. Instead, PLSCR1 was found to form a trimeric complex with NP and members of the importin α family, which inhibited the incorporation of importin β, a key mediator of the classical nuclear import pathway, into the complex, thus impairing the nuclear import of NP and suppressing virus replication. Our results demonstrate that PLSCR1 negatively regulates virus replication by interacting with NP in the cytoplasm and preventing its nuclear import.
| Influenza viral RNA is encapsidated by three polymerase proteins and the NP protein to form the vRNP complex, which is transported to the nucleus of infected cells for viral transcription and replication. The active nuclear import of the vRNP complex is mediated by the interaction between NP and importin α through the nuclear import pathway. Because the interactions between NP and the components of the nuclear import pathway are indispensable in mediating the nuclear import of the vRNP complex, the host has evolved mechanisms to antagonize influenza virus infection that target this crucial step. In this study, we identified PLSCR1 as an interacting partner of the influenza NP protein. We found that PLSCR1 negatively regulates influenza virus replication by inhibiting the nuclear import of the NP/vRNP complex. Importantly, we found that PLSCR1 did not disrupt the interaction between NP and importin α. Instead, NP, PLSCR1, and importin α formed a stable complex that blocked the interaction between importin α and importin β, thereby inhibiting the import of NP/vRNP complex through the nuclear import pathway. Our findings provide an example of a host restriction factor binding simultaneously to a nuclear import adaptor and to a cargo protein to inhibit the import of that cargo into the nucleus.
| Influenza A virus (IAV), a single-stranded, negative-sense RNA virus with an eight-segmented genome, is the causative agent of influenza in many animal species, including humans. Inside the virion, all eight viral RNA (vRNA) segments bind to the three RNA polymerases (polymerase basic protein 2, PB2; polymerase basic protein 1, PB1; and polymerase acidic protein, PA) and are encapsidated by the nucleoprotein (NP) to form viral ribonucleoprotein (vRNP) complexes [1]. The vRNP complex is the essential functional unit for the transcription and replication of the IAV genome [2]. Electron microscopy of isolated vRNPs has shown that both ends of the vRNA interact with each other to form a circular or supercoiled structure and that the RNA polymerase interacts with both ends of the vRNA segment [2–4]. The rest of the vRNA is encapsidated by the NP protein with approximately 24 nucleotides per molecule [5].
A prominent feature of the IAV life cycle is that the transcription and replication of the viral genome occur in the nucleus of infected cells [6, 7]. During the early phase of virus infection, after completion of endocytosis and uncoating, the vRNP complex is released into the cytoplasm and is translocated to the nucleus, which is mediated by the nuclear localization signals (NLSs) of the NP protein [8]. Two amino acid sequences have been identified as NLSs for the NP protein: an unconventional NLS in the N-terminus (residues 3 to 13; NLS1) [9, 10], and a bipartite NLS (residues 198 to 216; NLS2) [11]. The unconventional NLS appears to be the major determinant for NP nuclear import [12]. NP relies on the classical nuclear import pathway to enter the nucleus of infected cells. In this pathway, importin α functions as an adaptor by recognizing NLS sequences in cargo proteins and associating with the importin β receptor [13, 14]. Through a process that involves multiple rounds of interaction between importin β and nucleoporins of the nuclear pore complex (NPC), the trimeric importin α/β-cargo complex translocates into the nucleus [15]. NP interacts with various isoforms of importin α, including importin α-1, -3, -5, and -7 [10, 16, 17]. Previous studies have shown that the nuclear import of vRNP and newly synthesized NP to the nucleus of infected cells is a crucial step in the IAV life cycle [12, 18]. In addition to the central role played by importin α/β in modulating the nuclear transport of NP, host proteins could also be involved, such as α-actinin-4, Hsp40, or MOV10, which may promote or inhibit this active process [17, 19, 20]. However, the detailed mechanism that regulates the migration of vRNP complexes and newly produced NP into the nucleus remains obscure, and the identification of the potential host factors involved is not yet complete.
Phospholipid scramblase 1 (PLSCR1) was first identified in erythrocyte membranes, where it was activated under conditions of elevated calcium, resulting in disruption of phospholipid asymmetry across the plasma membrane [21]. Its function in remodeling the distribution of plasma membrane phospholipids in mammalian cells is still controversial because increase in PLSCR1 expression and gene depletion of PLSCR1 can occur in response to calcium without affecting the transmembrane movement of phospholipids [22–24]. In addition to its unresolved role as a scramblase, PLSCR1 appears to be involved in multiple biological processes. Several studies have shown that PLSCR1 plays a critical role in cellular maturation and terminal differentiation: PLSCR1 expression is markedly increased during the terminal differentiation of the monocytic and granulocytic lineages of hematopoietic precursor cells [25–27], and gene deletion of PLSCR1 in mice was found to impair the differentiation of hematopoietic precursor cells into mature granulocytes in response to select hematopoietic growth factors [24]. Although initially identified as a transmembrane protein, PLSCR1 also contains a nonclassical NLS and can be imported into the cell nucleus [28]. The nucleus-localized PLSCR1 can directly bind to the promoter region of the inositol 1,4,5-triphosphate receptor type 1 gene (IP3R1) to enhance its expression [29, 30], and can also interact with angiogenin (ANG) in the nucleus to positively regulate rRNA transcription [31]. Another important function of PLSCR1 is as an effector of the interferon (IFN) signaling pathway. PLSCR1 interacts with Toll-like receptor 9 (TLR9) and regulates its trafficking from the endoplasmic reticulum (ER) to the endosomal compartment in plasmacytoid dendritic cells (pDCs) [32], which is an important step in IFN production in pDCs. PLSCR1 harbors an IFN-stimulated response element in its first exon [23], is induced by IFN-α, -β, and -γ [23, 33, 34], and can enhance the expression of a subset of IFN-stimulated genes (ISGs) in response to IFN-β treatment to inhibit the replication of vesicular stomatitis virus (VSV) and encephalomyocarditis virus (EMCV) [33]. PLSCR1 mediates IFN-α-induced protection against staphylococcal α-toxin [35], is a main effector of IFN-γ-mediated antiviral activity against Hepatitis C virus (HCV) [34], and can also inhibit the replication of Hepatitis B virus (HBV) and human T-cell leukemia virus type-1 (HTLV-1) [36, 37].
In the present study, we discovered that the interaction between IAV NP and cellular PLSCR1 occurs in both transfected and infected mammalian cells. Importantly, overexpression of PLSCR1 significantly suppressed IAV replication, whereas siRNA knockdown or CRISPR/Cas9 knockout of PLSCR1 expression increased the virus titer, thereby demonstrating that PLSCR1 is a host restriction factor for IAV infection. We further found that PLSCR1 inhibited NP nuclear import and caused retardation of the virus life cycle. Strikingly, PLSCR1 formed an integrative complex with NP and different members of the importin α family, which inhibited the incorporation of importin β into the complex and impaired the import of NP via the nuclear import pathway.
To identify host cellular proteins that interact with influenza virus NP protein, we employed the yeast two-hybrid system to screen a cDNA library generated from a mixed human cell culture (A549, HEK293T, THP-1, and U251) as described previously [38]. The full-length NP protein from A/Anhui/2/2005 (AH05, H5N1) was used as bait. Putative positive clones were obtained after selection on QDO/X/A (Ade/–His/–Leu/–Trp/X-a-Gal/AbA) plates. After growing the putative positive clones in DDO (SD/−Leu/−Trp) medium, plasmids were isolated and sequenced to identify the potential NP interactants. One specific clone from this screen was found to contain the full-length open reading frame of PLSCR1 (GenBank accession no. NM_021105). The interaction between PLSCR1 and NP was then retested by yeast co-transformation, as described in the Materials and Methods. As shown in Fig 1, PLSCR1 specifically interacted with NP in yeast.
To further examine the PLSCR1-NP interaction, we performed co-IP experiments. HEK293T cells were transfected with V5-tagged WSN NP and Flag-tagged PLSCR1, individually or in combination. Cell lysates were immunoprecipitated with an anti-V5 mAb, followed by western blotting with rabbit pAb against V5 or the Flag tag (Fig 2A). Flag-tagged PLSCR1 was coimmunoprecipitated with V5-tagged NP of A/WSN/33 (WSN, H1N1) virus when they were coexpressed, but not in the absence of WSN NP, indicating that PLSCR1 interacts with influenza NP in mammalian cells. When a reverse co-IP experiment was performed with an anti-Flag mAb, V5-tagged WSN NP was also coimmunoprecipitated with Flag-tagged PLSCR1 (Fig 2B), further demonstrating the specificity of the NP-PLSCR1 interaction. The PLSCR1-NP interaction was also confirmed in a GST pull-down assay. WSN NP was pulled down by GST-PLSCR1, but not by GST alone (right panel, Fig 2C). Similarly, PLSCR1 was only pulled down by GST-WSNNP (right panel, Fig 2D).
A key function of influenza NP protein during the virus life cycle is to encapsidate viral RNA to form the vRNP complex in preparation for transcription, replication, and packaging [39]. To examine whether the interaction between PLSCR1 and NP is dependent on the RNA-binding activity of NP, we performed a co-IP assay with cell lysates that were first treated with 100 μl of RNase A/T1 (Fig 2E). Flag-PLSCR1 was still coimmunoprecipitated with V5-WSNNP, indicating that the interaction between PLSCR1 and NP did not rely on the RNA binding activity of NP.
We performed an additional co-IP experiment in A549 cells that were mock infected or infected with WSN virus at an MOI of 5. At 6 h post infection (p.i.), cell lysates were immunoprecipitated with a rabbit pAb against PLSCR1, followed by western blotting with a rabbit anti-PLSCR1 pAb for the detection of PLSCR1 and a mouse anti-NP mAb to reveal the presence of NP (Fig 2F). The results showed that WSN NP interacted with PLSCR1 during the natural viral infection.
We then attempted to define the region of NP that was critical for its binding with PLSCR1. We generated five truncated NP constructs (NP1-80, NP1-162, NP1-271, NP1-351, and NP268-498), which were fused to the C-terminus of GST, and then examined their interaction with PLSCR1 in HEK293T cells. We found that all five truncated versions of NP were well expressed, although there were differences in their expression levels (Fig 2G). The pull-down assay showed that NP1-271 and NP1-351 exhibited strong binding to PLSCR1. In contrast, the two short N-terminal NP mutants, NP1-80 and NP1-162, almost lost their ability to interact with PLSCR1. Further, the interaction between the C-terminal NP mutant, NP268-498, and PLSCR1 was also dramatically decreased compared with that of NP1-271 and NP1-351. These results indicate that neither the N-terminal nor the C-terminal region of NP is critical for its interaction with PLSCR1; rather, the middle region of NP is likely involved in the interaction with PLSCR1.
To study the role of the PLSCR1-NP interaction during the virus life cycle, we analyzed the effect of upregulating PLSCR1 on virus replication. We transduced A549 cells with a retrovirus encoding PLSCR1 to establish a stable PLSCR1-overexpressing cell line or with an empty retrovirus as a control cell line. As expected, PLSCR1 expression at both the mRNA and protein level was increased in PLSCR1-overexpressing cells compared with the empty retrovirus-transduced control cells (Fig 3A and 3B). The control and PLSCR1-overexpressing A549 cells were infected with WSN virus at an MOI of 0.1. Culture supernatants were collected at different timepoints after infection and titrated on MDCK cells. Strikingly, PLSCR1 overexpression led to a 20- to 100-fold decrease in virus titers at 12–48 h p.i. (Fig 3C). Similar reductions in virus titers were observed for influenza viruses AH05 (H5N1) (Fig 3D), A/Anhui/1/2013 (H7N9) (Fig 3E) and A/Fuzhou/1/2009 (H1N1) (Fig 3F) at both 24 h and 48 h p.i. In a separate experiment, we examined NP and PLSCR1 expression in WSN virus-infected cells at timepoints between 0 and 48 h p.i. (Fig 3G). In control A549 cells, viral NP expression was abundant at 12 h p.i., and remained high until 48 h. In clear contrast, less NP was detected at 24 h p.i., with somewhat more detected at 48 h p.i. in PLSCR1-overexpressing A549 cells. Moreover, the expression of PLSCR1 remained unchanged in virus-infected PLSCR1-overexpressing A549 cells, whereas in control A549 cells, the expression of PLSCR1 was upregulated at 6 h p.i., increased at 12 h p.i., and remained elevated at 48 h p.i. Together, these data indicate that endogenous expression of PLSCR1 is strongly induced by influenza virus infection, and stable overexpression of PLSCR1 significantly inhibits NP expression and virus replication.
We further analyzed the effect of PLSCR1 downregulation on IAV infection by means of small interfering RNA (siRNA)-mediated silencing. Real-time PCR and western blotting confirmed that the expression of PLSCR1 was significantly reduced in specific siRNA-treated A549 cells but not in cells treated with scrambled siRNA (Fig 3H and 3I). PLSCR1 downregulation had no major effect on cell viability as measured by a luminescent cell viability assay (Fig 3J). A549 cells treated with siRNA targeting PLSCR1 or with scrambled siRNA were infected with WSN virus. Culture supernatants were collected at 24 and 48 h p.i. and titrated on MDCK cells. As shown in Fig 3K, knockdown of PLSCR1 by specific siRNA increased the virus titer compared with that in scrambled siRNA-treated A549 cells. We further generated a PLSCR1-KO HEK293T cell line by using the CRISPR/Cas9 system. The knockout of PLSCR1 was confirmed by western blotting with a rabbit anti-PLSCR1 pAb (Fig 3L). The PLSCR1-KO HEK293T or control cells were infected with WSN virus at an MOI of 0.1, and the supernatants collected at 24 and 48 h p.i. were titrated on MDCK cells. As shown in Fig 3M, the titers of WSN virus in PLSCR1-KO HEK293T cells were dramatically increased compared with those of the control cells. Together, these data demonstrate that PLSCR1 negatively regulates IAV replication via its interaction with NP.
The cellular distribution of NP during the virus life cycle was investigated by using a time-course experiment in both PLSCR1-overexpressing and empty retrovirus-transduced control A549 cells infected with WSN virus at an MOI of 5. In the control A549 cells, NP had clearly accumulated in the nucleus of approximately 45% of cells at 4 h p.i (Fig 4A and 4C). By 6 h p.i., the percentage of cells with NP in the nucleus had increased to 54%. In addition, NP localized at both the edge of nucleus and the cytoplasm of 25% of the infected cells, an indication of vRNP export from the nucleus. At 8 h p.i., the distribution of NP was mixed, with approximately 37%, 32%, and 14% of cells showing clear nuclear localization, simultaneous localization at both the edge of the nucleus and the cytoplasm, and exclusive cytoplasmic distribution, respectively. At 10 h p.i., the newly synthesized vRNP complex was largely exported from the nucleus into the cytoplasm, as indicated by the cytoplasmic distribution of NP in 60% of the cells. NP was also primarily localized close to the cytoplasmic membrane in 15% of the cells. By 12 h p.i., the percentage of cells with NP distributed close to the cytoplasmic membrane was 90%, indicating that vRNP export was largely complete and active assembly and budding were underway. The endogenous PLSCR1 was predominantly localized in the cytoplasm of the control A549 cells throughout the observation period from 4 to 12 h p.i (Fig 4A). The colocalization of NP and PLSCR1 appeared in cells with obvious cytoplasmic distribution of NP at 8 h p.i. At 10 and 12 h p.i., a large amount of newly synthesized vRNP complex was visualized in the cytoplasm or close to the cytoplasmic membrane, where obvious colocalization of NP and PLSCR1 was observed. In comparison with the control A549 cells, the virus life cycle was significantly delayed in the PLSCR1-overexpressing A549 cells (Fig 4B and 4C). At 4 h p.i., NP did not accumulate in the nucleus of any of the visualized cells, suggesting that the nuclear import of the vRNP was inhibited by the overexpressed PLSCR1. At 6 h p.i., only approximately 6% of cells showed clear nuclear accumulation of NP. At 8 h p.i., NP had clearly accumulated in the nucleus of approximately 16% of cells, and roughly 3% of cells had NP at both the edge of the nucleus and the cytoplasm. At 10 h p.i., NP showed clear nuclear localization, simultaneous localization at both the edge of the nucleus and the cytoplasm, and exclusive cytoplasmic distribution in 25%, 5%, and 2% of cells, respectively. At 12 h p.i., the NP distribution pattern was similar to that at 10 h p.i. except that the percentages of cells showing clear nuclear localization, simultaneous localization at both the edge of the nucleus and the cytoplasm, and exclusive cytoplasmic distribution were further increased to 31%, 15%, and 6% respectively. As in the control A549 cells, PLSCR1 was almost exclusively localized in the cytoplasm of the PLSCR1-overexpressing cells (Fig 4B). In addition, co-localization of NP and PLSCR1 was detected in the cytoplasm of cells exhibiting considerable vRNP export at 10 and 12 h p.i. Taken together, these data demonstrate that overexpression of PLSCR1 significantly inhibits the nuclear import of the vRNP complex, and causes dramatic retardation of the virus life cycle.
At the early timepoints (i.e., 4 and 6 h p.i. in control cells, and 4, 6, and 8 h p.i. in PLSCR1-overexpressing cells), co-localization of NP and PLSCR1 was not observed in the cytoplasm, most likely because of the relatively low abundance of the vRNPs in the cytoplasm prior to their import into the nucleus. We attempted to determine whether the interaction between NP and PLSCR1 could directly inhibit the import of NP into the nucleus. To this end, we transfected A549 cells with a pCAGGS-WSNNP construct together with either pCAGGS-PLSCR1 or the empty vector. The localization of NP and PLSCR1 was visualized at 20 h post-transfection. As shown in Fig 4D, NP clearly accumulated in the nucleus of cells without exogenous PLSCR1 expression. In contrast, NP was predominantly retained in the cytoplasm and colocalized with PLSCR1 when PLSCR1 was substantially overexpressed. These results further confirm that PLSCR1 inhibits the import of NP into the nucleus.
We next validated the inhibitory effect of PLSCR1 on the nuclear import of NP with a cell fractionation experiment. The PLSCR1-overexpressing or empty retrovirus-transduced control A549 cells were infected with WSN virus at an MOI of 5. At 6 h p.i., the infected cells were lysed. The cytoplasmic and nuclear fractions were separated and subjected to western blotting. As shown in Fig 4E, the marker proteins GAPDH and LaminB1 were only detected in the cytoplasm and nucleus, respectively. PLSCR1 almost exclusively localized in the cytoplasm in both PLSCR1-overexpressing and control cells. A considerable amount of NP was detected in both the nucleus and the cytoplasm of the control cells. In contrast, NP was primarily detected in the cytoplasm and was only weakly detected in the nucleus of the PLSCR1-overexpressing cells.
We further investigated the inhibitory role of PLSCR1 on the nuclear import of incoming vRNPs by treating PLSCR1-overexpressing A549 cells or control A549 cells with cycloheximide (CHX) to inhibit protein synthesis. The treated cells were infected with WSN virus at an MOI of 5, and were separated into nuclear and cytoplasmic fractions at 2 h p.i., followed by western blotting to detect the NP in the nuclear and cytoplasmic fractions. We found that most of the NP was detected in the nucleus of the control A549 cells; however, in PLSCR1-overexpressing A549 cells, most of the NP was detected in the cytoplasm and NP was only weakly detected in the nucleus (Fig 4F). Since the only source of NP protein was from the incoming vRNPs under CHX treatment, this experiment demonstrates that PLSCR1 directly inhibits the nuclear import of incoming vRNPs.
Collectively, these results demonstrate that the expression of PLSCR1 suppresses the nuclear accumulation of NP/vRNP, thus inhibiting the virus life cycle.
We hypothesized that viral RNA transcription and replication would be impaired due to the retention of vRNP and NP in the cytoplasm caused by PLSCR1 expression. To test this hypothesis, we transfected HEK293T cells with specific siRNA targeting PLSCR1 or with scrambled siRNA for 48 h. Western blotting analysis showed that specific siRNA treatment indeed downregulated the expression of PLSCR1 (Fig 5A). The siRNA-treated cells were then transfected with protein expression constructs of the RNP complex proteins (PB2, PB1, PA, and NP), along with a reporter plasmid containing the terminal coding and noncoding sequences from the NS segment and the luciferase gene driven by the human RNA polymerase I promoter and terminator. Forty-eight hours later, the luciferase activity of the cell lysates was measured to reveal the RNP activity. We found that the RNP activity was increased by approximately 16-fold when the expression of PLSCR1 was knocked down by specific siRNA compared with that in scrambled siRNA-treated cells (Fig 5B), indicating that the endogenous PLSCR1 inhibited the transcription and replication of the viral genome.
To further determine the steps of viral transcription and replication that were affected by PLSCR1 expression, we infected the PLSCR1-overexpressing and empty retrovirus-transduced control A549 cells with WSN virus at an MOI of 5. At 6 and 10 h p.i., vRNA, mRNA, and cRNA derived from segment 5 were measured by quantitative reverse transcription PCR (RT-qPCR). At both timepoints, the levels of all three species of viral RNA were found to be significantly decreased in the PLSCR1-overexpressing cells compared with those in the control cells (Fig 5C and 5D). Among the three species of viral RNA, the reduction in the vRNA level was the most sizeable. This finding could indicate that vRNA synthesis occurred after the synthesis of the mRNA and cRNA, and reductions in the synthesis of mRNA and cRNA would lead to an accumulative defect in the amplification of the vRNA species.
The nuclear import of influenza NP protein can be regulated via its phosphorylation status [9, 40]. We therefore determined whether the effect of PLSCR1 on the import of NP is achieved by modulating NP phosphorylation. We infected either PLSCR1-overexpressing or empty retrovirus-transduced control A549 cells with WSN virus at an MOI of 5. At 6 and 8 h p.i., the NP and PLSCR1 expression levels in the infected cells were determined by western blotting (Fig 6A). In PLSCR1-overexpressing cells, the level of PLSCR1 remained relatively constant between the two timepoints, whereas the expression of viral NP protein was increased at 8 h compared with that at 6 h p.i. In contrast, in control A549 cells, the increase in PLSCR1 expression was more obvious than that of NP between the two timepoints. We then performed an immunoprecipitation experiment by using an anti-NP mAb to reveal the level of total NP, an anti-p-Ser mAb to determine the level of serine-phosphorylated NP, and an anti-p-Tyr mAb to detect the level of tyrosine-phosphorylated NP. At 6 h p.i., NP was clearly serine- and tyrosine-phosphorylated in control cells, and the extent of NP phosphorylation was further increased at 8 h p.i. In PLSCR1-overexpressing cells, NP phosphorylation was barely detectable at 6 h p.i., the timepoint when the expression of total NP was only weakly detected. In contrast, a considerable amount of NP was phosphorylated at 8 h p.i. In general, NP was less phosphorylated at both timepoints in the PLSCR1-overexpressing cells compared with the control cells. However, the proportion of phosphorylated NP of the total NP, as indicated at 8 h p.i., was similar between the PLSCR1-overexpressing cells and the control cells. These data indicate that overexpression of PLSCR1 does not affect the phosphorylation status of the viral NP protein.
Two phosphorylation sites in NP, S9 and Y10, are highly conserved among all influenza A viruses [41]. Mutations that abolish these two sites have been shown to significantly reduce virus replication [40, 41]. Here, we examined the growth properties of a phosphorylation mutant, S9A/Y10F, in PLSCR1-overexpressing A549 cells. As shown in Fig 6B, we observed an additive effect of the inhibitory role of PLSCR1 overexpression and that of mutation of key NP phosphorylation sites on virus replication. The replication of wild-type WSN virus was significantly inhibited in PLSCR1-overexpressing A549 cells compared with control cells (Fig 6B). Moreover, the replication of the phosphorylation mutant S9A/Y10F was decreased further in PLSCR1-overexpressing cells than in control cells. These results indicate that regardless of the phosphorylation status of the NP residues S9 and Y10, PLSCR1 overexpression consistently reduced virus replication, implying that the inhibitory role of PLSCR1 on virus replication is not played by influencing the phosphorylation status of NP.
PLSCR1 can potentiate the antiviral activity of IFN when exogenous IFN is present [33]. We therefore investigated the possibility that PLSCR1 indirectly inhibits virus infection by stimulating the IFN pathway. To this end, we measured the protein level of Mx1, a key antiviral effector protein of the IFN pathway, in both PLSCR1-overexpressing A549 cells and control A549 cells. As shown in Fig 6C, Mx1 expression was not detectable in either PLSCR1-overexpressing cells or control cells when they were not treated with IFN-α. In contrast, IFN-α treatment efficiently induced the expression of Mx1 protein in both types of cells. However, no difference in the level of Mx1 expression was observed between PLSCR1-overexpressing cells and control cells when they were treated with IFN-α. Therefore, the overexpression of PLSCR1 did not result in observable changes in the expression of Mx1 relative to the control cells. We also determined the luciferase activity of HEK293T cells transfected with an ISRE luciferase reporter gene, together with a PLSCR1 expression construct or an empty vector (Fig 6D). We found that the overexpression of PLSCR1 did not increase the expression of the ISRE luciferase reporter gene compared with that of the PLSCR1-non-overexpressing cells. Together, these results demonstrate that the inhibitory role of PLSCR1 in influenza virus replication does not involve stimulating the IFN pathways.
Importin α plays an important role in the nuclear import of proteins [42]. PLSCR1 has been reported to interact with importin α [28]. We therefore attempted to determine whether the interaction between PLSCR1 and NP interferes with the complex formation between NP and importin α, thus preventing the nuclear import of NP through the classical nuclear import pathway. To this end, we transfected HEK293T cells with V5-tagged NP and Myc-tagged importin α proteins, together with gradually increasing amounts of Flag-tagged PLSCR1 construct. At 48 h post-transfection, cell lysates were immunoprecipitated with an anti-Myc mAb, followed by western blotting with rabbit pAbs against Myc, V5, or the Flag tag to detect importin α, NP, and PLSCR1, respectively. As shown in Fig 7A, V5-tagged NP was coimmunoprecipitated with Myc-tagged importin α1 when they were coexpressed, but not in the absence of Myc-tagged importin α1, indicating that NP interacts with importin α1 in mammalian cells. When increasing amounts of Flag-tagged PLSCR1 construct were co-transfected with V5-tagged NP and Myc-tagged importin α1, the expression level of PLSCR1 also gradually increased in the cell lysates. More PLSCR1 was detected in the importin α1 immunoprecipitates as the amount of transfected PLSCR1 construct increased from 0.2 to 0.6 μg. Significantly, V5-tagged NP and Flag-tagged PLSCR1 were simultaneously co-immunoprecipitated with Myc-tagged importin α1, and the amount of NP coimmunoprecipitated with importin α1 was not reduced when the amount of transfected PLSCR1 was increased. This result indicates that the presence of PLSCR1 did not affect the formation of the complex between NP and importin α1; rather, a trimeric complex of PLSCR1, NP, and importin α1 was formed. We then performed similar co-IP experiments with PLSCR1, NP, and other members of the importin α family. V5-tagged NP and Flag-tagged PLSCR1 were coimmunoprecipitated with Myc-tagged importin α3 (Fig 7B), importin α5 (Fig 7C), or importin α7 (Fig 7D), and the gradually increased expression of PLSCR1 did not reduce the interaction between NP and these members of the importin α family. To validate this finding, we included the host factor MOV10 as a control in the co-IP experiment, because MOV10 has been shown to compete with importin α to interact with NP [17]. We found that co-expression of MOV10 reduced the amount of NP coimmunoprecipitated with importin α1, indicating that MOV10 indeed inhibited the interaction between importin α and NP (Fig 7E). In contrast, the expression of PLSCR1 did not affect the binding between importin α and NP. These data clearly indicated that the NP protein bound by MOV10 was no longer bound by importin α, but the NP protein bound by PLSCR1 could still bind to importin α. Taken together, these results demonstrate that PLSCR1 forms an integrative three-subunit complex with NP and importin α.
Thus, our results demonstrated that PLSCR1 forms a complex with NP and importin α and causes cytoplasmic retention of NP. We then performed another co-IP experiment to further reveal the underlying mechanism. HEK293T cells were transfected with plasmids expressing V5-WSNNP, Myc-importin α1, importin β, or together with Flag-PLSCR1. The cell lysates were immunoprecipitated with a mouse anti-Myc mAb, and the bound proteins were detected by western blotting with rabbit pAb against V5, Myc, Flag tag, or importin β. Strikingly, PLSCR1 expression significantly reduced the amount of importin β in the immunoprecipitates (Fig 7F), indicating that the formation of the complex of PLSCR1, NP, and importin α1 blocked the access of importin β, the key mediator of the classical nuclear import pathway, to the complex, thereby inhibiting the nuclear import of NP via the classical nuclear import pathway and suppressing virus replication.
The vRNP complex is responsible for the transcription and replication of the influenza viral genome in the nucleus of infected cells [16, 18, 43]. In addition to the three polymerase subunits with one copy of each, most of the vRNP complex is encapsidated by the viral NP protein [44, 45]. As the most abundant protein in the virus particle, except for M1 [46], NP inevitably becomes the main target of the host defense system. In this study, we identified the host cellular protein PLSCR1 as an interacting partner of the NP protein by using yeast two-hybrid screening. We demonstrated that NP and PLSCR1 interact in both transfected and infected mammalian cells. Western blotting analysis showed that PLSCR1 did not affect the phosphorylation status of NP. Instead, we found that PLSCR1 formed a complex with NP and members of the importin α family, inhibited nuclear import of vRNP/NP, and thereby suppressed virus genome transcription and replication and negatively regulated the propagation of different influenza virus subtypes.
The active nuclear import of the vRNP complex is mediated by an interaction between NP and importin α through the classical nuclear import pathway [10, 16, 17]. Host factors are reported to be involved in this active process. One such factor, Hsp-40, has been shown to be required for the efficient association between NP and importin α, thus promoting the nuclear localization of vRNP complex [20]. In contrast, MOV10 was found to disrupt the binding between NP and importin α, thereby causing the retention of NP in the cytoplasm and a reduction in virus replication [17]. In the present study, we found that the nuclear import of the vRNP complex was significantly retarded in virus-infected PLSCR1-overexpressing A549 cells compared with empty retrovirus-transduced control cells. Furthermore, NP clearly accumulated in the nucleus of cells that were not transfected with the PLSCR1 construct, whereas NP was predominantly retained in the cytoplasm and colocalized with PLSCR1 when PLSCR1 was significantly overexpressed by transfection. These results demonstrate that PLSCR1 inhibited the import of the NP/vRNP complex into the nucleus. Interestingly, we found that the inhibitory effect of PLSCR1 on the nuclear import of the NP/vRNP complex was not achieved by impairing the interaction between NP and importin α. Instead, NP, PLSCR1, and importin α formed a stable complex, which inhibited the interaction between importin α and importin β. Taken together, our findings favor a model in which influenza virus NP, derived from the newly synthesized NP or incoming vRNP, is bound by the heterodimeric import receptor, importin α/importin β, in the cytoplasm and is transported into the nucleus; in the presence of PLSCR1, the complex formed among NP, PLSCR1, and importin α in the cytoplasm prevents importin α from forming a functional nuclear import receptor complex with importin β, thereby suppressing the nuclear import of NP (Fig 8). We speculate that the simultaneous binding of two molecules by importin α may overload this nuclear import adaptor, or alter its structural property, thereby affecting its ability to interact with importin β.
As an IFN-inducible gene, PLSCR1 can enhance the antiviral activity of IFN by increasing the expression of IFN-stimulated genes [33]. In the presence of exogenous IFN-β, PLSCR1 has been shown to increase its antiviral activity against VSV and EMCV [33]. PLSCR1 has also been shown to inhibit HBV replication by reducing the synthesis of viral proteins, DNA replicative intermediates, and viral RNAs [36]. In addition to these indirect antiviral activities, PLSCR1 can directly bind to the Tax protein of HTLV-1 to reduce its transactivation activity by altering the subcellular distribution and homodimerization of Tax [37]. In the present study, we found that PLSCR1 directly binds to the NP protein of influenza virus and inhibits the nuclear import of NP/vRNP, thus demonstrating a new direct antiviral role for PLSCR1.
The localization of PLSCR1 is directly correlated with its function. Because it was initially identified as plasma membrane protein, its role in regulating the movement of plasma membrane phospholipids was intensively studied [22, 47, 48]. Moreover, cell surface-localized PLSCR1 can bind to the envelope proteins E1 and E2 of HCV and serve as an attachment factor for HCV entry [49]. As a result, downregulation of PLSCR1 expression inhibits HCV entry and infection. PLSCR1 can also be imported into the nucleus by the importin α/β import pathway [28], where it can bind to genomic DNA or nuclear proteins to perform different functions [29, 31, 50]. Yet, we found that PLSCR1 was localized predominantly in the cytoplasm of both PLSCR1-overexpressing and empty retrovirus-transduced control A549 cells infected with influenza viruses. The accumulation of PLSCR1 in the cytoplasm would likely enable it to efficiently participate in interactions with both importin α and viral NP, and effectively inhibit the nuclear import of vRNP/NP and virus replication.
Mutations of NP phosphorylation sites can reduce the binding affinity between NP and various members of the importin α family, resulting in the inhibition of nuclear import of NP and a reduction in virus replication [40]. In the present study, we found that the replication of influenza virus in PLSCR1-overexpressing A549 cells decreased the overall expression of viral NP protein. However, the extent of NP phosphorylation was similar between the PLSCR1-overexpressing cells and empty retrovirus-transduced control A549 cells. In addition, the replication of the NP phosphorylation mutant S9A/Y10F was further decreased in PLSCR1-overexpressing cells, thereby demonstrating an accumulative inhibitory effect on influenza virus propagation. Together, these results suggest that the effect of PLSCR1 on virus replication does not involve modulating the phosphorylation status of NP.
In summary, here we demonstrated that PLSCR1 is an interacting partner of the influenza NP protein. This interaction appears to downregulate virus replication since overexpression of PLSCR1 resulted in a significant reduction in virus titer in cell cultures of different virus subtypes, whereas siRNA knockdown or CRISPR/Cas9 knockout of PLSCR1 expression increased virus replication. Importantly, PLSCR1 inhibited the nuclear import of vRNP/NP, thus causing retardation of the virus life cycle. Moreover, we revealed that the mechanism by which PLSCR1 regulates influenza virus replication involves the formation of a complex with viral NP and importin α, which inhibits the incorporation of importin β into the complex and suppresses the nuclear import of NP. Collectively, our data suggest that PLSCR1 is an important host restriction factor against influenza virus.
HEK293T (ATCC CRL-3216), A549 (ATCC CCL-185), and MDCK (ATCC PTA-6500) cells were cultured in DMEM (Life Technologies, Grand Island, NY) containing 10% fetal bovine serum (FBS, Sigma-Aldrich, St. Louis, MO), in F12K (Life Technologies) with 10% FBS, or in MEM (Life Technologies) containing 5% newborn calf serum (NCS; Sigma-Aldrich), respectively. All media were supplemented with 100 units/ml penicillin and 100 μg/ml streptomycin (Life Technologies). All cells were cultured at 37°C with 5% CO2. A/Anhui/2/2005 (AH05, H5N1) and A/Anhui/1/2013 (AH13, H7N9) were grown in 10-day-old embryonated chicken eggs. A/WSN/33 (WSN, H1N1) and A/Fuzhou/1/2009 (FZ09, H1N1) were propagated in MDCK cells cultured in MEM containing 0.3% bovine serum albumin (BSA, Sigma-Aldrich) and 0.5 μg/ml L-1-tosylamide-2-phenylmethyl chloromethyl ketone (TPCK)-treated trypsin (Worthington, Lakewood, NJ).
All experiments with H5N1 and H7N9 viruses were conducted within the enhanced animal biosafety level 3 (ABSL3+) facility in the Harbin Veterinary Research Institute (HVRI) of the Chinese Academy of Agricultural Sciences (CAAS), which is approved for such use by the Ministry of Agriculture of China and the China National Accreditation Service for Conformity Assessment.
The yeast two-hybrid screen for protein-protein interactions was performed by using the matchmaker yeast two-hybrid system (Clontech, Mountain View, CA) as previously described [38]. NP of AH05 (H5N1) was constructed in pGBKT7, fused to the C-terminus of the GAL4-binding domain (BD), and used as bait. cDNAs prepared from a mixed human cell culture comprising A549, HEK293T, THP-1 (ATCC TIB-202), and U251 (Type Culture Collection of the Chinese Academy of Sciences, Shanghai, China) were cloned into pGADT7, fused to the GAL4-activation domain (AD), and used as prey. The yeast strain Y2HGold was transformed with the pGBKT7-AH05NP bait by using lithium acetate, and was then mated with the Y187 strain transformed with the pGADT7-based cDNA library. Transformants were selected on plates with synthetically defined medium lacking adenine, histidine, leucine, and tryptophan (SD/–Ade/–His/–Leu/–Trp) (quadruple dropout medium, QDO). The recovered colonies were grown on QDO plates containing 5-bromo-4-chloro-3-indolyl-α-d-galactopyranoside (X-α-Gal) and aureobasidin A (AbA) (SD/–Ade/–His/–Leu/–Trp/X-a-Gal/AbA, QDO/X/A). Blue colonies were selected and cultured in medium lacking leucine and tryptophan (SD/−Leu/−Trp) (double dropout medium, DDO). Plasmids were purified and sequenced to identify the potential cellular interactants with NP. To eliminate false-positive interactions, the bait and prey plasmids were cotransformed into the Y2HGold strain. Cotransformation of pGADT7-T (AD-T) with pGBKT7-p53 (BD-p53) into Y2HGold served as a positive control, and cotransformation of AD-T with pGBKT7-Lamin (BD-Lam) served as a negative control.
Human PLSCR1, importin β and the open reading frames (ORFs) of PB2, PB1, PA, and NP derived from WSN virus were cloned into the mammalian expression vector pCAGGS (a gift from Dr. Yoshihiro Kawaoka, University of Wisconsin-Madison). GST-tagged PLSCR1 and WSN NP were constructed in pCAGGS with a GST tag at the N-terminus. Plasmids pCAGGS-V5-WSNNP, pCAGGS-Flag-PLSCR1 and pCAGGS-Flag-MOV10 were generated by inserting the ORF of WSN NP, PLSCR1 and MOV10 fused with the V5 or Flag tag sequence at the N-terminus into the pCAGGS vector. Truncation mutants of GST-tagged WSN NP were generated by using a PCR approach and were cloned into the pCAGGS vector. pQCXIN-PLSCR1 was constructed by inserting the PLSCR1 ORF into the pQCXIN vector (Clontech). WSN NP S9A/Y10F, containing two mutations in the full-length NP gene, was generated by using a Fast Mutagenesis System (Transgen, Beijing, China) and was cloned into the pHH21 vector (a gift from Dr. Yoshihiro Kawaoka, University of Wisconsin-Madison). pHH21-SC09NS F-Luc, used to produce negative-sense RNA containing 176 bases of the 3’ end of the NS vRNA derived from A/Sichuan/1/2009 (SC09, H1N1), a firefly luciferase, a stop codon (TAA), and 179 bases of the 5’ end of SC09NS vRNA, was constructed by using the PCR method. The full-length ORFs of human importin α1, importin α3, importin α5, and importin α7 were cloned into pCAGGS with a Myc tag at the N-terminus. All plasmid constructs were confirmed by sequencing.
Mouse anti-NP monoclonal antibody (mAb) and rabbit anti-NP polyclonal antibody (pAb) were prepared in our laboratory by using conventional methods. The following primary antibodies were purchased from commercial resources: rabbit anti-V5 pAb (AB3792, Merck Millipore, Darmstadt, Germany); rabbit anti-GAPDH pAb (10494-1-AP), rabbit anti-importin β pAb (10077-1-AP), rabbit anti-LaminB1 pAb (12987-1-AP), rabbit anti-Mx1 pAb (13750-1-AP) and rabbit anti-PLSCR1 pAb (11582-1-AP) from Proteintech (Wuhan, China); mouse anti-actin mAb (sc-47778), mouse anti-p-Ser mAb (sc-81514) and mouse anti-p-Tyr mAb (sc-508) from Santa Cruz (Dallas, TX); mouse anti-Flag mAb (F3165), mouse anti-Myc mAb (M4439), mouse anti-V5 mAb (V8012), rabbit anti-Flag pAb (F7425) and rabbit anti-Myc pAb (C3965) from Sigma-Aldrich. The secondary antibodies used in the western blotting were DyLight 800 goat anti-mouse IgG (H+L) (072-07-18-06) and DyLight 800 goat anti-rabbit IgG (H+L) (072-07-15-06), purchased from KPL (Gaithersburg, MD); the secondary antibodies used in the confocal microscopy were Alexa Fluor 488 donkey anti-rabbit IgG (H+L) (A21206) and Alexa Fluor 633 goat anti-mouse IgG (H+L) (A21050) obtained from Life Technologies.
To examine the interaction of proteins in transfected cells, HEK293T cells were transfected with the indicated plasmids by using the Lipofectamine LTX and Plus Reagents (Invitrogen, Carlsbad, CA). To determine the interaction of proteins during natural viral infection, A549 cells were mock infected with PBS or infected with WSN virus at an MOI of 5. Cell lysates were prepared at 48 h post-transfection for transfected HEK293T cells or at 6 h p.i. for infected A549 cells. Briefly, the cells were washed twice with cold PBS and lysed with IP buffer (25 mM Tris-HCl pH 7.4, 150 mM NaCl, 1% NP-40, 1 mM EDTA, 5% glycerol; Pierce, Rockford, IL) containing complete protease inhibitor cocktail (Roche Diagnostics GmbH, Mannheim, Germany) for 30 min on ice and then centrifuged at 12,000 rpm at 4°C for 10 min. The supernatants were mixed with the respective primary antibodies, rocked overnight at 4°C, mixed with Protein G-Agarose beads (Roche) and rock for 6–8 h. The beads were washed four times with wash buffer (25 mM Tris-HCl pH 7.4, 150 mM NaCl, 1 mM PMSF). The bound proteins were then boiled in 2 × SDS sample buffer, separated by 12% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), and detected by western blotting.
HEK293T cells grown in 10-cm dishes were individually transfected with 10 μg of each plasmid (pCAGGS, pCAGGS-GST, pCAGGS-GST-PLSCR1, pCAGGS-NP, pCAGGS-GST-NP, or pCAGGS-PLSCR1) by using the Lipofectamine LTX and Plus Reagents. At 48 h post-transfection, cells were solubilized with 0.8 ml of IP buffer. Then, 300 μl of the cleared lysates from cells transfected with pCAGGS-GST, pCAGGS-GST-PLSCR1, or pCAGGS-GST-NP was mixed with 40 μl of Glutathione Sepharose 4 Fast Flow (GE Healthcare, Pittsburgh, PA) and rocked for 1 h at 4°C. After three washes with wash buffer, 300 μl of the cleared lysates from cells transfected with non-GST expressing constructs (i.e., pCAGGS, pCAGGS-NP, or pCAGGS-PLSCR1) was added and incubated for 2 h at 4°C. After three washes, the bound proteins were separated by SDS-PAGE. GST, GST-PLSCR1, or GST-WSNNP proteins in the eluates were detected by Coomassie blue (CB) staining, and non-GST tagged NP and PLSCR1 proteins were detected by western blotting.
Protein samples fractionated by SDS-PAGE were transferred onto nitrocellulose membranes (GE Healthcare). Membranes blocked with 5% skim milk in PBST were incubated overnight at 4°C with appropriately diluted primary antibody in PBST containing 2% BSA. After incubation with DyLight 800 goat anti-mouse IgG (H+L) and DyLight 800 goat anti-rabbit IgG (H+L), blots were visualized by using an Odyssey infrared imaging system (Li-Cor BioSciences, Lincoln, NE).
The AmphoPack-293 packaging cell line (631505, Clontech) cultured in 10-cm dishes was transfected with either retroviral construct pQCXIN-PLSCR1 or with the empty pQCXIN vector by using Lipofectamine LTX and Plus Reagents. At 48 h post-transfection, viral supernatants from the transfectants were collected and used to transduce A549 cells cultured in 6-well plates. Forty-eight hours later, the transduction was repeated to enrich for transductants. The confluent transduced cells were split and cultured in medium supplemented with 1000 μg/ml G418 for selection. The surviving cells were individually cloned in 96-well plates, propagated, and examined for PLSCR1 overexpression by quantitative reverse-transcription PCR (RT-qPCR) and western blotting. To study the effect of PLSCR1 overexpression on influenza virus replication, we used WSN (H1N1), AH05 (H5N1), AH13 (H7N9) or FZ09 (H1N1) to infect the PLSCR1-overexpressing cells or the empty retrovirus-transduced control A549 cells at an MOI of 0.1. Supernatants were collected at the indicated timepoints after infection and virus titers were determined by means of plaque assays on MDCK cells [38].
siRNA targeting PLSCR1 (5’-GCGGAAGAUACUGAUUGCU-3’) or scrambled siRNA (Genepharma, Shanghai, China) at a concentration of 30 nM was transfected into A549 cells seeded in 12-well plates by using the Lipofectamine RNAiMAX transfection reagent (Invitrogen). Forty-eight hours later, the knockdown efficiency was checked by means of RT-qPCR and western blotting. To study the effect of PLSCR1 knockdown on the growth of influenza virus, the WSN virus was used to infect siRNA-treated A549 cells at an MOI of 0.1. Supernatants were collected at 24 and 48 h post-infection (p.i.), and the virus titers were determined by means of plaque assays on MDCK cells.
Cell viability was determined by using the CellTiter-Glo kit (Promega, Madison, WI) as described previously [38]. Briefly, A549 cells seeded in opaque-walled 96-well plates were transfected with siRNA targeting PLSCR1 or with scrambled siRNA at a concentration of 30 nM. At 48 h post-transfection, 100 μl of CellTiter-Glo reagent was added directly into each well and incubated with the cells for 10 min on a shaker to induce cell lysis. The luminescence was measured with a GloMax 96 Microplate Luminometer (Promega).
PLSCR1-KO HEK293T cells were established using the CRISPR/Cas9 system. The PLSCR1 gene target sequence, 5’–CAGGATATAGTGGCTACCCT– 3’ (to target exon 4), was inserted into the guide RNA (gRNA) expression cassette of the pX330 vector [51], which also contains an expression cassette of Cas9. Six micrograms of the pX330 plasmid containing the PLSCR1 target sequence was then transfected into HEK293T cells with TransIT-LT1 (Mirus, Madison, WI). The transfected cells were trypsinized 24 h later into single cells, which were diluted and inoculated into 96-well plates for colony formation. Each colony was individually propagated into 24 well-plates, and the knockout of PLSCR1 expression was confirmed by western blotting. The PLSCR1-KO HEK293T or control cells were infected with WSN virus at an MOI of 0.1. Supernatants were collected at 24 and 48 h p.i., and virus titers were determined by means of plaque assays on MDCK cells.
A549 cells seeded in glass-bottom dishes were transfected with the indicated plasmids by using the Lipofectamine LTX and Plus Reagents. PLSCR1-overexpressing cells or empty retrovirus-transduced control A549 cells were infected with WSN virus at an MOI of 5. At 20 h post-transfection or 4, 6, 8, 10, and 12 h p.i., cells were fixed with 4% paraformaldehyde (PFA) in PBS for 1 h, and permeabilized with 0.5% Triton X-100 in PBS for 30 min. The permeabilized cells were blocked with 5% BSA in PBS for 1 h, and then incubated with primary antibodies (mouse anti-NP mAb, 1:500; rabbit anti-PLSCR1 pAb, 1:1000) for 2 h. The cells were washed three times with PBS and incubated with the secondary antibodies (Alexa Fluor 488 donkey anti-rabbit IgG (H+L), 1:10000; and Alexa Fluor 633 goat anti-mouse IgG (H+L), 1:10000) for 1 h. After four washes, the cells were incubated with DAPI (4’,6-diamidino-2-phenylindole, Thermo Fisher Scientific, Waltham, MA) for 15 min to stain the nuclei. Images were acquired by using the Leica SP2 confocal system (Leica Microsystems, Wetzlar, Germany).
HEK293T cells were treated with either siRNA specifically targeting PLSCR1 or with scrambled siRNA (30 nM) for 48 h, and were then cotransfected with the four protein expression plasmids of the RNP complex from WSN (pCAGGS-PB2, pCAGGS-PB1, pCAGGS-PA, and pCAGGS-NP; 1 μg of each), the construct pHH21-SC09NS F-Luc (0.1 μg), and an internal control pRL-TK (0.1 μg). At 48 h post-transfection, cell lysates were prepared by using the dual luciferase reporter assay system (Promega), and the luciferase activities were measured on a GloMax 96 microplate luminometer (Promega).
HEK293T cells grown in 24-well plates were transfected with the ISRE-Luc reporter plasmid (0.25 μg), pRL-TK control plasmid (0.02 μg), and the pCAGGS-PLSCR1 or empty pCAGGS plasmid (0.25 μg) for 20 h. The luciferase activity of the transfected cells was determined by using the dual-luciferase reporter assay.
PLSCR1-overexpressing cells or empty retrovirus-transduced control A549 cells grown in 12-well plates were left untreated or treated with 100 U/mL of IFN-α (Sigma-Aldrich) for 24 h. Cell lysates were then prepared and subjected to western blotting with a rabbit anti-Mx1 pAb to determine the expression level of Mx1 protein.
The mutant WSN virus WSN NP S9A/Y10F, which possesses two mutations in the viral NP protein, was generated by use of reverse genetics as described previously (34). Briefly, the eight plasmids for the synthesis of viral RNA (vRNA) and the four supporting plasmids to express the PB2, PB1, PA, and NP proteins were transfected into HEK293T cells with the Lipofectamine LTX and Plus Reagents. At 48 h post-transfection, the transfection supernatant was harvested and used to infect MDCK cells to produce stock viruses. To ensure that the mutant virus contained the desired mutation, vRNA was extracted from the stock viruses using a QIAmp viral RNA mini kit (QIAGEN, Valencia, CA), reverse transcribed into cDNA with Superscript III reverse transcriptase (Invitrogen), and amplified by PCR with gene-specific primers. The complete NP segment was sequenced by using an ABI 3500xL genetic analyzer (Applied Biosystems, Carlsbad, CA).
The PLSCR1-overexpressing or empty retrovirus-transduced control A549 cells grown in 12-well plates were infected with wild-type WSN virus or the NP mutant, WSN NP S9A/Y10F, at an MOI of 0.1. The cells were incubated with F-12K medium containing 0.3% BSA at 37°C. Virus-containing supernatant was harvested at the indicated timepoints and was subjected to plaque assays on MDCK cells to determine the virus titer.
To quantify the level of PLSCR1 mRNA, total RNA was extracted from PLSCR1-overexpressing A549 cells or siRNA-treated A549 cells at 48 h post-transfection by using an RNeasy kit (QIAGEN). The first-strand cDNA was generated with oligo(dT) primer using Superscript III reverse transcriptase. Real-time PCR was conducted using SYBR premix Ex Taq II (TaKaRa, Dalian, China) and 0.4 μM PLSCR1 primers according to the manufacturer’s instructions. Relative RNA quantities were determined by using the comparative cycle-threshold method, with cellular GAPDH serving as the endogenous reference and empty retrovirus-transduced A549 control cells or scrambled siRNA-treated cells serving as the control.
The PLSCR1-overexpressing or empty retrovirus-transduced control A549 cells grown in 6-well plates were infected with WSN virus at an MOI of 5. Total RNA was extracted by using an RNeasy kit at 6 h and 10 h p.i. Relative quantities of viral NP genomic RNA (vRNA), complementary RNA (cRNA) and mRNA were determined by qRT-PCR as described previously [52]. Relative RNA quantities were determined with GAPDH serving as the endogenous reference.
The PLSCR1-overexpressing or empty retrovirus-transduced control A549 cells grown in 6-well plates were infected with WSN virus at an MOI of 5. At 6 h p.i., the cells were separated into nuclear and cytoplasmic fractions by using NE-PER Nuclear and Cytoplasmic Extraction Reagents (Pierce) according to the manufacturer’s procedure. The amount of NP and PLSCR1 in each fraction was determined by western blotting with a rabbit anti-NP pAb and a rabbit anti-PLSCR1 pAb, respectively. LaminB1 and GAPDH, nuclear and cytoplasmic fraction markers, respectively, were detected by western blotting with a rabbit anti-GAPDH pAb and a rabbit anti-LaminB1 pAb, respectively.
In another experiment, PLSCR1-overexpressing or control A549 cells grown in 6-well plates were pretreated with 50 μg/mL CHX (Sigma-Aldrich) for 1 h, and then infected with WSN virus at an MOI of 5. The virus-infected cells were maintained in culture medium containing CHX for 2 h, and were then subjected to cell fractionation and western blotting as described above.
Unless otherwise indicated, all experiments were performed at least three times; data from representative experiments are shown. Data were statistically analyzed by using the Student’s t test. A mean difference was considered statistically significant if the P value was < 0.05.
|
10.1371/journal.pcbi.1002210 | Estimating the Relevance of World Disturbances to Explain Savings, Interference and Long-Term Motor Adaptation Effects | Recent studies suggest that motor adaptation is the result of multiple, perhaps linear processes each with distinct time scales. While these models are consistent with some motor phenomena, they can neither explain the relatively fast re-adaptation after a long washout period, nor savings on a subsequent day. Here we examined if these effects can be explained if we assume that the CNS stores and retrieves movement parameters based on their possible relevance. We formalize this idea with a model that infers not only the sources of potential motor errors, but also their relevance to the current motor circumstances. In our model adaptation is the process of re-estimating parameters that represent the body and the world. The likelihood of a world parameter being relevant is then based on the mismatch between an observed movement and that predicted when not compensating for the estimated world disturbance. As such, adapting to large motor errors in a laboratory setting should alert subjects that disturbances are being imposed on them, even after motor performance has returned to baseline. Estimates of this external disturbance should be relevant both now and in future laboratory settings. Estimated properties of our bodies on the other hand should always be relevant. Our model demonstrates savings, interference, spontaneous rebound and differences between adaptation to sudden and gradual disturbances. We suggest that many issues concerning savings and interference can be understood when adaptation is conditioned on the relevance of parameters.
| Trying to explain how humans adapt to new motor behaviors and retain them over time is a central focus in motor control. Many aspects of adaptation, including savings and interference, have proven difficult to explain in a coherent manner. Linear dynamical models have been successful at describing the observed increase in performance while subjects familiarize themselves with an experimental perturbation. Many aspects of these experiments however, remain unexplained. In particular, while subjects display the ability to remember new motor behaviors for long periods of time, these linear models cannot. In this work we extend our previous body-world model of motor adaptation by estimating the relevance of inferred world disturbances. When these parameters are estimated to be relevant, they are used (and motor behaviors are adapted), and when they are estimated to not be relevant they are stored (and motor behaviors are remembered without being lost). Our model offers explanations for many observations on motor adaptation, savings and interference.
| There is a large body of evidence to suggest that the nervous system maintains internal representations of variables that are relevant to the production of movement [1], [2], [3]. Internal models allow us to make repeatable and reliable movements despite a highly variable world and body, and our noisy perceptions of them. Ideally, these internal models ought to distinguish between the properties of the body and world, a crucial ability when generalizing movements [4]. Such a representation requires many parameters to represent how to control the body when interacting with external objects in the world. This in turn implies that many parameters of both the body and the world need to be estimated.
When estimating changes in the many parameters necessary to describe the interaction of the body and the world, it seems sensible that some of these parameters will change rapidly, while others change more slowly. Consequently a number of recent studies have constructed linear time invariant models that model adaptation unfolding over multiple time scales [e.g. 5,6,7]. These models have explained a wide range of temporal adaptation and savings phenomena.
While many linear models can explain motor phenomena associated with rapid re-adaptation, they are limited in their ability to explain phenomena of even short-term adaptation, as in savings after “washout” trials [e.g. 8], let alone the long-term effects of adaptation. For instance, linear models predict that aftereffects should decay with the same rate behaviors are adapted to, in contrast with experimental evidence [9], [10]. Linear models also predict that once a disturbance has been removed, its influence on movement is de-adapted and completely forgotten. This is clearly not the case, and subjects retain the ability to compensate for previously adapted behaviors over long periods of time [11], [12], [13]. In summary, while there are clearly multiple time scales at work, a linear time invariant process is not capable of explaining motor adaptation.
Since neither washout nor intervening days delete motor adaptations, there must be some mechanism that guards newly adapted parameter values against de-adaptation when they are no longer relevant. Motor architectures that can guard entire forward and inverse models of limb dynamics by switching them on and off have been proposed [14], [15], [16]. However, it is unclear how these models can account for the patterns of apparently incomplete generalization observed experimentally [4]. What's more, these models do not make a distinction between the parameters of the body and the world, but rather estimate when an appropriate model of the coupled body and world dynamics is applicable. In contrast, we propose the nervous system should separately estimate the properties of the body and the world and when those individual parameters are relevant for control. For example, our estimate of a coffee cup's weight is only relevant while we are holding it and not after we have set it down on the table. Estimates of our arm's weight, on the other hand, are always relevant for limb movements. Conditioning on such obvious relevance the nervous system can know not to adapt estimates of the cup's weight unless we are holding it. Parameter relevance, however, is not always this obvious. If this relevance could be estimated, then the nervous system could guard newly adapted behaviors and later retrieve them when they are relevant again.
To examine this idea, we designed an idealized model for computing the probability of relevance, and then using this estimated relevance to adapt. In a previous study we proposed a statistical inference model for motor adaptation that estimated a large number of parameters for the body and the world [4]. In a different study we proposed that the nervous system constantly estimates the relevance of errors for motor adaptation [17]. Here we combine these two approaches. We assume that parameters associated with the body are always relevant, whereas world parameters are only relevant under specific conditions. If the probability of a parameter's relevance is high, then it is subject to adaptation. If not, the motor errors may be due to sensorimotor noise or changes in body parameters.
In contrast with the coffee cup example, the kind of experimental disturbances subjects are exposed to are not as evident. Therefore, we estimate relevance using a model that can predict the consequences of a class of world disturbances. As such, relevance defined here does not depend on a particular parameter value, but rather the particulars of that type of parameter's influence on motor behaviors. When movement patterns are consistent with a large world disturbance, regardless of the observed movement error (see Fig 1), then the likelihood of that parameter being relevant is high. For example, if the presence of a coffee cup in our hand, any coffee cup, can account for unexpected limb motions and forces on our hand, then parameters representing the cup's inertial properties should be subject to adaptation. If not, then those parameters should not be updated. In effect this allows for a rudimentary long-term memory, allowing for the retention and later retrieval of newly acquired world parameter values.
We simulated a series of experiments to investigate how our model behaves when adapting to multiple motor behaviors in succession. The model was restricted to four free parameters, which were held constant for all simulations. The models' predictions are consistent with the findings of savings, interference, spontaneous rebound and the differences between adaptation to gradual and abrupt disturbances. Our model offers a formalization of how the nervous system may estimate and store motor parameters when adapting to disturbances.
The model used here is based on that used in a previous study [see 4 for details and code]. Briefly, the human upper limb is modeled as a nonlinear 2-link, 2 degree of freedom mechanism driven by feedforward torque components to compensate for estimated world and body dynamics, plus a feedback component to stabilize movements about a nominal, minimum jerk trajectory. For the results shown here only two parameters were inferred, a body-centric visuomotor rotation θb (due to some possible combination of proprioceptive errors and relative head or torso rotations) and a world-imposed visuomotor rotation, θw, the experimental disturbance of the cursor. The system observation, y(t), is the visually observed (displayed cursor) position vector, x and velocity vector, dx/dt of the limb's endpoint (or hand) in a Cartesian reference frame, y = [x(t), dx(t)/dt]T. We assume this observation is corrupted by measurement noise, n(t), with zero mean and covariance R.
We collate the parameters to be estimated in the vector, p = [θb, θw]T. To infer these parameters, we assume that they vary according a random walk, with a small forgetting factor,
where wi is a zero mean random variable drawn from a normal distribution with variance, σi2. These parameters influence the nonlinear dynamics of the limb, and the subsequent effects on movement are then observed in the output, y. However, we assume that influences of the world parameter, θw, are only observed when the limb is perturbed. To denote this state of being perturbed by a visuomotor rotation, we define the relevance variable, λrot. The variable can take on one of two values, one or zero. If a world rotation parameter is relevant then the relevance variable is one, if not, zero. The system's output then depends on the relevance in the following way,
where y(θ, t) is shorthand for the observed output when visually rotated by θ. Though binary, we assume that the relevance parameter is also Markovian, and has a small non-zero probability of transitioning from one value to the other. We define a transition model (or mixing matrix), M = [0.999 0.001; 0.001 0.999], ensuring our prior probability of relevance never becomes fixed at 0 or 1. In total the model has 4 free parameters, ab and aw, σb and σw. However, to assure that estimates of the world would be retained over long periods of time, we held aw fixed at 0.99999.
The model uses its observations of the limb's endpoint (y) to infer the probability that an external parameter is relevant and update its belief in the parameters in p. In this study we were focused on the interference and savings of visuomotor adaptations, therefore we limited our computations of relevance to the visuomotor variable, λrot. Key in this computation is how the likelihood of relevance is computed. Before examining this, we first briefly describe how the posterior probability of relevance is computed using a Bayesian update. For ease of notation, we shall refer to the visuomotor relevance variable, λrot, as λ for the remainder of this section. As defined above, λ = 1 if it's variable is relevant, and λ = 0 if not. The posterior probability that the world's rotation parameter is relevant, P(λ(t) = 1|y(t)) is found through Bayes' rule,
where P(y(t)|λ(t) = 1) is the likelihood (P(y|λ = 1) for brevity), and P(λ(t) = 1) is the prior (P(λ = 1) for brevity). Note that the prior is the posterior found in the previous time step, modulo the transition model, m11 P(λ = 1) + m12 P(λ = 0), since it summarizes the probability of being relevant based on all the observations made up to that time (we assume λ is Markovian).
Our definition of relevance is based on a type of parameter's ability to explain disturbances. To illustrate, consider reaches early during adaptation. The body and world parameter estimates of a visuomotor rotation are zero and there are large movement errors. These errors are consistent with a large world disturbance and the probability of the visuomotor parameter being relevant is computed as high (Fig 1A). After adapting for some time, an updated estimate of the body parameter partially compensates for the disturbance. The newly estimated world disturbance further compensates for the disturbance. Any remaining errors are used to update these parameter values with a Bayesian update (see below). However, even if the errors are driven to zero, there remains a large apparent error between the observed movement, and how much the body parameter can account for (Fig 1B). If this mismatch can be explained as the result of a relatively large world rotation, then the likelihood of a world disturbance is high. Therefore the corresponding likelihood of relevance is based on the probability of observing the cursor (y), given our current estimate of the body parameter, and a world rotation of any value perturbing our observations. To compute this we must integrate the probability of a perturbed observation over all possible world rotations,
P(y|θ,λ = 1) is the likelihood of observing the limb's endpoint with a given rotation, θ. Since body parameters are always relevant, this likelihood is a normal distribution centered on the internal model's prediction, N(y(θ+θb), R). Rather than integrate this distribution over the forward model's prediction over each movement and all possible world rotations we made the following simplifying assumption. Since the visuomotor disturbance influences movement observations in a relatively simple and unique manner (a constant rotation), we redefined this likelihood using only visuomotor angles. We used the hand trajectory, to identify the unique rotation, θy, that minimized the root mean squared error between the observed limb path, y and the estimated path when only compensating with the, always relevant, body estimate. We then use a Normal distribution over θ centered on θy, with the variance associated with an observation of the rotated limb, σθ2 (see below). Although the normal distribution is defined over all real numbers, the variance of this distribution is much smaller than our limits of integration, and can very accurately be described as restricted between − π to π.
To define the prior over visuomotor angles, P(θ/λ = 1), we note the following considerations: relevance is based on the ability of any visuomotor disturbance to explain the data, and we want to avoid biasing the inference. The prior should be flat over all non-zero rotations, but avoid assigning high probabilities to the degenerate case of small (relative to our observation noise) or zero rotations. Based on these considerations we defined the prior as (1-exp(−θ2/2σθ2))/Z where Z is an appropriate normalizing constant. Just as above, given that the variance for the Gaussian term is much smaller than the domain, 2π, Z is very accurately approximated as 2π. This form of a prior assigns high probability to all large valued rotations, and low probability to rotations that are near zero, or small relative to the size of the observation noise, σθ2.
After integrating the above equations we find an expression solely in terms of the rotation that corresponds to our observation, θy,
To summarize, this likelihood assigns high probability when the observed rotation, θy is large relative to the observation noise. We also note that in this idealized model, θy is the angular displacement relative to a movement predicted using the current estimate of a body disturbance, θb. Thus estimated body disturbances influence the forward model's belief of where in space the limb is. Finally, we also need to compute the likelihood of the unperturbed condition, λ = 0,
We can define P(y|θ,λ = 0) with a Normal distribution just as before. However, since this is for the case when the rotation is not relevant, this distribution should be centered on θ = 0. The prior, however, will be different. The prior should only assign large probability to rotations that are small, or small relative to the size of the observation noise. Therefore we define the prior as a Normal distribution with zero mean and variance, σθ2. Again, since the variance for these distributions is very small relative to the limits of integration, both the likelihood and prior can be accurately approximated as restricted between −π to π. After integrating, we arrive at
With this final term found, we can express, P(y) = P(y|λ = 1) P(λ = 1) + P(y|λ = 0)P(λ = 0), and compute both the posteriors, P(λ = 1|y), and P(λ = 0|y).
The variance, σθ2, was found by noting that the angle subtended by the arm's length, L, and one standard deviation of the observation noise in either direction, is approximately 2σ/L, where σ is 0.01 meters. Using either the upper or lower arm length for L, the angle is approximately 1.7°. Using the whole arm length for L, the angle is 3.4°. Therefore, we defined σθ2 = (2.5 degrees)2. During the error clamp simulations the model's observation was artificially constrained to have zero error, regardless of the parameter estimates used to generate motor commands, or their relevance. The model's observations of movements that attempted to compensate for disturbances were no different from estimated movements without disturbances. To model this uncertainty, we held the likelihood fixed at 0.5 during these circumstances. We note that denying the model the evidence necessary to compute a likelihood (as may occur in error clamps) also has the same effect, as the transition matrix relaxes the probability of relevance to 0.5 as time passes.
With the relevance probabilities in hand, we can then infer estimates of the parameters. The estimate of the world's rotation used by the model to make predictions and compute commands is conditioned on the prior probability of being relevant,
since (the rotation when not operating in a visuomotor rotation) is assumed to be zero. This expected world estimate along with the body estimate is collated in the vector .
If the probability of relevance is one, then the update for the rotations is the extended Kalman filter update,
where A is a matrix with ab and aw on the diagonal. However, if the probability of relevance is zero, then the world rotation is guarded against adaptation, and the update is
Therefore, we approximate the update with the maximum likelihood update,
The parameters' covariance, P, was updated in a similar fashion. Defining Pn+1 = APnAT + Q, and the updated covariance , then the posterior covariance was approximated as
We note that multiple approximations to the updates for the parameters and their covariance were attempted and the qualitative results did not change. Furthermore, the transitions from low to high relevance are relatively quick (2–3 trials). As such, the approximations for inference during the intermediate state of relevance/non-relevance (0<P(λ = 1)<1) have only a limited influence on the estimated parameter values.
The limb parameter values were based on [18]. For all simulated experiments, the targets and reaching distances were equivalent to that used in the studies. For all simulated movements we assumed the nominal limb trajectory was that of a minimum jerk profile specified by the target locations, via points (8 equally spaced locations) and movement times reported. Parameter estimates were updated 6 times per movement, and movement targets were randomly selected. The probability of relevance was computed once per movement. The three free parameters, ab, σb and σw, were tuned by hand to create qualitative fits to the data from [19]. These values were then used for the remaining simulations.
Our simulated visuomotor experiments display trial-by-trial adaptations, whereas experimental plots of the same data are of cycles (data averaged over 8 consecutive trials). We have not made a distinction between trials and cycles because of the rescaling properties of the inference process. A single trial in our formulation need not represent a single trial or a cycle. The model is time invariant in this regard and we can scale all the parameters (jointly) to scale time by any specific value.
In our previous model [4] parameters were always relevant and subject to adaptation. For variables that describe the body this makes intuitive sense. Variables that describe the environment, however, may only be relevant in a particular circumstance [17]. We thus amended the source estimation model, partnering world parameters with relevance variables. The probability of being relevant is found by comparing the observed movement with the movement predicted if the estimated world disturbance were neglected. The estimate of a world parameter is then adapted using a Kalman update weighted by the probability of being relevant (see Methods). This contextualization allows for the storage and later retrieval of newly acquired parameter values.
In this study we focus on the paradigm of visuomotor adaptation, restricting the model to estimate two variables, a body-centric visual rotation (e.g. a rotation of the head relative to the torso and/or arm) and world-imposed rotation (the experimental manipulation). As a result, the model can only entertain one visual disturbance due to the body and one due to the world. We restrict the model to four free parameters: two parameters to describe the magnitude of noise associated with them, and two decay rates or time scales. However, we further assume the decay rate for world parameters is essentially zero, allowing for the long-term retention of that estimate. The existence of a fast and slow time scale are consistent with previous findings [5], and our previous work [4] which suggests the uncertainty associated with body parameters is large, and estimates should vary quickly. The resulting model offers predictions for how adaptation should proceed when it is statistically optimal.
Though the relevance model we present here is nonlinear in both the limb dynamics and the adaptation scheme, the results we present share many similarities with those of previously published linear models of adaptation. Specifically, when adapting to a visuomotor rotation of the model's hand location the motor errors appear linear in the estimated disturbances. Furthermore, although these disturbances are not adapted with a fixed rate (but instead estimated with an extended Kalman filter), trial-by-trial changes in the estimates are small and the resulting motor errors follow typical exponential trajectories. Due to these similarities the relevance model has the appearance of a linear estimation process with a nonlinearity that switches the estimated world disturbance in and out of the adaptation process.
To examine short-term motor adaptation, many experiments expose subjects to a disturbance twice in quick succession, with either a counter disturbance or a washout period in between. Savings are observed on the second presentation of the disturbance in both cases. Linear models can explain savings after adaptation in the form of an increased learning rate when adapting to the counter disturbance paradigm [5], [6]. However, linear (time invariant) models are not capable of explaining this same type of savings after a sustained washout period [8]. Once the perturbation has been removed, the model necessarily de-adapts its parameters. Therefore, a washout period lasting as long as the adaptation period would reverse any savings; a second exposure to the disturbance would proceed just as the initial one. Without a mechanism for guarding parameters against de-adaptation, linear models are incapable of displaying even this form of short-term motor adaptation.
Consider how the model presented here adapts while making reaches with a visuomotor perturbation. Initially the model cannot predict the consequences of, nor compensate for, a visual disturbance, and there are large motor errors (see Fig 1A, 2A). These errors drive adaptation of the estimated body rotation. At the same time, the model estimates that a large angular rotation of the hand's path is consistent with the observed reach (Fig 1A). This large potential angular perturbation indicates that the probability of the world's visuomotor rotation relevance is high (approximately 1, Fig 2B). As a result the world's rotation estimate is adapted and rises to help compensate for the experimental perturbation (Fig 2B). Although the motor errors progressively decrease, the model is still aware that a large visuomotor rotation is consistent with the ongoing observations; there remains a large discrepancy between the observed reaches and the model's estimate of an uncompensated reach. An estimate of the uncompensated reach is found by predicting a reach made without compensating for the estimated world rotation. The estimated body rotation however, is still used, and biases this estimate (see Fig 1). A large angular perturbation continues to be estimated and the probability of relevance remains high throughout the adaptation process. After an adequate number of trials, the contribution from the body and world rotations largely cancels the visual disturbance and the errors are small (Fig 2A). The overall motor behavior is qualitatively consistent with adaptation to a novel visuomotor disturbance. Both linear models and our nonlinear model can correctly describe the resulting patterns of adaptation.
Continuing with the short-term adaptation paradigm, when washout trials are subsequently presented, consistent with experimental findings, the relevance model produces large motor errors in the opposite direction (Fig 2A). The model, now biased by its previously adapted body rotation, mistakenly estimates an angular perturbation now in the opposite direction. The probability that the world's visuomotor rotation estimate is relevant remains high and both the body and world estimates de-adapt (this produces a short lasting overshoot in the error, Fig 2 green panel). As the body estimate quickly de-adapts the probability of relevance decreases back to zero. This change in the world's estimated relevance halts adaptation of the world rotation parameter. In contrast with similar linear multi-rate models, the motor errors are now only used to estimate the body's rotation parameter (which is always relevant). The body's estimate continues to de-adapt and the motor errors vanish. This combination of the fast change in the world parameter's relevance, along with the fast adaptation rate for the body parameter, results in the relatively quick de-adaptation back to nominal reaches. When the disturbance is turned on again, large errors result. Just as before, the probability that the world's visuomotor rotation parameter is relevant increases. This quick change in the estimated relevance results in a relatively fast decrease in errors, as the world's rotation estimate begins to compensate for the rotated reaches (Fig 2C). Thus the estimation of relevance allows the model to explain fast re-adaptation.
The adaptation and short-term savings we have reviewed above have similar analogues over longer time frames. To examine savings over multiple days, subjects adapt to a disturbance, and then are presented with the same motor disturbance on a subsequent day. In another paradigm, subjects adapt to two disturbances in quick succession, the later often a counter disturbance, and then evidence for savings or interference is examined on a subsequent day [e.g. 13,19]. Both experimental paradigms demonstrate that many of the features of short-term motor adaptation also exist over longer time frames. Unfortunately linear models are not capable of describing some of these phenomena over these longer time frames.
Using the relevance model to examine its predictions for saving over days, we simulated these experimental paradigms [e.g. 19]. No changes were made to the model for these long-term adaptation results. The first day's adaptation to a visuomotor disturbance proceeds just as described above (Fig 3A). There is a subsequent washout period just as before, where the probability that a world rotation estimate is relevant quickly decreases towards zero. The world's parameter estimate, no longer relevant, is left for later use should it become relevant again. The body estimate then rapidly de-adapts (Fig 3A).
In this particular experiment, subjects do not undergo a period of washout with the robot, but instead leave the experimental setting. Therefore this washout is different from those of the previously described short-term experiments on two counts. First, the washout trials predicted by the model correspond to natural movements made by the subjects after the experiment has ended. Our model thus predicts that there should be aftereffects that persist after the subject has let go of the robot handle. This is consistent with recent evidence [e.g. 20]. Second, since this washout period does not take place while grasping the robot handle, the last interactions subjects have with the robot are associated with a disturbance; the robot is an unambiguous proxy for the relevance of a visuomotor rotation. Therefore we assume that the model's initial probability of the visuomotor parameter's relevance should be similar when the model next returns to the experiment.
When adaptation on a subsequent day is simulated, the probability of the visuomotor parameter being relevant is initialized to a high value (0.75), as discussed above, and the world's rotation estimate is believed to be relevant. The model is again presented with the same visuomotor rotation and the initial motor errors are lower than those of the previous day (Fig 3B). At the end of this second day of adapting to the visuomotor disturbance the movement errors are lower than on the previous day. This is due to the relatively large contribution from the world estimate. The model's predictions for long-term savings are consistent with the observed experimental findings (Fig 3C).
We simulated the second experimental paradigm, now presenting two visuomotor rotations of opposite orientations in succession [19]. Adaptation to the first motor behavior proceeds just as above. When the model is presented with a second, oppositely directed rotation, the model again estimates a large angular discrepancy between the observed hand path and an estimated path that neglects the current world rotation estimate. However, this estimated rotation of the hand's path is now in the opposite direction. Regardless, the probability that the visuomotor rotation parameter is relevant remains high. Both the world and body parameters begin adapting to a rotation with the opposite sign (Fig 3D). When the washout trials begin, the probability of relevance quickly decreases, the world's rotation estimate is no longer used, and adaptation is halted. However, by this point all adaptation to the first visuomotor rotation has largely been lost. Just as above, on a subsequent day the model begins with a belief in a visuomotor rotation's relevance, and uses its estimated world rotation. Yet, the small estimate has little influence on the movements. Consistent with experimental findings of interference, the model performs as if naïve on the second day's presentation of the disturbance (Fig 3E, F).
Our new model explains long-term savings in the form of retention of a previously adapted motor behavior and decreased initial errors. Further, the model demonstrates how adaptation to two similar disturbances can cancel each other's influences and result in interference. Both findings are widely observed in motor adaptation studies.
Most studies examine adaptation after the sudden introduction of a perturbation. However, recent evidence has found marked differences when subjects adapt to a perturbation that is gradually introduced. These gradually introduced perturbations have been used to examine both interlimb generalization, and savings of motor behaviors across multiple days. In one study, subjects adapted to a force field that was either suddenly or gradually introduced [21]. After adapting, savings were examined when making test reaches with the non-dominant limb in the same force field (at full strength). The test reaches made after adaptation to the gradually introduced perturbation exhibited relatively larger deviations from a straight path. The initial errors were roughly twice as large as those found after adapting to the suddenly introduced perturbation, suggesting generalization of the adapted force field to the other limb was relatively poor when the perturbation is gradually introduced. In another study examining the differences between gradually and abruptly introduced force fields, post-adaptation reaches made without grasping the robot handle were examined [22]. The aftereffects on these free reaches were larger when subjects adapted to a gradually introduced perturbation. This suggested adaptation to a gradually introduced force field, may have altered the way subjects controlled their limb. Another study examined savings across days with a visuomotor rotation that was either gradually or suddenly introduced [23]. After adapting on one day, subjects made reaches in the same visuomotor perturbation (full strength) on a subsequent day. Subjects that had adapted to the gradually introduced perturbation made slightly larger errors initially, even though they adapted over more trials than the other group. These three results, and other studies like them, with their distinctions in savings, may offer testable predictions for how the nervous system adapts.
To examine our model's predictions we simulated the same gradual perturbation as the one used in [23]. During the early trials the motor errors are small and the body estimate quickly adapts to them. Because these errors are small the body estimate does an adequate job of compensating for the perturbation. The model does not detect a large angular perturbation and does therefore not believe the world's rotation estimate (initially zero) to be relevant. Only during later trials as the perturbation strength increases does the model believe the world's parameter is relevant. Thus, much of the adaptation is accounted for by the body estimate (Fig 4A). After the simulated experiment has ended, the model has a world estimate that is little more than half as strong as would be otherwise (compare with Fig 3). Our model predicts three findings of interest. First, we can conclude that during a generalization trial with the other limb, the model's errors would be approximately twice as large as if the perturbation was suddenly introduced, consistent with experimental evidence [21]. Second, because the perturbation is largely attributed to the body, the model predicts relatively large aftereffects during reaches made without the force field, when the robot handle is not grasped and the probability of a disturbance parameter's relevance is zero [22]. Third, since the world estimate of a rotation is smaller than would be otherwise, movement errors on a subsequent day are larger initially, just as was found experimentally (compare Fig 4A, B). Our model thus provides an interpretation of the effects that are associated with fast versus slow introductions of perturbations.
One additional set of phenomena may be important to characterize the properties of motor adaptation. In several recent studies subject's motor behaviors are examined when they make reaches in an “error clamp”, or “force channel”, wherein force disturbances are removed and movements are constrained to be straight. This is done to examine how and if subjects alter their motor strategies in the absence of kinematic errors. In an early study, after subjects adapted to a velocity-dependent force field, an error clamp was unexpectedly turned on [24]. Even though there was no longer any need to compensate for the force field, subjects continued to produce considerable forces as if it were still present. These forces slowly decayed, over a longer period of time than the subjects required to adapt or de-adapt in the absence of an error clamp. This suggested that these erroneous forces and their slow decay were the result of some altogether different process.
We can examine what the model would predict by simulating similar circumstances. The model is first presented with a visuomotor rotation, and then the reaches are “clamped” to constrain movement errors to be zero. Adaptation proceeds just as we have seen before (Fig 5A). Under the simulated error clamp condition, regardless of what the model (or subjects) does to compensate for a perceived disturbance, they observe the same error-less outcome. The model cannot observe the consequences of using its estimated perturbations; this results in uncertainty in the relevance of the visuomotor parameter (see Methods). As a result the model partially uses the world's estimate to compensate and both the body and world rotation estimates slowly decay towards zero. The results are qualitatively similar to experimental findings (Fig 5B).
In a somewhat different paradigm, after subjects adapt to one disturbance they are briefly presented with a counter disturbance and subsequently make reaches while errors were clamped [5]. Under these circumstances subjects temporarily make reaches as if they are compensating for the counter disturbance, even though it is not present. This phenomenon, termed spontaneous rebound, has been observed under a variety of conditions [25], [26], [27]. Ideally models of motor adaptation should be able to describe such a behavior.
How would the source relevance model explain such findings of spontaneous rebound? We can simulate the model's predictions to the same paradigm with a visuomotor rotation first, then a counter rotation, and then a “clamp” where we artificially constrain the movement errors to be zero. The model can predict spontaneous rebound through the interaction of two mechanisms. As with other linear multi-rate models there is the interaction of two or more processes with different adaptation rates [e.g. 5]. But more importantly for our model, under the simulated error clamp condition, the model (and subjects) observes an error-less outcome, regardless. This results in uncertainty in the relevance of the visuomotor parameter and the model partially uses the world estimate. The model appears to overcompensate for a nonexistent rotation and the results are similar to experimental observations (Fig 5C, D). Though other multi-rate models can explain spontaneous rebound, our model offers a different explanation in terms subject's difficulty in gauging the circumstances under which they are adapting.
Here we have extended a body-world, multi-rate model to infer not only the parameter values but also their relevance to the current motor conditions. The discrepancy between observed movements and those predicted when neglecting world estimates is used for the computation of relevance. World parameters that are estimated as having little relevance are not used to generate motor commands and are not adapted. Body parameters, however, are assumed to always be relevant and subject to adaptation. In effect, this allows for a rudimentary long-term memory of world parameters, allowing for the retention and later retrieval of newly acquired parameter values. The entire process is dynamic and requires no intervention for describing behavior across short or long time frames. We have demonstrated that such a model can explain a wide range of findings on human motor control. Our results are consistent with the basic findings of savings and interference, error clamp results, and the differences between adapting to gradual and abruptly introduced disturbances.
Though there are some clear similarities between the model we present here and other computational descriptions of motor control and adaptation, there are important distinctions. Our model makes a categorical distinction between parameters that represent the body and those that represent the world; thus it shares similarities with two-rate models [5], [6]. Indeed, our model makes nearly identical predictions for short-term savings, interference and reduced learning rates with increased adaptation duration [28]. Since these models are linear, however, they cannot explain adaptation on longer time scales, as all their adapted parameters relax back to zero. Perhaps a more fundamental distinction, it is not clear what the “fast” and “slow” variables in multi-rate models represent computationally, although they may be related to distinct neural structures at the implementation level. The model we present offers explanations for a range of findings on both short and long-term motor adaptations as well as generalization [4]. Further, we model the estimation of body and world variables that can be tested through future experimentation.
Since our model switches the world parameter values in and out based on their probability of relevance, it bears some resemblance to the other models that switch modules on and off, such as the mixture of experts and MOSAIC [14], [15], [16]. However, our representation of world and body parameters within a dynamical model is distinct from the MOSAIC controller's modules of paired forward and inverse models of whole body-world dynamics. The MOSAIC controller does not independently represent the body and the world (which is a cornerstone of our model). In fact, even if the MOSAIC were altered to represent the body and the world in two different modules, they could not be “summed” to represent whole body-world dynamics, as these descriptions are coupled and highly nonlinear. Our proposed model represents distinct parameters within a model of the limb and body dynamics. Therefore it can uniquely adapt these parameters, and use them for generalization in a manner MOSAIC cannot.
Furthermore, our use of a relevance parameter is distinct from the notion of context used in these switching controllers. In the MOSAIC model, modules are switched on and off based on the similarity between their predictions and the observed motor behavior. Each module's predictions are uniquely described by the current parameter values that make up that module (e.g. its current estimate for a visuomotor rotation or force field). As a result, a module for a particular force field will not be switched on unless the limb makes reaches in a very similar force field. Our computation of relevance is based not on a parameter's value, but on the manner it influences motor behaviors. For example, the parameter for an inertial perturbation is likely whenever limb movements are consistent with an inertial perturbation of any sufficiently large value.
In large part due to these differences in relevance and context, it is not clear if the MOSAIC model could also explain some of the findings we have presented here. For instance, consider adapting to a 30° visuomotor rotation. A module representing the perturbed limb dynamics would modify its parameters to compensate for the disturbance. When a −30° rotation is then presented, this module's prediction errors (now ∼60°) would in fact be larger than a baseline, null condition module (only ∼30°). As a result the context variable for the module associated with the visuomotor rotation would be switched off, and this module would not continue adapting to the counter rotation; the model would not predict interference. Through a similar line of reasoning it is not clear how the MOSAIC model could explain the phenomena of spontaneous rebound.
Other studies have used the idea of context in different ways. In one study context was defined as the implicit memory of the limb segments used during a motor behavior [29]. In a sense, this assigned relevance to different body effectors. In a more recent study context indicated visuomotor rotations of different magnitudes [6]. In those studies context was known unambiguously, not estimated based on errors or changes in the environment, as we have done here. Further, here we define relevance (similar to context) in terms of the existence of external disturbances, regardless of what limb segment is used or the strength of the particular magnitude of the disturbance. Our study can thus be seen as a generalization of these studies to unobserved contexts and changes in the environment, which makes new experimentally testable predictions about the role of relevance.
In this work we have examined the effects of adapting to a visuomotor rotation, however, this model could be extended to adapt to other types of disturbances as well. In particular, several experimental studies have investigated how adapting to visuomotor rotations and altogether different motor disturbances in quick succession, effect interference and savings [19], [30], [31]. Interestingly, the results of these studies, having contrasting findings on savings, have motivated distinct interpretations concerning the nervous system's ability to represent kinematics and dynamics uniquely. Within that context, these results were argued to be incompatible. Our model makes no distinction between kinematics and dynamics but instead a distinction between parameters that represent how to control the body and how to interact with the world. Furthermore, our model predicts that if the effects of two different perturbations were similar (in terms of their resulting motor errors and sensory consequences) then their accompanying world estimates (e.g. estimated world rotation, or estimate world force field) would both be assumed relevant for adaptation. Therefore, adapting to a visuomotor rotation, and then a force field that perturbed the limb in a similar manner, might produce interference [19], whereas the subsequent adaptation to a force field dissimilar to a visuomotor rotation wouldn't [30]. As such future work using this model may offer a unique perspective to examine the findings of these and similar experimental studies.
For the sake of focus and instruction, we have modeled one estimate per disturbance, i.e. one estimated world rotation. This assumption played a crucial role in some of our findings on interference. If instead we had allowed for multiple estimates of a world rotation, it is not clear how the model would predict interference when modeling adaptation to counter disturbances across multiple days. Indeed, other studies have found that under appropriate conditions, a newly acquired motor behavior can be consolidated and resist retrograde interference [32], [33]. Our model does not predict these findings but extensions that could also explain these effects would be interesting. Such extensions might be possible by introducing parameters to describe multiple visuomotor disturbances, each with their own uncertainty. Such a model could implement a form of supervised adaptation; after adapting to, or operating within, a specific visuomotor disturbance for a long time the model could grow certain of this parameter value. Then, adapting to a similar but oppositely directed disturbance would require adapting another, less certain, visuomotor parameter. Such a scheme might implement adaptation to multiple disturbances, consistent with the idea of consolidating a motor behavior and learning a second, distinct behavior without interference.
We feel that much if not all of the model's value lays in the intuition it yields in trying to explain motor behavior phenomena. The studies and accompanying simulated results we present are those that we feel the model may help to explain. However, as with all models, this model is necessarily false [34], and there are experimental findings the model either cannot explain or that are flatly at odds with its predictions. For example, though our model is consistent with the findings on adapting to gradually versus abruptly presented perturbations in the Klassen et al. study, a more recent examination found distinct results. In this new study rates of motor decay were probed during short-term adaptation to a force field, either abruptly or gradually introduced [35]. Though the aim and experimental protocol of this study was very different from the Klassen study, some apparent contradictions were found in that there were no effects on the re-adaptation to the force field between the abrupt and gradual groups. To be clear, this finding was made under conditions of short-term savings of a force field (not long-term retention of a viruomotor disturbance), and obtained with the use of error clamps. However, despite their differences, it is not obvious to us how our model could account for these two distinct findings.
In contrast with the gradual vs. abrupt findings presented above however, our model makes an interesting prediction that could readily be tested. According to our model the amount of adaptation for world parameters is due to both the size of disturbance and the amount of training; the larger the disturbance and the more training time, the more a world parameter is adapted. Similarly, the more world parameters are adapted, the more savings should be observed on a second day's presentation of the disturbance. Surprisingly though, our model predicts that even with a gradually introduced perturbation, and one that never reaches the strength of the abruptly presented one, more savings can be observed on a second day. If the model is presented with a visuomotor rotation that is ramped up slowly over many trials, the world estimate will have relatively more time to adapt, and the body estimate more time to de-adapt. As a result, even if adaptation ends before the visuomotor disturbance has reached, say 30°, the world estimate will surpass that seen in the abruptly presented paradigm. Thus more savings, not less, will be observed on the second day. The results of such an investigation would be very informative for the study of adaptation.
Another study of force field adaptation offers both supporting and contradicting evidence for our model. In this study the rates at which subjects adapted (as quantified through movement errors) were compared when adapting either to the null field or a scaled down version of the force field [9]. Consistent with our model, de-adapting to the null field is much faster than adapting to the force field. It was also found that subjects adapted to the scaled down force field even faster than they did the null field. In contrast with this finding, our model would predict that both the body and the world would adapt to the scaled down force field, resulting in a relatively slow process. This is in sharp contrast with their findings and will provide an interesting target for future modeling efforts.
In our model we have assumed that movement predictions always utilize body estimates. Since the body is always relevant, this seems sensible. One consequence of this is that the model is in effect “blind” to changes it has inferred are due to the body; the model cannot make predictions for movements that do not compensate for these adapted body estimates. Even if the inferred body estimates are due to an experimental perturbation, the model will have an altered prediction of where the limb will be in space. In effect, the act of adapting alters the model's perception of the limb. Interestingly, there is a growing body of experimental evidence for this same effect. In particular, the act of adapting to a visuomotor disturbance biases the perception of subjects' movement and hand position in a manner consistent with our model [36], [37]. This perceptual bias was found to be nearly half of the adapted rotation, also consistent with our model. Importantly, this bias was found to be associated with the limb alone, and not the result of a global recalibration of visual space [36]. A similar finding demonstrates that adapting to a force field alters the perception of the limb in space as well [38]. On the whole, these results suggest a further link between our model's use of body and world parameters and how the nervous system adapts to new motor behaviors.
Some of our results on interference rely on the relative duration of the counter disturbance, behavior B, in the A-B-A paradigm. Since both disturbances are presented for the same length of time, the counter disturbance almost completely degrades any estimate of the world parameter estimate. This results in motor patterns consistent with interference. If the counter disturbance was presented for approximately twice as long, our model predicts that the typical pattern of interference would not be observed. Rather than producing errors similar to naïve subjects, our model predicts subjects should produce larger errors, consistent with the expectation of the counter disturbance. As far as the authors are aware, this particular experimental result has not yet been performed and would be particularly informative.
In this study we have implicitly assumed that savings is a form recall; previously adapted information is called upon resulting in reduced motor errors relative to naïve conditions. However, other researchers have asserted that savings could be a form of meta-adaptation instead, wherein adaptation rates are facilitated and motor errors decrease faster than during naïve conditions [e.g. 8,39]. By the same token, interference could either be a form of re-adaptation and hijacking of previously adapted behaviors (as we have assumed), or an inability to recall previously adapted information. To the best of the authors' knowledge, both of these options for savings and interference are consistent with the known empirical evidence. However, our model does make some predictions that might speak to these possibilities. For example, our model predicts that on repeated days of training, the estimate of a world-imposed disturbance progressively increases. Assuming cues such as the experimental apparatus are salient for estimating relevance, each day's initial errors should be smaller than the previous. This implies that subjects should eventually display “one-shot” learning of a disturbance. This would be strong evidence that subjects were in fact recalling knowledge, rather than nearly instantaneously adapting. Future studies could examine a similar line of predictions to distinguish between savings as recall, and savings as meta-adaptation.
Relevance as we have defined it here is a relatively simplistic indication of the motor system's current operating condition, or context. Clearly there is more to context than motor errors. For instance, whether or not one is holding the handle of a robot is a clear indicator of the kind of disturbances one might expect [20]. Similarly, while they may not be as salient, cues such as tones and colors may also serve for disambiguating context [40]. Finally, in this study we have completely neglected forces, both the contact forces between the limb and the robot handle, and the forces required to produce movements. This is clearly an oversimplification and it is known that these forces are relevant when adapting [e.g. 41,42,43]. Why some cues are easy to indicate context and others are difficult remains and open question. Which variables the nervous system uses to distinguish context are similarly unknown. We expect that future studies will shed more light on these issues.
|
10.1371/journal.pgen.1006030 | Germline Defects Caused by Smed-boule RNA-Interference Reveal That Egg Capsule Deposition Occurs Independently of Fertilization, Ovulation, Mating, or the Presence of Gametes in Planarian Flatworms | Few animals are known to lay eggs in the absence of ovulation or copulation, as it is presumably energetically wasteful and subjected to negative selection. Characterization of Smed-boule, a member of the DAZ family of germline RNA-binding proteins, revealed that egg capsule (or capsule) production and deposition occurs independently of the presence of gametes in the planarian flatworm Schmidtea mediterranea. Reduction of Smed-boule expression by RNA-interference (RNAi) causes ablation of spermatogonial stem cells and the inability of ovarian germline stem cells to undergo oogenesis. Although animals subjected to Smed-boule RNAi lose their gametes and become sterile, they continue to lay egg capsules. Production of sterile capsules is even observed in virgin Smed-boule(RNAi) and control planarians maintained in complete isolation, demonstrating that egg production in S. mediterranea occurs independently of ovulation, fertilization, or mating. Evidence suggests that this is a conserved feature amongst Platyhelminthes, and therefore relevant to the pathology and dissemination of parasitic flatworms. These findings demonstrate that Smed-boule functions at different stages during male and female germline stem cell development, and also demonstrate that egg capsule production by planarian flatworms occurs independently of signals produced by mating or ova.
| Our work shows that production and deposition of egg capsules by planarian flatworms does not require fertilization, mating, ovulation, or even the existence of gametes. We also uncovered evidence for the existence of gender-specific germline stem cells in Schmidtea mediterranea, a hermaphroditic species of flatworm that develops germ cells post-embryonically. These findings surfaced from the characterization of Smed-boule, a member of the Deleted in AZoospermia gene family of RNA-binding proteins required for germline development in a broad range of animals. These findings lead to a better appreciation of the evolutionary diversity in approaches to oviparity. Additionally, discovering that egg capsule production occurs independently of germline or mating activities may carry a potential applied aspect with regards to regulating the dissemination and pathology of parasitic flatworms (such as blood flukes and tapeworms), if conserved in these organisms.
| The characterization of developmental processes involved in sexual reproduction has important implications towards reproductive medicine, stockbreeding, farming, and for controlling the dissemination of infectious disease. Evolutionarily conserved molecular processes involved in metazoan germline development have been identified through decades of research using model organisms. For example, post-transcriptional regulation of gene expression by conserved germline-specific RNA-binding proteins is one of the conserved molecular processes that ensure development of gametes [1–3]. On the other hand, there is great diversity in the processes that occur during and after fertilization, many of which are the outcome of speciation events [4,5].
Planarian flatworms belong to the phylum Platyhelminthes, and are well known for their extraordinary regenerative abilities, which are founded in the availability of a pluripotent stem cell population throughout their life [6–9]. The evolutionary history of these organisms has yielded extreme divergence of reproductive strategies, both between and within populations of different planarian species. For example, there are planarians that rely exclusively or temporally on asexual reproduction, which involves transverse fission and stem cell driven regeneration [7,10]. There are also populations of planarians that reproduce predominantly through parthenogenesis (Pongratz et al., 2003). However, the default reproductive strategy of turbellarians is believed to be hermaphroditic sexual reproduction [11], more specifically for planarians, through cross-fertilization and oviparity [12]. By contrast, some parasitic flatworms (i.e. schistosomes or blood flukes) have complex life cycles that involve dioecious and asexual reproductive phases during transitions between vertebrate and invertebrate hosts, respectively [13]. Since the complex life cycle of schistosomes complicates husbandry and experimentation in laboratory settings, researchers have begun to use planarian flatworms as a model to dissect the molecular mechanisms behind the extensive lifespan and reproduction of their parasitic cousins [14]. One aspect of particular interest is the continuous production of thousands of eggs that both facilitate dissemination and sustain the pathology of schistosomes by populating organs of their host [13,14].
Planarian flatworms have become useful models for the study of metazoan germline development [12,15]. In general, the specification of germline stem cells can occur through mechanisms that involve: 1) inherited material deposited in the cytoplasm of the maturing oocyte (preformation); or 2) embryonic stem cell differentiation in response to inductive cell-to-cell interactions (epigenesis) [16,17]. Inductive determination occurs in mice and is also observed in planarians, both initially and during regeneration of fragments that lack germ cells, and it occurs through differentiation of pluripotent somatic stem cells called neoblasts [18]. In the planarian species Schmidtea mediterranea, germline stem cells are first detected as dorsolateral clusters in the area where testes develop [18]. In other planarian species, such as Dugesia ryukyuensis, germline stem cells are first detected in the area of the ovaries [19,20]. Upon feeding and growth, planarians that reproduce sexually develop a hermaphroditic reproductive system and their gonads begin continuous production of gametes [12,19]. Germline stem cells in the ovary enter oogenesis and produce oocytes of approximately 40 μm in diameter that exit the ovary and are fertilized by sperm deposited in the tuba [12,15]. Even-though oocytes are large in comparison to other planarian cells (e.g. neoblasts are ~8 μm diameter) these do not hold the nutrients necessary for embryonic development, as is normally observed in eggs of insects, amphibians or fish (to name a few). Instead, planarian yolk glands (vitellaria in other flatworms) produce separate cells that provide material required for egg capsule shell formation and nurturing embryonic development [21,22].
The development of planarian germline stem cells depends on conserved post-transcriptional regulators such as Nanos and Bic-C [15,18]. Boule is an RNA-binding protein encoded by the basal member of the Deleted in AZoospermia (DAZ) gene family, which is required for germ cell development in species ranging from sea anemone to humans [23,24]. How DAZ family homologs contribute to germline development in planarians remains unknown. In this study, we characterize a Boule homolog in the planarian Schmidtea mediterranea and demonstrate that it functions at different stages during male and female germline development. Functional analyses by RNA-interference (RNAi) revealed that Smed-boule is required for development and maintenance of spermatogonial stem cells, but disposable for the existence of their oogonial counterparts, uncovering the presence of sex-specific germline stem cells in planarian hermaphrodites. Long-term analysis of Smed-boule knockdowns revealed that egg capsule deposition in planarians is not triggered by gametogenesis, ovulation, oocyte activation, fertilization, or mating. These results demonstrate that egg capsule formation occurs regardless of signals from sexual activity or germ cell activity in S. mediterranea. These findings also provide a unique opportunity to identify internal mechanisms that influence capsule production in Platyhelminthes, which is central in the dissemination and pathology of parasitic members of this phylum.
We identified a boule homolog in the planarian flatworm S. mediterranea with a region of amino acid sequence 55% identical with that of the RNA recognition motif of human BOLL (E-value = 1e-23; Fig 1A). The protein encoded by this gene shared highest homology with members of the Boule-like subfamily, as compared with other members of the DAZ family of proteins (Fig 1B). Expression of this gene (from here on referred to as Smed-boule or boule) was detected by whole-mount in situ hybridization (ISH) in testes and ovaries of sexually mature planarians that are actively laying egg capsules (Fig 2A–2D). Smed-boule expression was also detected in testis primordia of hatchlings and sexually immature animals (Fig 2D and 2D’). To better understand the distribution of Smed-boule expression in testes and ovaries, we performed detailed analysis by double fluorescent in situ hybridization (FISH) with the germline stem cell marker Smed-nanos [18] (Fig 3). Detection of Smed-boule mRNA overlapped with that of Smed-nanos in testes (Fig 3A) and partially in ovaries (Fig 3B). The presence of Smed-boule mRNA was also robustly detected in the spermatogonial layer of the testes (Fig 3A). Detection of Smed-boule expression was not apparent in the soma, and thus we conclude that expression of this gene is restricted to the germline in S. mediterranea.
To test the function of Smed-boule in planarian germline development and sexual reproduction we subjected planarians to three months of RNAi. Planarians continuously turn over all cells in their body from a continuous population of pluripotent stem cells, which allowed us to assess whether Smed-boule is required for normal germline development in sexually mature adults using germ cell markers (Fig 4A–4F; S1 and S2 Figs). Groups of seven sexually mature planarians were fed liver supplemented with 100 ng/μl of double-stranded RNA (dsRNA) twice per week. Smed-boule knockdowns (Smed-boule(RNAi)) were compared to control(RNAi) planarians. DsRNA corresponding to a planarian Cytoplasmic Polyadenylation Element Binding Protein 1 homolog (CPEB1), which is required for yolk gland development and egg capsule production (below), was administered to an additional group (CPEB1(RNAi)) as readout of RNAi effectiveness overtime. At the end of three months of RNAi, we observed that both oocytes (Fig 4B; S2B Fig) and sperm (Fig 4E; S2B” Fig) were absent in Smed-boule(RNAi). No defects in oocyte or sperm development were observed in control(RNAi) planarians (Fig 4A and 4D; S2A–S2A” Fig). The testes of CPEB1(RNAi) samples were fully developed (Fig 4F), but their ovaries displayed abnormal morphology and distribution of oocyte marker expression (Fig 4C). From these results we concluded that Smed-boule is required for development of sperm and ova in S. mediterranea.
Neoophoran flatworms rely on a particular approach to oviparity in which nutrients for the developing embryo (yolk) are not accumulated in the developing ova. Instead, nutritional support is contributed by yolk cells (vitellocytes), which are transferred from yolk glands to the planarian uterus and encapsulated with embryos during egg capsule deposition. We checked for the presence of yolk glands using the yolk cell marker Smed-surfactant b (S1 Fig), which proved to be of comparable abundance and distribution in control(RNAi) and Smed-boule(RNAi) animals (Fig 4G and 4H). However, the presence of yolk glands in CPEB1(RNAi) animals was severely reduced (Fig 4I). We looked for other possible defects in the development of somatic reproductive structures but were unable to find any abnormalities other than the absence of accumulated sperm in the seminal vesicles of Smed-boule(RNAi) (Fig 5), which is due to their inability to produce sperm (Fig 4E). There was also a noticeable difference in size of CPEB1(RNAi) planarians, which were on average 30.8% larger than control animals maintained under the same conditions (n = 14; unpaired two-tailed t-test, p < 0.05; S3 Fig). The normality and functionality of the accessory reproductive system in Smed-boule(RNAi) was further supported by quantitative analyses of egg capsule production (below).
As expected, from planarians with underdeveloped yolk glands (Fig 4I), CPEB1(RNAi) ceased laying eggs within a month of RNAi (Fig 4J). The rate of egg production in Smed-boule(RNAi) was unaffected during the three months of RNAi treatment (Fig 4J). Both the control(RNAi) and Smed-boule(RNAi) groups continuously laid eggs for the entirety of the experiment (Fig 4J). In fact, an increase of 33% and 30% in egg capsule production was observed in Smed-boule(RNAi) when compared to control(RNAi) planarians during the second and third months of RNAi treatment, respectively (unpaired two-tailed t-test, p < 0.05; Fig 4J). Given the surprising result that planarians devoid of gametes continued to deposit egg capsules, we monitored and quantified the number of fertile capsules (yielding progeny) produced by the different knockdown groups for two months after capsule deposition. From this, we discovered that egg capsules produced by animals subjected to two months of Smed-boule RNAi completely ceased to hatch (Fig 4K). Egg capsules produced by control(RNAi) groups hatched 22% to 48% of the time (Fig 4K). From these results, we concluded that Smed-boule function is required for germline development and sexual reproduction in S. mediterranea, but dispensable for production of egg capsules. Furthermore, the continuous production of egg capsules by groups of Smed-boule(RNAi) planarians (Fig 4J) suggested that production and deposition of egg capsules do not require fertilization, contributions from sperm, ovulation, or the presence of oocytes.
Given the fact that Smed-boule(RNAi) planarians were capable of producing sterile egg capsules in the absence of germ cells (and therefore fertilization events), we hypothesized that control animals would also produce sterile egg capsules in the absence of fertilization events. To test this hypothesis, we obtained ≤ 1 week-old hatchlings (which lack ovaries, testes, yolk glands, and accessory reproductive organs) and maintained them in isolation for four months under continuous RNAi regimens. Planarians were maintained in isolation throughout the experiment, which allowed us to test whether egg capsule production is independent of signals produced during mating or the presence of potential mates altogether. Since planarians in this experiment were subjected to Smed-boule RNAi within a week of being born, which is a point when no sperm has developed, this approach also allowed us to verify that lingering sperm in adult knockdowns used in the previous experiment was not contributing to egg capsule production. Two categories of isolated virgins were maintained on either liver containing Smed-boule dsRNA or control dsRNA and were fed twice per week. These animals were expected to grow and eventually reach sexual maturity under these husbandry conditions. The production of egg capsules would only occur if independent from stimuli produced during mating, fertilization, embryonic development and, in the case of Smed-boule(RNAi), the absence of gametes.
Indeed, both control and Smed-boule(RNAi) isolated animals produced egg capsules during the third and fourth months of the experiment (Fig 6A). The number of capsules produced during the length of the experiment by isolated individuals from each category ranged from none to six (Fig 6B). The average number of capsules deposited by individuals in the control category was slightly, but not significantly higher than those of Smed-boule(RNAi) (unpaired two-tailed t-test, p = 0.25) (Fig 6B). As expected from results observed in animals subjected to RNAi in the presence of potential mates (Fig 4J and 4K), none of the egg capsules produced by Smed-boule(RNAi) individuals were fertile (n = 0/28 capsules). Capsules produced by control RNAi animals were also completely sterile (n = 0/43 capsules), suggesting that the production of egg capsules in these animals were not due to self-fertilization or parthenogenesis. We verified that normal gamete development was present in control animals at the end of the isolation experiment (Fig 6C and 6E) and absent in Smed-boule(RNAi) flatworms (Fig 6D and 6F), which was expected from analyses of knockdowns not maintained in isolation (Fig 4A, 4B, 4D and 4E). We also validated successful development of yolk glands in control and Smed-boule(RNAi) planarians raised in isolation (Fig 6G and 6H). Collectively, these results demonstrate that production of egg capsules in S. mediterranea occurs in response to internal triggers that do not require the presence of a mate, mating, or fertilization events. Furthermore, the production of egg capsules by Smed-boule(RNAi) planarians suggests that this trigger is detached from signals originating from sperm and oocyte development or ovulation.
We decided to evaluate the severity of germline development defects caused by Smed-boule RNAi. The most severe phenotype would be the loss of germline stem cells, which are specified and maintained post-embryonically through neoblast differentiation [18]. Germline stem cells in the planarian ovaries and testes can be identified by the characteristic expression of germinal histone H4 and nanos [18,20,25,26]. We tested for the presence of germline stem cells in control(RNAi) and Smed-boule(RNAi) by nanos ISH after 3–4 months of RNAi (at the end of experiments in Figs 4J–4K and 6A–6B). Whole-mount ISH analysis of germinal histone H4 and nanos expression revealed the presence of germline stem cells in the testes region of control(RNAi) animals (Figs 7A and 8A). However, germline stem cells were completely absent from the testes region of Smed-boule(RNAi) planarians (Figs 7B and 8B). Surprisingly, germline stem cells in the ovary region of both control(RNAi) and Smed-boule(RNAi) planarians were readily detectable (Figs 7A’, 7B’, 8C and 8D). Identical results were observed from hatchlings raised subjected to Smed-boule RNAi while maintained in isolation (S4 Fig). Furthermore, analysis of germline stem cells in presumptive testis primordia present in asexual strains of S. mediterranea (Wang et al., 2007) were also lost after Smed-boule RNAi (S5 Fig). From these results, we conclude that spermatogenesis defects in Smed-boule(RNAi) are due to the absence of male germline stem cells, whereas defects in oogenesis occur further downstream in the differentiation pathway.
To better evaluate the progression of oogenesis in Smed-boule(RNAi) ovaries, we analyzed control and Smed-boule knockdowns stained with DAPI by confocal microscopy (Fig 8E and 8F; S1 and S2 Movies). DAPI is retained by DNA and allowed for the visualization of numerous large oocytes with condensed chromosomes in the ovaries of control samples (Fig 8E; S1 Movie). In contrast, neither oocytes, cells with condensed chromosomes, or otherwise recognizable mid- to late-oogenic intermediates, were detectable in ovaries of Smed-boule(RNAi) flatworms (Fig 8F; S2 Movie). From these results we conclude that Smed-boule function is required during the initial stages of oogenesis, sometime before development of primary oocytes, but after specification of ovarian germline stem cells.
The different outcomes observed on germline stem cells of testes and ovaries following Smed-boule(RNAi) reveal that these are two fundamentally distinct germline stem cell populations that require Smed-boule function at different developmental stages. Smed-boule function is necessary for neoblast differentiation into male germline stem cells and/or maintenance of male germline stem cells, whereas ovarian germline stem cells only require Smed-boule for progression through early stages of oogenesis (Fig 8G). Furthermore, the severe defects in germline development observed after Smed-boule RNAi further support the hypothesis that egg capsule production and deposition occur independently of gametes, ovulation, parthenogenesis, fertilization, mating, or embryonic development in S. mediterranea.
Collectively, the data presented demonstrate that production and deposition of the egg capsules that ensure development of planarian embryos occur independently of fertilization events. Rather, it seems that egg capsule deposition, at least in S. mediterranea, is driven by intrinsic signals that are activated once these flatworms grow past a certain size and develop their yolk glands and other accessory reproductive organs. Given these findings, conclusions regarding planarian “fecundity” previously calculated from the rate of capsule production [27,28], may need to be re-evaluated. Additionally, knowledge of the separation between capsule deposition and fertility should assist in the study of planarian germline and embryonic development, as well as in generation of methodologies for transgenesis, which have proven elusive to this point.
Planarian reproduction can occur asexually through transverse fission, or sexually through post-embryonic development of a hermaphroditic reproductive system [7]. In planarians committed to sexualization, the development of gonads and gametes precedes formation of the oviducts, sperm ducts, and copulatory organs [19]. Yolk gland development in S. mediterranea, which is essential for production of egg capsules, is initiated towards the end of sexual development depending on sufficient nutritional intake and growth. The rate of egg capsule production observed in our experiments (1 to 5 egg capsules per animal per month; Fig 4J) is comparable to those observed in different planarian species both in their natural habitat and raised under laboratory conditions following a similar liver-only diet [29,30]. Therefore, we believe that the conditions used for husbandry of S. mediterranea in the laboratory are conducive to normal egg capsule production rates, and that this is not the limiting factor in reproductive output. However, the low yield of fertile egg capsules observed from control animals in our experiments (22% to 48%; Fig 4J and 4K) suggests that optimal laboratory husbandry conditions need to established to promote oocyte production, ovulation, or mating (either of which may be rate limiting in actual reproductive output).
How can triggering egg production independently of fertilization be an efficient approach to survival of planarian populations? First, we must consider that in terms of sexual reproduction, S. mediterranea performs rather poorly under laboratory conditions. This is supported by the studies of Jenkins and Brown [29] who observed D. dorotocephala yield an average of 16.5 hatchlings per egg (approximately 10-fold higher from what is observed in our laboratory for fertile egg capsules of S. mediterranea). Studies in S. polychroa have shown that siblings emerging from a single egg result from different fertilization events, which is possible because sperm from one or more partners can be stored for at least a month after insemination [31]. The ability to store sperm for an extended period of time after insemination, combined with the delay in development of yolk glands in comparison to the rest of the reproductive system, presents a scenario that would benefit from a mechanism that triggers capsule formation independently of copulation. In fact, it would be optimal if the activation of capsule formation also triggers ovulation of the many fully-grown oocytes present in ovaries of sexually mature planarians (Consequential Model; S6A Fig). Since sperm can be stored in the tuba, massive ovulation could maximize the number of hatchlings generated per capsule. Alternatively, encapsulation of multiple embryos in a single egg capsule could also be facilitated by extended storage of zygotes prior to capsule deposition (Complete Autonomy Model; S6B Fig). We are currently unable to differentiate between these two possibilities, or the possibility that passage of oocyte precursors (e.g. oogonial or female germline stem cells) may activate capsule formation. Indeed, oocytes were not detected in Smed-boule(RNAi) planarians (n = 0/20; Figs 4B and 6D; S2B Fig), but ovaries and oogonial stem cells were readily observed (n = 14/15; Figs 7B, 8D and 8F; S4D Fig; S2 Movie). It is possible that the release of early oocyte precursors from the ovary triggers capsule formation. However, this hypothesis is challenged by the fact that dozens of hatchlings often emerge from single capsules of different planarian species, and the observation that the rate of capsule production was not compromised in Smed-boule(RNAi) when compared to control planarians (which contained both oocytes and precursors). Nevertheless, current and previous observations do support a model by which a sustainable approach to oviparity could rely on a trigger for capsule formation that is independent of mating, fertilization, or ovulation.
Inside the phylum Platyhelminthes, free-living species (such as S. mediterranea) are evolutionarily distant from members of parasitic groups (Trematoda, Monogenea, and Cestoda). However, the non-causative relationship between ovulation/fertilization and capsule production appears to be conserved in some cestodes and trematodes, whose dissemination and pathology depend on the continuous production of egg capsules. Parasitic flatworms of the genus Schistosoma have been reported to produce egg capsules from females after single-sex infections of mammalian hosts [32,33]. Although female schistosomes depend on interactions with a mate to fully grow and develop their gonads, they are also able to develop some vitelline cells and immature ovaries on their own. Shaw [33] observed the production of infertile capsules from females without male stimuli, probably through mechanisms conserved with those reported here for planarian flatworms. Similarly, parasitic flatworms belonging to the class Cestoda (tapeworms) have been reported to produce unviable egg capsules in the absence of fertilization events when cultured in vitro [34,35]. Thus, given that continuous production and deposition of egg capsules is central to dissemination and pathology of different types of parasitic flatworms, the molecular machinery involved in egg capsule production (and not germline development alone) becomes a desirable target for therapeutic developments.
A laboratory sexual strain of Schmidtea mediterranea [36] was used all experiments, except for those presented with asexual planarians [37] in S5 Fig. Planarian cultures were maintained in 0.75x Montjuïc Salts at 18°C under dark conditions, whereas 1x Montjuïc Salts and 21°C were used for asexuals as per [37]. Planarians were exposed to room temperature and light during weekly feedings of pureed organic beef liver (Vantage USA, Oak Park, Illinois). Experimental animals were starved at least seven days before experimentation.
Smed-boule was identified from a collection of S. mediterranea contig sequences assembled from RNAseq and conventional cDNA expressed sequence tag reads ([38]; https://www.ideals.illinois.edu/handle/2142/28689). A PCR product corresponding to Smed-boule ORF sequence was amplified from oligo(dT)-primed total RNA cDNA using 5’-GTTGTTTCAACGGTTCTACTGGCATC -3’ and 5’- GATTATTCCGGACAAAGCTGGACAAG -3’ forward and reverse primers (respectively) and ligated to pJC53.2 [39] after Eam1105I restriction digest. The insert sequence was validated and deposited into NCBI under accession number KT709533.
Fixation and preparation of samples for whole-mount in situ hybridization and DAPI staining were performed as per King and Newmark [40]. Colorimetric development for visualization of riboprobes was performed as described by Pearson et al. [41]. Smed-boule riboprobes were synthesized using SP6 RNA Polymerase. Smed-CPEB1 (NCBI accession number KU990884), Smed-nanos, were also synthesized from a pJC53.2-based construct [39], whereas Smed-synaptotagmin XV, Smed-granulin (grn), Smed-surfactant b, germinal histone H4, and a tuba/oviduct marker were synthesized from pBluescript-based clones (PL04017B1F10, PL05005A1F08, PL010001001D12, pBS-gH4, and PL04015A2A02, respectively [18,36,42]) using T3 RNA Polymerase. Colorimetric and low magnification analyses of DAPI signals from testes were imaged on a Zeiss Axio Zoom.V16 stereoscope. Confocal images were captured on an Olympus FluoView FV1000 confocal microscope hosted at Wright State University’s Microscopy Core Facility.
Double-stranded RNAi feedings were performed twice every seven days and the protocol was followed as previously described [43]. DsRNA corresponding Escherichia coli ccdB sequence, which does not affect planarian development or behavior was used for unaffected control groups. For isolated RNAi samples, each planarian was fed individually and in isolation. For other experiments, planarians were maintained in groups of seven animals.
Groups of seven sexual planarians of 0.5 to 0.7 cm size and with a visible gonopore were maintained in glass Petri dishes and subjected to dsRNA feedings as described above. For isolated experiments, single ≤ 1 week-old hatchlings were maintained in isolation in glass Petri dishes throughout the experiment, under the husbandry conditions described above. Isolated planarians were fed liver containing control or Smed-boule dsRNA twice per week, at which point any capsules present were collected and isolated. DsRNA corresponding to E. coli ccdB sequence was used for control samples. Egg capsules were monitored for hatchling events weekly for a period of three months after deposition.
|
10.1371/journal.ppat.1001205 | Tick Histamine Release Factor Is Critical for Ixodes scapularis Engorgement and Transmission of the Lyme Disease Agent | Ticks are distributed worldwide and affect human and animal health by transmitting diverse infectious agents. Effective vaccines against most tick-borne pathogens are not currently available. In this study, we characterized a tick histamine release factor (tHRF) from Ixodes scapularis and addressed the vaccine potential of this antigen in the context of tick engorgement and B. burgdorferi transmission. Results from western blotting and quantitative Reverse Transcription-PCR showed that tHRF is secreted in tick saliva, and upregulated in Borrelia burgdorferi-infected ticks. Further, the expression of tHRF was coincident with the rapid feeding phase of the tick, suggesting a role for tHRF in tick engorgement and concomitantly, for efficient B. burgdorferi transmission. Silencing tHRF by RNA interference (RNAi) significantly impaired tick feeding and decreased B. burgdorferi burden in mice. Interfering with tHRF by actively immunizing mice with recombinant tHRF, or passively transferring tHRF antiserum, also markedly reduced the efficiency of tick feeding and B. burgdorferi burden in mice. Recombinant tHRF was able to bind to host basophils and stimulate histamine release. Therefore, we speculate that tHRF might function in vivo to modulate vascular permeability and increase blood flow to the tick bite-site, facilitating tick engorgement. These findings suggest that blocking tHRF might offer a viable strategy to complement ongoing efforts to develop vaccines to block tick feeding and transmission of tick-borne pathogens.
| Ticks are distributed worldwide and affect human and animal health by transmitting diverse infectious agents. Safe and effective vaccines against most tick-borne pathogens are not currently available. Typical vaccines target microbes directly, using extracts of the organism, or recombinant antigens as the immunogen; the transmission of tick-borne pathogens can also theoretically be prevented by interfering with the ability of ticks to feed on a mammalian host. In this study, we have characterized a putative histamine release factor (tHRF) from I. scapularis ticks, the predominant vector of B. burgdorferi, the agent of Lyme disease in North America. Our results suggested that tHRF is presented in tick saliva and critical for tick feeding; blocking tHRF markedly reduced the efficiency of tick feeding, and reduced the B. burgdorferi burden in mice. This finding provides novel insights into the molecular mechanisms of tick feeding and provides a potential vaccine target to block tick feeding and pathogen transmission.
| Ticks are distributed worldwide and affect human and animal health by transmitting diverse infectious agents. Ticks are considered to be second to mosquitoes as major vectors of human diseases [1], [2]. For example, Ixodes spp., transmit Borrelia burgdorferi (the Lyme disease agent), Anaplasma phagocytophilum (the cause of human granulocytic anaplasmosis), Babesia microti, and tick-borne encephalitis virus (TBEV), among other pathogens [1], [3].
Effective vaccines against most tick-borne pathogens are not currently available and there is an urgent need for the control of ticks and their associated pathogens [4]. Typical vaccines target microbes directly, using extracts of the organism, or recombinant antigens as the immunogen. For example, B. burgdorferi outer surface protein A has been extensively studied and resulted in an Federal Drug Administration-approved vaccine that was commercially available from 1998 until 2002 [5], [6]. Currently one vaccine is approved and available for protection against a tick-borne pathogen – TBEV, which is transmitted by I. ricinus in Northern Europe and Asia [4].
The transmission of tick-borne pathogens can also theoretically be prevented by interfering with the ability of ticks to feed on a mammalian host [7]. A pilot study by Allen and Humphreys several decades ago, suggested that vaccines based on tick gut antigens successfully reduced Boophilus engorgement on cattle [8]. Recently, immunization of guinea pigs with a tick salivary antigen, sialostatin L2, diminished the capacity of Ixodes scapularis nymphs to feed [9]. While reducing the ability of tick feeding, tick-based vaccines may have another equally important impact – to decrease the chance of transmission of tick-borne pathogens [10]. Immunization of cattle with Bm86 vaccines resulted in lower infestations as well as decreased incidence of babesiosis and Anaplasma marginale infection in some regions [1], [7]. Repeated exposure of guinea pigs to ticks causes acquired resistance of the animals to subsequent tick bites [11], [12], and this development of “tick-immunity” can decrease tick-transmitted B. burgdorferi infection [13]. B. burgdorferi need to replicate within the ticks during blood feeding and are transmitted to the host after 36–48 h of tick attachment [2], [14], [15], [16]. Thus, impairing I. scapularis feeding could be another useful strategy to reduce B. burgdorferi transmission.
Tick saliva contains molecules that are important for formation and maintenance of the feeding cavity in the host dermis, as well as the transmission of tick-borne pathogens [17], [18]. These activities include anti-hemostatic, anti-inflammatory and immunomodulatory effects, among others [17], [19]. Histamine binding proteins are well characterized and suggested to be important to neutralize the inflammatory effect of histamine, which is secreted by host immune cells at the tick feeding site and critical for Ixodes ticks to successfully attach to the host [20], [21]. Interestingly, Dermacentor variabilis ticks also express a protein in their saliva, which shares high homology with mammalian histamine release factor [22]. Given the deleterious effects of histamine on tick physiology, it is very surprising that ticks encode a histamine release protein that would presumably stimulate histamine secretion. The role of the tick histamine release factor in vivo during tick feeding is not understood and warrants detailed examination.
Tick feeding can be divided into a series of 9 stages [2] beginning with host seeking, and culminating in engorgement on the host followed by detachment and dropping off the host. A feeding lesion is established about 24 h post attachment, and during this early phase of feeding there is minimal blood intake. Blood meal ingestion begins slowly around 48 h post tick attachment followed by rapid feeding to repletion around 72 h–96 h post tick attachment (late stage). While it is recognized that the I. scapularis salivary gland proteome changes during these early and late phases of feeding [13], a molecular understanding of these events remains to be elucidated.
In this study, we have characterized a putative histamine release factor from I. scapularis, the predominant vector of B. burgdorferi, the agent of Lyme disease in North America. We invoke a pivotal role for I. scapularis HRF during the rapid phase of tick feeding, and address the vaccine potential of this antigen in the context of tick engorgement and B. burgdorferi transmission.
To identify tick proteins that may be utilized by B. burgdorferi to facilitate transmission, 2-dimensional fluorescence difference gel electrophoresis (DIGE) was performed using extracts from B. burgdorferi-infected, and uninfected, I. scapularis salivary glands. Seventeen differentially expressed proteins (5-fold or more expression levels in Borrelia-infected salivary glands) were selected for mass spectrometric analysis, and 4 I. scapularis proteins were unambiguously identified with significant MASCOT scores (p<0.05) of 79 (Table S1). In this study, we characterize one of the most highly induced proteins, named tHRF because it shares high homology with a murine histamine release factor (57.1% similarity and 40.1% identity at amino acid level). tHRF mRNA levels were induced during I. scapularis engorgement, and significantly higher in B. burgdorferi-infected, than in uninfected, ticks (p<0.01). Immunoblots using tHRF antiserum further demonstrated a ∼2.5 fold up-regulation of tHRF in B. burgdorferi-infected ticks (Figure 1, A, C–D). tHRF was present in tick saliva, as well as in the salivary glands and midgut, indicating that it is a secreted protein (Figure 1E).
To analyze the potential role of tHRF in tick feeding, and also during B. burgdorferi transmission, tHRF-deficient I. scapularis nymphs were generated by RNA interference (RNAi). Buffer-injected (MOCK), SSRB (another tick gene- Single Sequence Receptor Beta- found in our 2DIGE list, used as a control) or tHRF double-stranded RNA (dsRNA)-injected B. burgdorferi-infected nymphs were allowed to engorge on mice. The silencing of tHRF and SSRB were confirmed by quantitative RT-PCR (Figure 2A). After 3 days, the weighs of tHRF-deficient ticks were significantly lower than control ticks (Figure 2B). Q-PCR revealed a decrease in spirochete levels in tHRF-deficient ticks, as well as in the skin of mice that were fed upon by tHRF-deficient ticks (Figure 2, C and D). At 3 weeks, when spirochetes have disseminated to diverse organs, the B. burgdorferi burden in the heart and joints was also lower in mice infected by tHRF-deficient ticks, compared to that in mice infected with control ticks (Figure 2, E and F).
To further show that tHRF directly influences tick feeding, an RNAi study was performed with nymphs that were not infected with B. burgdorferi. Consistent with the results using B. burgdorferi-infected ticks, the tick weight was significantly decreased in tHRF-dsRNA-treated uninfected I. scapularis after feeding (Figure 2G). These data show that tHRF is critical for tick feeding, regardless of whether B. burgdorferi are present within ticks.
To further examine the importance of tHRF during tick feeding, and its influence on B. burgdorferi transmission, a passive immunization study was performed in naive mice. Groups of 5 mice were administered 200 µl of tHRF antiserum, or control sera (normal rabbit serum or Salp25D antiserum; Salp25D is a tick salivary protein that does not influence tick feeding [23]). One day later, 6 B. burgdorferi-infected ticks were placed on each mouse and tick weights were assessed after 3 days of feeding. Ticks engorging on tHRF antiserum-treated mice weighed significantly less than ticks that fed on control mice (Figure 3A). The spirochete burden in ticks was also substantially lower in I. scapularis that fed on tHRF antiserum-immunized mice (Figure 3B). B. burgdorferi burden was also markedly reduced in tHRF antiserum-immunized mice. The spirochete load in murine skin at day 7 post-infection and in joints and hearts at 3 weeks post-infection was markedly lower in tHRF antiserum-immunized group compared to control serum immunized group (Figure 3, C–E). About 20–27% of the tHRF antiserum-immunized mice (N = 15) were fully protected (based on the absence of a detectable flaB signal in Q-PCR), while 100% of the control animals were infected (N = 30) (Figure 3, C–E).
Uninfected I. scapularis nymphs also fed less efficiently on tHRF antiserum-treated mice (Figure 3F) as seen by decreased engorgement weights compared to ticks fed on control antiserum-treated mice.
We then assessed the ability of the ticks to feed on mice actively immunized with tHRF. Group of 5 mice were immunized with recombinant tHRF, or adjuvant (control). Immunoblots confirmed that mice generated antibodies against tHRF following active immunization (Figure 4A). The tick weights were significantly decreased when B. burgdorferi-infected, or uninfected, nymphs fed on tHRF immunized mice compared to ticks that fed on control mice (Figure 4, B and C). The spirochete load was also markedly reduced in ticks fed on tHRF-immunized mice (Figure 4D) and in murine skin (at day 7 post-infection) (Figure 4E) and in joints and hearts (at 3 weeks post-infection) in the tHRF-immunized group compared to that in control mice (Figure 4, F–G). 20–33% of tHRF immunized mice (N = 15) were PCR negative, while 100% of the mice in the control groups (N = 15) were PCR positive for B. burgdorferi flaB amplicon (Figure 4, E–G).
Our above experiments focused on 72 h post tick attachment- a specific time point at which 30–40% of the ticks from the control groups successfully complete engorgement and drop off the mice, and the remaining ticks nearing engorgement. To address the role of tHRF on 72–96h post tick attachment, all the ticks were allowed to feed to repletion on tHRF antiserum immunized mice or control mice. While, ticks in the control group fed to repletion and detached around 72–84 h of attachment, ticks fed on tHRF-immunized animals fed to repletion around 96 h after attachment. Further, 10–20% of ticks from the tHRF group remained attached to the mouse even after 96 h, (Figure S1A). The engorgement weights of ticks fed on tHRF-antiserum immunized mice were also significantly less than the engorgement weights of ticks fed on control mice (Figure S1B), consistent with our data obtained from 72 h fed ticks (Figure 2–4).
Mammalian histamine release factor binds to basophils and stimulates histamine release [24]. Since tHRF shares substantial homology with mammalian histamine release factors, we postulated that tHRF might also adhere to host basophils and induce histamine secretion. An in vitro binding assay was performed using a rat basophil cell line and recombinant tHRF (tHRF shares 57% similarity with rat HRF at amino acid level, and rat HRF is 100% identical to mouse HRF). Flow cytometry and confocal imaging showed that recombinant tHRF bound to rat basophils (Figure 5, A and B). To examine the influence of tHRF on histamine release, basophils were incubated with recombinant tHRF, or nymphal tick salivary gland extracts (T.SGE). Recombinant tHRF and tick salivary gland extracts stimulated histamine release from basophils (Figure 5C).
Ticks are sensitive to histamine during the early stage of blood feeding, and express histamine binding proteins to counteract this effect. However, tick sensitivity to histamine wanes after 3 days of attachment to a host [20]. Quantitative RT-PCR analysis showed preferential expressions of 3 histamine binding proteins in the salivary glands of I. scapularis nymphs at 24–48 h post tick attachment (Figure 6A–C). However, tHRF was preferentially expressed at 48–72 h post tick attachment (Figure 6 D). Since tHRF induces histamine release, histamine might play an under-appreciated role in the late/rapid phase of tick feeding. To confirm this, histamine or recombinant tHRF was injected into the skin - at the I. scapularis bite site - 60 h after tick-attachment. The tick weights at 72 h were significantly increased when I. scapularis nymphs fed on mice given histamine or recombinant tHRF compared to ticks fed on control mice (Figure 6, E and F). The B. burgdorferi burden was also higher in ticks fed on tHRF-treated mice compared to ticks that fed on control mice (Figure 6G).
The incidence of tick-borne diseases has steadily increased over the past few years, and effective vaccines against most tick-borne pathogens are not currently available [4]. I. scapularis is the major vector of Lyme disease in the USA [2], [15]. Further, I scapularis can serve as efficient vectors of A. phagocytophilum, B. microti, and Powassan virus (a tick-borne encephalitis causing virus). The last decade has seen an increased functional understanding of tick salivary proteins and their critical interactions with the host and pathogen [17], [18], [25]. This information has also offered a new approach to develop effective vaccines against ticks and the pathogens they transmit by simultaneously targeting the pathogen and the tick [26].
The identification of tick proteins potentially involved in pathogen transmission is an important step in the development of effective tick vaccines [1], [18]. The presence of B. burgdorferi within ticks may alter the expression level of selected genes that encode antigens in saliva [19], [27]. One of best characterized genes is salp15 [27]; our recent study suggests that immunization with Salp15 could reduce the transmission of B. burgdorferi from infected ticks to mice, although Salp15 antibodies did not influence the ability of ticks to feed. The action is mainly due to the interaction between Salp15 antibody, Salp15 and Borrelia [26].
We performed a 2DIGE analysis to identify additional tick salivary proteins modulated by spirochetes. We found that tHRF was up-regulated in Borrelia-infected tick salivary glands. HRF is an evolutionally conserved multiple-function protein [28], also a novel cytokine that provokes the release of histamine by both IgE-dependent and IgE-independent mechanisms from mammalian basophils and mast cells [29]. In addition to mammalian HRF, HRF homologs have also been identified in Plasmodium falciparum parasite [30], Dermacentor variabilis [22], [31], [32] and Dermanyssus gallinae [33]. The latter study further indicated that antibodies against HRF increased the mortality of the mites after engorgement, suggesting its potential as a vaccine antigen [33].
Histamine, secreted by basophils in blood and mast cells in tissues, plays a deleterious role during tick feeding. Histamine is a mediator of the itch response and promotes the recruitment of pro-inflammatory cells to the tick bite site – and these immune response prevent tick attachment to the skin of the host [20], [34]. However, the Ixodes tick encodes several histamine binding proteins (HBPs) to counteract the effect of histamine [21], [35]. The elaboration of a histamine release factor in tick salivary glands therefore seemed counterintuitive, since such an activity would be detrimental to tick feeding. Mulenga et al. [22] suggested that ticks might need either HBPs or HRF, depending on its feeding phase. Tick feeding involves a complex series of 9 sequential stages [2]. Host seeking and engagement with the host precede actual tick attachment and establishment of the feeding lesion. The early phase of tick feeding that lasts about 24 h post attachment is sensitive to histamine [20]. We observed increased expression of Histamine Binding Proteins (HBP) in I. scapularis nymphal salivary glands during this early phase of tick feeding and might be critical to counter the effect of histamine (Figure 6). Ticks imbibe very little blood during this early phase of feeding. About 60–72h post tick attachment, which includes the rapid feeding phase, tick sensitivity to histamine significantly declines [20], [34], and ticks fully engorge. During this phase the expressions of HBPs appear to be significantly decreased, and the expression of tHRF increases (Figure 6). We speculate that this reciprocal expression of HBPs and tHRF might help increase the local concentration of histamine at the tick-feeding site during the rapid feeding phase. Increased histamine concentration might modulate the vascular permeability to enhance blood flow into the tick feeding site and facilitate tick engorgement.
B. burgdorferi replicate after the tick begins to take a blood meal, and transmission to the host begins about 36–48 h post tick attachment, a time coincident with active spirochete replication and migration to the salivary glands [14], [36]. Temperature and host blood are critical signals for B. burgdorferi replication and dissemination from the midgut. Since the feeding ability of the tHRF-deficient ticks was significantly impaired, the replication and dissemination of Borrelia inside the ticks was also significantly decreased (Figure 2C). Consequently, the spirochete transmission from tick to mouse was also reduced (Figure 2D), and 20–30% of the mice immunized with tHRF were fully protected from Borrelia infection based on the absence of a detectable flaB signal in Q-PCR (Figure 3 and 4). It is also conceivable that the vasodilatory effect of histamine, might additionally contribute to the efficient dissemination of Borrelia from the original tick-feeding site, where they are deposited, to distal sites.
In summary, for the first time, we demonstrate that the I. scapularis salivary protein tHRF is critical for the tick engorgement, and consequently also facilitates Borrelia transmission to the murine host. We show that B. burgdorferi upregulates the expression of tHRF and immunization with tHRF significantly impairs tick feeding, and decreases B. burgdorferi burden in mice. Importantly, these observations underscore the dynamic nature of the temporal interactions between the vector, the host and the pathogen. While vaccine targeting of tHRF alone might not be sufficient to thwart tick feeding and spirochete transmission, blocking tHRF might offer a viable strategy to complement ongoing efforts to develop vaccines to block tick feeding and transmission of tick-borne pathogens.
An infectious and low passage isolate of B. burgdorferi N40 was used to generate B. burgdorferi-infected ticks. Larval, nymphal, and adult I. scapularis were maintained in our laboratory. Clean larvae were fed either on naïve C3H mice to generate naïve nymphs or on B. burgdorferi-infected C3H mice to generate infected nymphs. Female C3H/HeJ (C3H) mice, 4 to 6 weeks of age, were obtained from the Jackson Laboratory.
Animals were housed and handled under the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The animal experimental protocol was approved by the Yale University's Institutional Animal Care & Use Committee (Protocol Permit Number: 2008-07941). All animal infection experiments were performed in a Bio-safety Level 2 animal facility, according to the regulations of Yale University.
A quantitative analysis of the I. scapularis salivary gland proteome was carried out by 2D fluorescence differential gel electrophoresis (DIGE) at the W.M. Keck Facility at Yale University. Salivary gland extracts from 200 clean and Borrelia-infected I. scapularis nymphs fed for 66–72 h were suspended in a cell lysis buffer (7M urea, 2M thiourea, 4% CHAPS, 25 mM Tris, pH 8.6 at 4°C) and equal amounts of protein (50 µg) from Borrelia-infected and clean salivary gland extracts were then differentially labeled in vitro with Cy3 and Cy5 N-hydroxysuccinimidyl ester dyes as described in the Ettan DIGE manual (GE Healthcare, NJ) and electrophoresis and analysis performed essentially as described earlier [13]. The gel was sequentially scanned using the Typhoon 9410 Imager (GE Healthcare, Piscataway, NJ) and images exported into the DeCyder (GE Healthcare, NJ) software package to assess differentially expressed protein spots. The protein spots that were increased at least 5-fold in Borrelia-infected salivary glands were excised for identification. The gel spots of interest were robotically digested using trypsin prior to analysis on an Applied Biosystems 4800 MALDI-Tof/Tof mass spectrometer. The data was analyzed using the Applied Biosystems GPS Explorer software with Mascot analysis against the NCBI nr database, and a combined peptide mass fingerprint/MS/MS search was done. Spots identified with significant MASCOT scores (P<0.05) of 79 were tabulated.
Fed-nymph salivary gland cDNA was prepared as described [37] and used as template to amplify cDNA of tHRF (GenBank accession no. DQ066335), and SSRB (Signal sequence receptor beta, another tick gene used as a control) (GenBank accession no. DQ066202). The primer sequences are indicated in Table S2 (P11, 12 for tHRF; P13, 14 for SSRB). The resultant amplicons were purified and cloned into the SacII-XhoI sites of the L4440 double T7 Script II vector [37]. dsRNA complementary to the DNA insert was synthesized by in vitro transcription using the Megascript RNAi kit (Ambion, Austin, TX). The dsRNA was purified and quantified spectroscopically. The microinjection of dsRNA was performed as described previously [37]. Briefly, we injected ≈4 nl of dsRNA (1×109 molecules per nl) or buffer alone (MOCK) into the ventral torso of the idiosoma of nymphal I. scapularis. The ticks were allowed to rest for 4∼6 hrs before feeding on mice.
DNA was extracted from mouse tissues and ticks using a DNeasy tissue kit (QIAGEN, Valencia, CA) according to the manufacturer's protocol. The nymphal ticks (unfed or fed for 24, 48, and 72 h) were dissected under the microscope to get the tick salivary gland and midgut. Total RNA was extracted using RNeasy mini spin columns (QIAGEN). RNA was converted into first-strand cDNA using random hexamers and Superscript III reverse transcriptase (Invitrogen, Carlsbad, CA) according to the manufacturer's protocol.
All quantitative PCR (Q-PCR) assays were performed with an iCycler (Bio-Rad Laboratories, Hercules, CA) using gene-specific primers, and IQ SYBR green quantitative PCR system (Bio-Rad) or a Taqman quantitative PCR system (Applied Biosystems, CA) with a program consisting of an initial denaturing step of 3 min at 95°C and 45 amplification cycles consisting of 30 s at 95°C followed by 1 min at 60°C. The gene-specific primers (and probes, for Taqman Q-PCR) used for Q-PCR were indicated in Table S2.
The full open reading frame of tHRF was amplified from the tick salivary cDNA library using gene specific primers P15, 16 (Table S2). The PCR product was subcloned into the pGEX-6P2 vector (Invitrogen, CA) and transfected into E.coli BL21/DE3 strain for protein expression. The recombinant tHRF was purified by GST sephorose 4B and the GST tag was removed by the precision protease on column according to the manufacturer's protocol.
To make recombinant protein using the Drosophila S2 cell system, the full open reading frame of tHRF was subcloned into pMT/BiP/V5-His A vector (Invitrogen) using primers P17, 18 (Table S2) and transfected into Drosophila S2 cells (Invitrogen, CA) in combination with the hygromycin selection vector pCOHYGRO for stable transfection. The stable transformants were selected using 300 µg/ml hygromycin-B for 3–4 weeks. The recombinant tHRF with 6-His tag were induced and purified with Talon affinity column as described previously [38].
To generate polyclonal antisera, tHRF (without the GST tag) produced in E. coli was emulsified in complete Freund's adjuvant and injected into groups of 2–3 rabbits (100 µg/animal). The animals were boosted twice at 3-week intervals with the same dose of antigen in incomplete Freund's adjuvant, and the sera were collected 2 week after the second boost.
A western blot was performed to analyze the protein expression of tHRF in adult tick saliva, nymphal salivary gland extract, midgut extract and whole nymphs. The tick saliva and tissue extract were prepared as described [22]. Protein preparations were separated on a 4–15% gradient poly-acrylaminde gel and transferred on to a PVDF membrane. The membranes were probed with polyclonal anti-tHRF antibody followed by HRP-conjugated anti-Rabbit IgG and detected with enhanced luminol-based detection (ECL) kit (GE bioscience).
Groups of 5 mice were passively immunized with 200 µl of normal rabbit serum, anti-Salp25D antiserum (as controls) or anti-tHRF antiserum, respectively (Salp25D is a tick salivary protein that does not influence tick feeding [23]). 24 h after immunization, 6 B. burgdorferi-infected nymphal ticks or 10 non-infected nymphs were placed on each mouse. After 72h, the ticks were collected and weighed to analyze the feeding efficiency. For the B. burgdorferi-infected tick experiment, the Borrelia burden in ticks as well as in the localized skin specimen at 7 post tick repletion and in the murine heart and joints at 3 weeks post-infection were determined by measuring flaB copies using quantitative PCR.
To address the role of tHRF on 72–96 h post tick attachment, all the ticks were allowed to feed to repletion on tHRF antiserum immunized mice or control mice (normal rabbit serum immunized). After 60 h post tick attachment, the mice were examined every 12 h and the number of tick detached from the mice were recorded. The weights of ticks after repletion were measured as described above.
In the active immunization study, groups of 5 mice were immunized by subcutaneously injecting 10 µg of purified recombinant tHRF suspended in complete Freund's adjuvant, or adjuvant alone (mock control). Mice were boosted with 5 µg of antigen suspended in incomplete Freund's adjuvant every two weeks. Before tick challenge, mice were bled and the anti-tHRF antibody titer was analyzed by western blot. The tick challenge and pathogen burden analysis were performed using the same methods described above.
To test whether tHRF binds to mammalian basophils, an in vitro binding assay was performed as described previously [24]. Briefly, a rat basophilic leukemia cell line RBL-2H3 was purchased from American type culture collection (ATCC, Manassas, VA). Cells were cultured to confluence in a 6-well plate and then incubated with recombinant tHRF (generated from E. coli) or GST in 1% FCS or buffer alone at 4°C for 2hrs. After 3 washes with PBST (PBS+ 0.1% Tween 20), the cells were incubated with purified Alexa 488 labeled anti-tHRF IgG or Alexa 488-anti-GST IgG in 1% FCS plus 1% rat isotype IgG buffer. The IgG labeling was performed using the Alex488 easy labeling kit (Invitrogen, CA) according to manufacture's direction. After 3 washes with PBST, the cells were fixed with 4% PFA and permeablized with 1% Triton X-100, and the nuclei were stained with TOPRO3. The cells were then analyzed by microscopy and Flow Cytometry.
To investigate whether tHRF can induce histamine secretion from basophils, a histamine release assay was performed using the method described [22]. The possible endotoxins from all test protein preparations were eliminated by passing the protein preparations through endotoxin-free columns (PIERCE, Rockford, IL). Varying concentrations of 5, 0.5 and 0.1 µg ml−1 of endotoxin-free HRF (DES) or GST (as negative control) were added to confluent RBL-2H3 cells (in 2 ml media) and incubated at 37°C for 30 min. To determine whether, native tHRF in tick tissue extracts could also induce histamine release, 1.0 µg ml−1 nymphal tick salivary gland extracts were assayed for histamine release. Substance C48/80 (Sigma, St Louis, MO), a calcium ionophore was used at 0.5 µg ml−1 for positive control. A histamine ELISA kit purchased from Research diagnostic Inc. (Flanders, NJ) was used to determine histamine concentrations in culture supernatants.
To investigate the role of histamine and tHRF on late stage of tick feeding, B. burgdorferi infected nymphal ticks were fed on naïve mice for 60h. Then 10mM of histamine or 10 µg of recombinant tHRF were injected into the mouse skin at the tick bite site (usually around the ear). Control mice were given the same amount of PBS or recombinant GST. At 72h post-tick attachment, the tick weights were measured and B. burgdorferi burden in ticks were analyzed by Q-PCR.
Results are expressed as the mean ± the SEM. The significance of the difference between the mean values of the groups was evaluated by Student's t test with StatView software (SAS Institute).
The GenBank accession numbers for the genes related with this study: tHRF/DQ066335; SSRB/DQ066202; Salp25D/AF209911; HBP1/DQ066014; HBP2/DQ066128; HBP3/DQ066002.
|
10.1371/journal.pcbi.1004571 | Mathematical Modelling of Molecular Pathways Enabling Tumour Cell Invasion and Migration | Understanding the etiology of metastasis is very important in clinical perspective, since it is estimated that metastasis accounts for 90% of cancer patient mortality. Metastasis results from a sequence of multiple steps including invasion and migration. The early stages of metastasis are tightly controlled in normal cells and can be drastically affected by malignant mutations; therefore, they might constitute the principal determinants of the overall metastatic rate even if the later stages take long to occur. To elucidate the role of individual mutations or their combinations affecting the metastatic development, a logical model has been constructed that recapitulates published experimental results of known gene perturbations on local invasion and migration processes, and predict the effect of not yet experimentally assessed mutations. The model has been validated using experimental data on transcriptome dynamics following TGF-β-dependent induction of Epithelial to Mesenchymal Transition in lung cancer cell lines. A method to associate gene expression profiles with different stable state solutions of the logical model has been developed for that purpose. In addition, we have systematically predicted alleviating (masking) and synergistic pairwise genetic interactions between the genes composing the model with respect to the probability of acquiring the metastatic phenotype. We focused on several unexpected synergistic genetic interactions leading to theoretically very high metastasis probability. Among them, the synergistic combination of Notch overexpression and p53 deletion shows one of the strongest effects, which is in agreement with a recent published experiment in a mouse model of gut cancer. The mathematical model can recapitulate experimental mutations in both cell line and mouse models. Furthermore, the model predicts new gene perturbations that affect the early steps of metastasis underlying potential intervention points for innovative therapeutic strategies in oncology.
| We provide here a logical model that proposes gene/pathway candidates that could abrogate metastasis. The model explores the mechanisms and interplays between pathways that are involved in the process, identifies the main players in these mechanisms and gives some insight on how the pathways could be altered. The model reproduces phenotypes of published experimental results such as the double mutant Notch+/+/p53-/-. We have also developed two methods: (1) to predict genetic interactions and (2) to match transcriptomics data to the attractors of a logical model and validate the model on cell line experiments.
| Understanding the etiology of metastasis is very important in clinical perspective. Despite the progress with treatment of the primary tumours, the chances of survival for a patient decrease tremendously once metastases have developed [1]. It is estimated that metastasis accounts for 90% of cancer patient mortality [2]. It is now understood that the metastatic process follows a sequence of multiple steps, each being characterised by a small probability of success: 1) infiltration of tumour cells into the adjacent tissue, 2) migration of tumour cells towards vessels, 3) intravasation of tumour cells by breaching through the endothelial monolayer, 4) travelling in the circulatory or in the lymphoid system, 5) extravasation when circulating tumour cells re-enter a distant tissue, and 6) colonisation and proliferation in distant organs [3]. The early stages of invasion and migration are tightly controlled in normal cells and can be drastically affected by malignant mutations. It has been shown indeed that primary and secondary tumours have a common gene signature [4] that mediates the initial stages of metastasis while extravasation and colony formation by a (tumour) cell does not require malignant gene alterations [5], supporting the idea that the later stages of metastasis are affected by the anatomical architecture of the vascular system [6].
Here, we focus on the ability of cancer cells to infiltrate and migrate into the surrounding tissue. The first step towards the formation of secondary tumours is acquiring the ability to migrate. In order to gain motile capacity, epithelial cells need to change their morphology through Epithelial to Mesenchymal Transition (EMT), a process which occurs during development (EMT type 1), wound healing (EMT type 2) and under pathological conditions such as cancer (EMT type 3) [7,8]. EMT type 3 is characterised by both loss of E-cadherin (cdh1) and invasive properties at the invasive front of the tumour [9]. Gene expression of E-cadherin is inhibited by the transcription factors Snai1/2, Zeb1/2 and Twist1, while gene expression of N-cadherin (cdh2) is induced by the same transcription factors [8,10,11]. These transcription factors activate other genes that initiate EMT [11–13] and are induced by several signalling pathways including TGF-β, NF-κB, Wnt and Notch pathways [8,14,15]. On the contrary, the transcription factor p53 has been shown to inhibit EMT via degradation of Snai2 [16]; however, a p53 loss of function (LoF) alone is not sufficient to induce EMT [17]. After the switch of E-cadherin to N-cadherin expression, the cell-cell contacts are weakened [18,19] and cancer cells can pass the basal membrane to infiltrate the surrounding tissue [20]. The process of local invasion can be active since tumour cells can secrete Matrix Metalloproteinases (MMPs) that dissolve the lamina propria [21]. MMPs are also able to digest other components of the extracellular matrix (ECM) and thereby to release growth factors and cytokines that are attached to the ECM [21,22] which in turn activate the tumour cell’s ability to propagate the dissolvement of the lamina propria. Finally, after dissolving the lamina propria and invading the (local) stroma, cancer cells can migrate to distant sites by intravasation and extravasation of the vascular system [2]. To gain insight in the regulation of the metastatic process, several groups have developed mathematical models of various aspects of it [23–29] (S1 Text).
Our aim is to understand the role of gene alterations in the development of metastasis. In many (experimental or in silico) models, EMT is described as a very important step in acquiring metastasis and even considered to be synonymous to appearance of metastasis [30–32]. Due to EMT role in metastasis, much research has been performed to elucidate its regulation. The regulation of EMT is known to be complex and simple intuition is not sufficient to comprehend how genetic alterations (mutations and copy number variations) affect it. Logical modelling can give qualitative insight on how they could affect EMT and subsequently metastasis.
Previously, we have constructed a detailed map of molecular interactions involved in EMT regulation which is freely available at [33], and based on its structural analysis, we hypothesized a simple qualitative mechanism of EMT induction through upregulation of Notch and simultaneously deletion of p53. This prediction has been experimentally validated in a mouse model of colon carcinoma [31].
In the present study, we significantly extend the biological context and provide a mathematical framework for the description of the necessary conditions for having metastasis, going beyond the regulation of EMT only. We take into consideration the gained motility and ability to invade as determinants of the metastatic process. For this purpose we largely extended and re-designed the signalling network including more molecular players and phenotypes, and translated the network into a formal mathematical model, allowing prediction of the metastasis probability and the systematic analysis of mutant properties. Therefore, this work represents a significant progress with respect to the previous results, allowing reconsideration of the qualitative hypothesis suggested before using a formal mathematical modelling approach.
First, we introduce a logical model that recapitulates the molecular biology of the early steps in metastasis. The construction of the influence network and the choice of the logical rules are both based on knowledge derived from scientific articles. The final readouts of the model are the phenotype variables CellCycleArrest, Apoptosis and the aggregated phenotype Metastasis that combines the phenotypes EMT, Invasion and Migration.
We have chosen those final read-outs, as we believe that a metastatic phenotype depends on the occurrence of EMT, invasion and migration. In addition, apoptosis is of importance to the system as during healthy conditions, the cells undergo apoptosis when the cells detach from the basal membrane [34]. Suppressing apoptosis during migration is a required key feature. Our interest in cell cycle arrest is due to results of the mouse model [31] that show decreased proliferation. We try to model this feature in our logical model by looking at the regulation of cell cycle arrest. We did not focus on other phenotypes (or cancer hallmarks) such as proliferation explicitly, senescence, or angiogenesis. These are often considered in cancer studies but they were out of the scope of this work, which focused on depicting early invasion modes and not specifically on how tumour growth is regulated. The model inputs have been selected to represent external signals necessary for the metastasis initiation pathways. The Boolean model that we show here describes a possible regulation of the metastatic potential of a single tumour cell and not of multiple cells or a tissue.
We provide a simplified version of the model where some genes are grouped into modules (or pathways) allowing an analysis based on pathways rather than individual genes. Both versions of the models are validated by reproducing the phenotypic read-outs of published experimental mouse and cell line models.
We then analyse the two models and formulate several types of predictions: at the level of individual genes, e.g. exploring the individual role of each EMT inducer in metastasis; and at the level of pathways, e.g. investigating the functional role of each pathway in triggering metastasis. The logical models can suggest a systematic mechanistic explanation for the majority of experimentally validated mutations on the local invasion and migration processes. Moreover, we were able to establish a link between the solutions of the mathematical model and the gene expression data from cell lines in which EMT was transiently induced. In addition, we have applied this method to the analysis of transcriptomes of tumour biopsies.
Lastly, we investigate how genetic interactions between different gene mutations can affect the probability of reaching a metastatic outcome. Our analysis predicts the effect of single mutations and the genetic interactions between two single mutations with respect to several cellular phenotypes. Our model proves an exceptionally efficient synergetic effect of increased activity of Notch in combination with a decreased activity of p53 on metastasis in accordance with our previous work [31].
The construction of the influence network is based on scientific articles that describe the interactions between nodes of the model. We first selected the main genes or proteins that may contribute to activating EMT, regulating early invasion and triggering metastasis. We then searched for experimentally proven physical interactions that would link all these players, and simplified the detailed mechanisms into an influence network. For example, it has been shown experimentally that AKT protein phosphorylates and stabilises MDM2, which in turn inhibits p53 by forming a complex, leading to protein degradation of p53. We simplified the biochemical reactions by a negative influence from AKT to p53. The influence network is then translated into a mathematical model using Boolean formalism (see below for details). We verified the coherence of the network by comparing the outcome of the perturbed model to the observed phenotypes of mutants found in the literature. The final model is the result of several iterations that led to the accurate description of most of the published mutants related to the genes included in our model. Once the model was able to reproduce most of the published mutant experiments, we simulated mutants and conditions not yet assessed experimentally and predicted the outcome.
From the influence network recapitulating known facts about the processes, we develop a mathematical model based on the Boolean formalism. To do so, we associate to a node of the influence network a corresponding Boolean variable. The variables can take two values: 0 for absent or inactive (OFF), and 1 for present or active (ON). The variables change their value according to a logical rule assigned to them. The state of a variable will thus depend on its logical rule, which is based on logical statements, i.e., on a function of the node regulators linked with logical connectors AND (also denoted as &), OR (|) and NOT (!). A state of the model corresponds to a vector of all variable states. All possible model states are connected into a transition graph where the nodes are model states and the edges correspond to possible transitions from one model state to another. The transition graph is based on asynchronous update, i.e., each variable in the model state is updated one at a time as opposed to all together in the synchronous update strategy. Attractors of the model refer to long-term asymptotic behaviours of the system. Two types of attractors are identified: stable states, when the system has reached a model state whose successor in the transition graph is itself; and cyclic attractors, when trajectories in the transition graph lead to a group of model states that are cycling. In this model, no cyclic attractors were found for the wild type case. However, we do not guarantee the non-existence of cyclic attractors in some of the perturbed cases, as perturbations of the model may create new dynamics.
A logical rule is written for each variable of the model, corresponding to a node in the influence network, in order to define how its status evolves (ON or OFF). In this rule, the variables of the input nodes are linked by logical connectors according to what is known about their combined activities. There are several cases to consider: (1) The simplest logical rule that can be assigned is when a variable depends on the activity of a single input: for instance, the transcription factor Twist induces the transcription of the cdh2 gene (see Table 1); (2) In the case of an input that has a negative effect on the activity of its target, the Boolean operator “NOT” or “!” is used: EMT is, for example, activated by CDH2 but inactivated by CDH1, thus for EMT to be activate, CDH1 should be OFF and CDH2 should be ON. The complete logical rule for the activation of EMT will be EMT = 1 (ON) if CDH2 &! CDH1 (see Table 1); (3) In some cases, a gene can be activated by two independent genes reflecting two different conditions and thus inputs are linked by an OR, e.g., DKK can be activated either by CTNNB1 or by NICD, independently of each other; (4) In some other cases, two activators are linked by an AND connector, e.g., ZEB2 whose activity depends on several inputs including TWIST1 & SNAI2 which are needed simultaneously: it has been observed experimentally that both transcription factors Twist1 and Snai2 are required to induce gene expression of zeb2. All models are available in GINsim format in S1 File.
MaBoSS is a C++ software for simulating continuous/discrete time Markov processes, defined on the state transition graph describing the dynamics of a Boolean network. The rates up (change from OFF to ON) and down (from ON to OFF) for each node are explicitly provided in the MaBoSS configuration file together with logical functions, which allows working with physical time explicitly. All rates are set to be 1 in this model since it is difficult to estimate them from available experimental facts. Probabilities to reach a phenotype are computed as the probability for the phenotype variable to have the value ON, by simulating random walks on the probabilistic state transition graph. The probabilities for the selected outputs are reported for each mutant based on predefined initial conditions (which can be all random). Since a state in the state transition graph can combine the activation of several phenotype variables, not all phenotype probabilities appear to be mutually exclusive. For example, Apoptosis phenotype variable activation is always accompanied by activation of CellCycleArrest phenotype variable (because p53 is a transcription factor of p21, responsible for cell cycle arrest, and the miRNAs, activated by the p53 and its family members, lead to a cell cycle arrest), and activation of the Metastasis phenotype variable is only possible when all three EMT, Invasion and Migration phenotype variables are activated.
With MaBoSS, we can predict an increase or decrease of a phenotype probability when the model variables are altered, which may correspond to the effect of particular mutants or drug treatments. If mutation A increases the Apoptosis probability when compared to the Apoptosis probability in wild type, we conclude that mutant A is advantageous for apoptosis. All models are available in MaBoSS format in S1 File.
The pathway activity (synonymously, module activity) score in a tumour sample is defined as the contribution of this sample into the first principal component computed for all samples on the set of the module target genes, as it was done in [35]. This way, we test target gene sets selected from MSigDB [36] and KEGG [37] databases together with few tens of gene sets assembled by us from external sources. The gene lists for each module is provided in S5 Table. Differential activity score of each module was estimated by t-test between metastatic and non-metastatic groups and significantly active/inactive modules were selected according to p-value <0.05 condition.
We conducted our study on the publicly available data of human colon cancer from TCGA described in [38]. By excluding rectal cancers from the original dataset, the remaining 105 tumour samples were included in our analysis, classified into two groups (‘metastatic’ M1 = 17 tumours and ‘non-metastatic’ M0 = 88 tumours) according to clinical information about metastasis appearance during cancer progression.
We used gene expression data generated from A549 lung adenocarcinoma cell line that was treated with TGF-β1 ligand at eight different time points [39]. In short, gene expression was measured for three replicates at each time point using Affymetrix Human Genome U133 Plus 2.0 Array. For more information about treatment and growth protocols see [40].
We followed the following six steps to link transcriptome data to the stable states of the model (described in detail in S2 Text): (1) We first matched the genes of the model with their HUGO names. For phenotypes such as Apoptosis, Migration or Invasion, the genes coding for CASP9, CDC42, and MMP2 were used as biomarkers, respectively. These readouts were selected as the most representative of the process; they were chosen based on the changes of the expression of a list of candidate genes we explored throughout the experiments. (2) We averaged the genes over the 3 replicates for time point T0 (initiation of experiment with no TGF-β), for T8 (identified as the beginning of EMT), for T24 (EMT in process) and for T72 (last point). (3) Using several methods (binarization algorithms available in [41]), we identified a threshold of expression and binarized the data accordingly. Among our list of genes, only 11 of them have significant expression dynamics along the experiment: cdh1, cdh2, ctnnb1, egfr, mapk1, mmp2, smad3, snai2, tgfb1, vim and zeb1. The other genes were either always ON or always OFF throughout the 72 hours of experiments because the expression is either above or below the threshold we set. (4) We associated a label (phenotype) to the 9 stable states of the logical model based on the activity status of the phenotype variable. (5) The similarity matrix was computed according to the following rule: for each stable state and for each time point, if a gene is ON (= 1) or OFF (= 0) in both the vector of discretized expression data and the vector of the stable state, we set the entry in the similarity matrix to 1, otherwise, it is set to 0. (6) For each time point and each stable state, we then summed up the corresponding similarity matrix row, and set an expression-based phenotypic (EBP) score for each stable state. The highest EBP score for each time point corresponds to the phenotype that is the closest to the studied sample and is representative of the status of the cells.
The non-linear principal manifold was constructed for the distribution of all single and double mutants of the model in the space of computed model phenotype probabilities, using elastic maps method and ViDaExpert software [42–44]. We preferred using a non-linear version of principal component analysis (PCA) for data visualisation in this case, because it is known to better preserve the local neighbourhood distance relations and allows more informative visual estimation of clusters compared to the linear PCA of the same dimension [42]. For data analysis, only those “mixed” phenotypes were selected whose probability expectation over the whole set of single and double mutants was more than 1%. It resulted in a set of 1059 single and double mutants embedded into 6-dimensional space of phenotype probabilities for which the principal manifold was computed.
The results of double mutants were used to quantify the level of epistasis between two model gene defects (resulting either from gain-of-function mutation of a gene or from its knock-out or loss-of-function mutation) with respect to metastatic phenotype. The level of epistasis was quantified using the simplest multiplicative null model applied for the event of not having metastasis: ε = (1-p12)-(1-p1)(1-p2), where p1 and p2 are the probabilities of having metastasis in single mutants, and p12 was the probability of having metastasis in the double mutant. Therefore, negative values of the epistasis score E correspond to synergistic interactions when two gene defects amplify each other’s effect stronger than expected in the multiplicative model. On the contrary, positive values correspond to alleviating effect, when the effect of one gene defect could be masked (sometimes, even reduced to zero) by the second mutation. For genetic network visualisation, we kept the most significant interactions with ε<-0.2 or ε>0.3 values. These thresholds were chosen because at these levels we observed gaps in the distribution of ε values. The complete list of interactions together with p1, p2, p12 and ε values can be found as a Cytoscape 3 session (S2 File).
Mesenchymal cells are characterised by their increased motility, loss of cdh1 (coding for E-cadherin) expression, increased expression of cdh2 (coding for N-cadherin), and presence of vimentin (Vim) [7,10,45]. The EMT program can be initiated by the transcription factors snai1, snai2, zeb1, zeb2 and twist1. They are considered to be the core regulators of EMT as each has been shown to down-regulate cdh1 [46–50]. In turn, the genes coding for these core EMT-regulators are subjected to regulation by other signalling pathways. The TGF-β pathway has been reported to be able to induce EMT [7,51], but other pathways are also involved in EMT including Wnt, Notch and PI3K-AKT pathways [52–56].
Furthermore, microRNAs regulate the Snai and Zeb family members. For example, miR200 targets snai2, zeb1 and zeb2 mRNA [57–59] whereas miR203 targets snai1 and zeb2 mRNA [59], and miR34 targets snai1 mRNA [60]. The transcription of these microRNAs is under the control of p53 [61–64]. The miR200 expression can also be induced by p63 and p73 proteins, while miR34 is only induced by p73 but is down-regulated by p63 [65–67]. The microRNAs can be down-regulated by the EMT-inducers Snai1/2, and Zeb1/2 [59,60,68]. Note that the proteins p63 and p73 have been identified as members of the p53 protein family since their amino acid sequences share high similarity with that of p53 [69]. They are able to bind to the promoters of the majority of the p53-target genes and therefore have overlapping functions in cell cycle arrest and apoptosis [70,71]. The p53-family members are involved in cross-talks with Notch and AKT pathways: p63 protein is inhibited by the Notch pathway, p53 by AKT1 and AKT2 [69,72–76] while p73 is down-regulated by p53 (itself negatively regulated by p73), AKT1, AKT2, and Zeb1 [69,72,77].
The PI3K-AKT pathway has been considered to be important in evading apoptosis and cell cycle arrest by modulating the TRAIL pathway, down-regulating pro-apoptotic genes and phosphorylating p21 [78–80]. More recently, AKT has been assigned additional but important roles in the development of metastasis. AKT1 suppresses apoptosis upon cell detachment (anoikis) of the ECM [34]. The different isoforms of AKT seem to have opposing roles in the regulation of microRNAs: AKT1 inhibits miR34 and activates miR200 while AKT2 inhibits miR200 and activates miR34 [81]. Another opposing role for both AKT isoforms has been found in migration. AKT1 inhibits migration by phosphorylating the protein Palladin; phosphorylated Palladin forms actin bundles that inhibit migration. AKT2 increases the protein Palladin stability and upregulates β1-integrins stimulating migration [82,83] or by inhibiting TSC2 that, in turn, activates RHO [84]. Furthermore, AKT1 inhibits cell cycle arrest while AKT2 activates it [85,86] (all these effects are shown implicitly in Fig 1A).
Extracellular stimuli are also included in the logical model. Growth factors (GF) are soluble ligands that can be excreted locally or from longer distances and are able to activate the PI3K-AKT, and MAPK pathways [87,88]. Another extracellular stimulus might be the extracellular microenvironment (ECMicroenv) with components that are not soluble including the extracellular matrix. The ligands for the TGF-β pathway can be imbedded in the extracellular matrix [89–91] and the ligands for the Notch pathway are transmembrane proteins from adjacent neighbouring cells [92,93].
These mechanisms are depicted in an influence network (Fig 1A). The network is composed of nodes and edges, where some nodes represent biochemical species (proteins, miRNAs, processes, etc.) and others represent phenotypes, and edges represent activating (green) or inhibitory (red) influences of one node onto other node. Each edge is annotated and supported by experimental papers (see S1 Table). Throughout the article, we will use the general term “phenotypes” to refer to “phenotype variables”, which correspond to the four outputs: CellCycleArrest, Apoptosis, Metastasis (depending on EMT, Migration and Invasion), and Homeostatic State (HS) as presented below.
To assess the importance of each pathway on metastasis, apoptosis and cell cycle arrest, we simulated a gain of function or a loss of function, in the reduced model, for each module and for all combinations of inputs. These simulations mean that when an important entity in a pathway is altered, it affects the whole pathway activity. The model shows that mutations leading to either GoF or LoF of each pathway have opposing results in the occurrence of migration and for the occurrence of metastasis (S2 Table). The Notch_pthw is an exception in this: both a GoF and LoF of the Notch pathway can lead to a stable state solution with metastasis ON. This might indicate that Notch (pathway) activity must be in a certain range in order to have a non-pathological effect or that Notch is important for the functioning of some dynamic feedback controls preventing metastasis (so fixing it at a particular value would destroy these feedbacks). In addition, GoF of the Notch_pthw, TGFb_pthw, ERK_pthw, EMT_reg or AKT2 shows their inhibitory role in the apoptotic process as it has been demonstrated before [113–117]. For the p53, TGF-β, EMT_reg and miRNA pathways, mutations leading to activation or inhibition have opposing results in regulating invasion when either the pathway is activated or inhibited. This effect on invasion is a direct result of having an activating or inhibiting role on EMT except for the TGF-β pathway.
The role of TGF-β pathway has been investigated. The activation of TGF-β pathway might be dependent on the micro-environment as its ligands can be found in the extracellular matrix [89–91]. The triple mutant: Notch_pthw GoF, p53 LoF and TGFb_pthw LoF leads to one stable state in which the EMT_reg is ON but no metastasis, migration, invasion or apoptosis are reachable (S2 Table) indicating that activation of TGF-β pathway (e.g., by the peripheral tumour cells more exposed to the micro-environment) is required to have metastasis in the double mutant by activating invasion [118,119].
To identify for each EMT regulator (Snai1, Snai2, Zeb1, Zeb2, Twist1) their specific role in the different cell fates considered in our model, we simulated LoF and GoF mutants and observed that all GoF, except for that of Snai2, led to the loss of apoptosis (S3 Table). Metastasis can be reached for all GoF mutants but other phenotypes can still be reached depending on the combinations of inputs. The single deletions of each EMT regulator show that Zeb2 and Twist1 are required for metastasis. Zeb2 controls migration mainly through VIM but has no direct impact on invasion. Twist1 LoF, on the contrary, modulates negatively the possibility to reach not only the metastatic phenotype but also EMT, migration and invasion. Twist1 controls EMT through Cdh2 that controls migration and EMT. Other factors, such as CTNNB1 (β-catenin) or TGF-β, play a role in triggering the metastatic process by modulating invasion or migration, but our model suggests that the main EMT regulators are Zeb2, Twist1 or Snai2, either as loss of function for Zeb2 and Twist1, or gain of function for Snai2. Note that by definition, Cdh2 is absolutely required for metastasis to occur because of its direct role in controlling EMT and migration. In our model, Cdh1 inhibits EMT (directly) and migration (through CTNNB1 and VIM) but not invasion. Since all three phenotypes are required for metastasis, the process is thus impaired when Cdh1 is over-expressed [121,122].
The probability of achieving the metastatic phenotype for all possible single and double mutants was systematically computed using MaBoSS [123]. Each single and double mutant is characterised by the distribution of phenotype probabilities. A non-linear PCA analysis was performed as described in Methods, which allowed to group together single and double mutants having similar effect on the model phenotypes (Fig 2A). In this plot, one can distinguish six major clusters (a to f) which can be tentatively annotated as “almost wild-type” (no significant changes in the phenotype probabilities compared to the wild-type model), “risk of metastasis” (elevated probability of having metastasis though not equal to 1), “apoptotic” (for these mutants Apopotosis and CellCycleArrest phenotypes are activated), “EMT without migration” (for these mutants, presented as two clusters, the formation of metastases cannot be accomplished because the cells did not acquire ability to migrate), “cell cycle arrest only” (these mutants are found arrested without starting EMT or invasion or apoptotic programs). The direction of increased metastasis probability is shown by dashed line in Fig 2A, which ends at NICD GoF/p53LoF double mutant for which the probability of having metastasis equals to 1, according to the model (whereas single p53 LoF mutation belongs to “almost wild type” and single NICD GoF mutation belongs to “risk of metastasis” clusters respectively).
In this study, we propose a logical model focusing on the specific conditions that could allow the occurrence of metastasis. Our model of the metastatic process represents its early steps: EMT, invasion and migration. A cell acquires the capability to migrate when both EMT and invasion abilities have been acquired. These steps are regulated by several signalling pathways, where genetic aberrations could influence the efficiency of metastatic process. Both the influence network and the assignment of logical rules for each node of this network have been derived from what has been published from experimental works as of today. With this model, we were able to explore known conditions (and predict new ones) required for the occurrence of metastasis. Our influence network describes the regulation of EMT, invasion, migration, cell cycle arrest and apoptosis known from the literature. In this regulatory network, cell cycle arrest and apoptosis are mechanisms or phenotypes that maintain homeostasis of organs [127] or ways to evade metastasis. Cell migration depends on pathways involving AKT, ERK, Vimentin, miR200 and p63 but also on the acquisition of EMT and invasive abilities such as producing MMPs to dissolve the laminae propria enabling migration to distant sites. Cells that have only invasive properties are not able to move as they are still well attached to their surrounding neighbouring cells resulting in absence of cell migration. Only when those two requirements are met and the other pathways allow migration, can metastasis occur.
The role of each EMT regulator, for acquiring invasive properties, has been investigated and the model shows that each individual EMT regulator is sufficient to induce EMT when over-expressed and with the appropriate initial conditions. The model also predicts that a LoF mutation of the EMT regulators does not affect metastasis except for ZEB2 and TWIST1: ZEB2 inhibition leads to abrogation of migration, while a TWIST1 LoF leads to inhibition of EMT, since TWIST1 is the only transcription factor that can induce transcription of cdh2 gene which is required to have EMT. These regulators are interesting targets for therapy since both are more downstream in the metastasis’ cascade knowing that most activating mutations occur relatively more upstream e.g. KRAS and EGFR mutations.
The model has been validated using experimental data by matching the transcriptomic data with stable state solutions of the logical model. The direct comparison between stable states and gene expression of tumour samples shows no conclusive results. This may be due to that only at the front of tumours, cells undergo EMT and this signal is obscured by the bulk of the tumour [30,128]. On the other hand, the model matches well the transcriptomic data from a time course experiment of lung carcinoma cell lines in which EMT was induced by increasing concentration of TGF-β.
Qualitative simulations of the model using MaBoSS confirmed that single mutations are not sufficient to enable metastasis. Therefore, we systematically computed the level of epistasis of each two-gene mutation with respect to reaching the metastatic phenotype. We determined which double mutations are the most efficient for inducing metastasis with NICD GoF/p53 LoF mutations being the most efficient combination of gene knock-out and over-dosage, as this double mutant leads in silico to 100% probability of having metastasis.
In our previous work, this specific double mutation NICD GoF/p53 LoF has been carried out experimentally in a mouse model, by crossing the villin-CreERT2 mice [129] (in this study referred as p53 LoF) and RosaN1ic mice [130] (in this study referred as NICD GoF) with the isogenic C57BL/6 animals to generate the NICD GoF/p53 LoF compound mice. These compound mice develop intestinal tumours with metastatic tumours to distal organs [31]. Our logical model successfully reproduces experimental observations of the compound mouse and proposes mechanisms explaining the metastatic phenotype with high penetrance in mice. In addition, we have investigated the role of TGF-β pathway in metastasis and showed its crucial role in the metastatic phenotype in the double mutant. Suppressing the TGF-β pathway might be an interesting target therapy to control metastasis, however future studies are required.
We also explored the activity of the Wnt pathway in the double mutant. Increased activity of the Wnt pathway due to mutations in the apc and ctnnb1 genes leads to tumourigenesis of many cancers [131–133] and subsequently to metastasis [134,135]. Our mathematical model predicts phenotypes that correspond to adenocarcinomas as a result of linear progression of acquired mutations during sporadic colorectal cancer (CRC) suggested by the “Vogelstein sequence” [136] but no metastasis is reached with the model. Indeed, when we simulate APC LoF, KRAS GoF and p53 LoF (the Vogelstein sequence), the model predicts stable states of cells that are not arrested in the cell cycle, can undergo EMT and can invade (see S4 Table). Thus our logical model supports the hypothesis that the Wnt pathway contributes to tumour initiation [137]. However, there is still a debate if the Wnt pathway is actively involved in metastasis. For example, a negative correlation has been demonstrated between the presence of β-catenin and metastasis in breast cancer [138], in lung cancer [139–141], and in CRC [142–144]. It has been also demonstrated that the canonical Wnt pathway (β-catenin-dependent pathway) is suppressed at the leading edge of the tumour [145] and this might happen without affecting the β-catenin protein levels [146,147]. In the mouse model with Notch GoF /p53 LoF double mutation, in some tumours samples, mutations in apc and ctnnb1 have been found but also tumours without those mutations have been shown to acquire metastasis. Both truncated APC and mutations in β-catenin correspond in our mathematical model to full activation of CTNNB1 and this will induce activation of AKT1. In our model, activation of AKT1 will inhibit migration and therefore inhibit metastasis. Appearance of metastasis in the mouse model with activated Wnt pathway might be putatively explained if one looks at the length of the truncated APC isoform for tumours with apc mutation. The APC mutation found in the Notch GoF /p53 LoF mouse model results in a relatively large truncated APC isoform that might still have inhibitory effect on β-catenin [148]. More details about the APC isoforms and its effect on β-catenin can be found in S3 Text.
Another explanation for having metastasis in tumours with active Wnt pathway might be the involvement of another mutation that affects the akt1 or the akt2 gene. According to our model, the Wnt pathway inhibits metastasis by up-regulation of AKT1. There are tumours in CRC patients (TCGA data from http://cbioportal.org, [31]) that can have an akt2 gene amplification or a homozygous deletion or missense mutation of akt1. AKT2 induces migration while AKT1 inhibits migration thus the ratio AKT1 to AKT2 might be an important determinant for acquiring metastasis in the colon. Indeed studies have shown that AKT2 is predominant in sporadic colon cancer [149] and have a critical role in metastasis in CRC [150].
A Boolean model of EMT induction has been recently published, where the theoretical prediction that the Wnt pathway can be activated upon TGF-β administration was validated experimentally by measuring increased gene expression of the Wnt target gene axin2 in Huh7 and PLC/PRF/5 cell lines [151]. Those cell lines are derived from hepatocellular carcinomas [152,153] and both can harbour known mutations [154] and unconfirmed mutations (http://tinyurl.com/l6mjd8y) that affect the signalling pathways: the Wnt pathway has constitutive activity in the Huh7 cell line [137,155]. An alternative explanation could be that our model is more specific for epithelial cancers as the model depicts many reactions observed in epithelial cells; it has been shown that different types of cancer have different protein (or isoforms) abundance [112,149]. Therefore, our model might be less adequate in predicting the activity for certain nodes for hepatocellular carcinoma and lung adenocarcinoma.
EMT is considered to be the first step and is very often modelled as an equivalent of having metastasis once it is activated. We provide here a logical model that proposes the involvement of three independent processes in order to have metastasis: EMT, invasion and migration. These phenotypes are controlled by an intricate network and only when EMT, invasion and migration do occur, can metastasis happen. The logical model explores the mechanisms and interplays between pathways that are involved in the processes, identifies the main players in these mechanisms and gives insight on how these pathways could be altered in a therapeutic perspective. Note that other mechanisms involving other alterations in the pathways that we model, or in other pathways might also take place, and we do not claim that our approach cover all possibilities of inducing metastasis. Still, our approach provides candidate intervention points for designing innovative anti-metastatic strategies.
|
10.1371/journal.pgen.1007243 | Tissue-specific and mosaic imprinting defects underlie opposite congenital growth disorders in mice | Differential DNA methylation defects of H19/IGF2 are associated with congenital growth disorders characterized by opposite clinical pictures. Due to structural differences between human and mouse, the mechanisms by which mutations of the H19/IGF2 Imprinting Control region (IC1) result in these diseases are undefined. To address this issue, we previously generated a mouse line carrying a humanized IC1 (hIC1) and now replaced the wildtype with a mutant IC1 identified in the overgrowth-associated Beckwith-Wiedemann syndrome. The new humanized mouse line shows pre/post-natal overgrowth on maternal transmission and pre/post-natal undergrowth on paternal transmission of the mutation. The mutant hIC1 acquires abnormal methylation during development causing opposite H19/Igf2 imprinting defects on maternal and paternal chromosomes. Differential and possibly mosaic Igf2 expression and imprinting is associated with asymmetric growth of bilateral organs. Furthermore, tissue-specific imprinting defects result in deficient liver- and placenta-derived Igf2 on paternal transmission and excessive Igf2 in peripheral tissues on maternal transmission, providing a possible molecular explanation for imprinting-associated and phenotypically contrasting growth disorders.
| A humanized mouse line carrying a mutation of the H19/IGF2 imprinting control region demonstrates how tissue-specific and mosaic imprinting alterations result in growth disorders with opposite clinical pictures and asymmetric growth of bilateral organs.
| Imprinted genes show monoallelic and parent-of-origin-dependent expression and play key roles in the control of growth and development. In humans, altered expression of imprinted genes is associated with Imprinting Disorders (IDs) that are characterized by growth, metabolic and behavioural disturbances [1–2]. Most imprinted genes are organized in clusters, in which their parental-specific expression is dependent on Imprinting Control Regions (ICRs). ICRs correspond to 2–5 kb-long sequences with differential DNA methylation on their maternal and paternal alleles. Parental-specific ICR methylation is acquired during gametogenesis and maintained in the zygote and somatic cells throughout development despite extensive demethylation occurring in the embryo before implantation and de novo methylation after implantation [3]. An evolutionary conserved cluster of imprinted genes of about 1 Mbp is located on human chromosome 11p15.5 and mouse distal chromosome 7. The cluster is organized in two functionally independent domains, each with its own ICR. In the telomeric domain, the H19/IGF2 intergenic differentially methylated region (also known as and herein termed Imprinting Center 1, IC1) controls the reciprocal imprinting of the maternally expressed H19 and paternally expressed Insulin like Growth Factor 2 (IGF2) genes [3]. IGF2 is required for normal foetal growth [4]. The liver is the main endocrine source of IGF2 in post-natal life, but autocrine/paracrine activity is found in most embryonic tissues, particularly in placenta, where it is needed for correct allocation of maternal resources to fetal growth [5, 6]. H19 is a long non-coding RNA with inhibitory activity on foetal growth [7]. Both IGF2/Igf2 and H19 are down-regulated after birth in both humans and mice but their deficiencies have long-lasting effects on somatic growth [4, 7–11].
In mouse embryos, H19 and Igf2 are co-expressed in endoderm- and mesoderm-derived tissues, and their expression depends on the same downstream enhancers on the maternal and paternal chromosomes, respectively [12–15]. IC1 is structurally different in humans and mice—human IC1 (hIC1) is ~ 5kb-long and contains seven CCCTC-binding factor (CTCF) target sites (CTS), whereas mouse IC1 (mIC1) is ~ 2kb-long and contains four CTS (Fig 1A). CTCF binding to IC1 is required for the formation of an insulator with enhancer blocking activity in both species [16–17]. Because IC1 is methylated on the paternal allele and CTCF binding is inhibited by DNA methylation, the insulator is formed only on the maternal chromosome where it prevents the activation of IGF2 but allows activation of H19 by the enhancers. The opposite happens on the paternal chromosome, where IGF2 is activated and H19 silenced.
Opposite hIC1 methylation and imprinting defects are associated with the Beckwith-Wiedemann syndrome (BWS, MIM #130650) and the Silver Russell syndrome (SRS, MIM #180860), two IDs characterised by congenital overgrowth and congenital undergrowth, respectively [1]. In particular, hIC1 gain of methylation (GOM) resulting in IGF2 activation and H19 repression on the maternal chromosome is found in 5–10% of BWS cases. Conversely, hIC1 loss of methylation (LOM) leading to IGF2 repression and H19 activation on the paternal chromosome occurs in about 50% of SRS patients. In a number of BWS cases with hIC1 GOM, small deletions and single nucleotide variations within hIC1 co-segregate with the clinical phenotype and abnormal methylation upon maternal transmission but lead to normal phenotype on paternal transmission [11, 18–25]. Different paternally inherited hIC1 deletions have been recently described in a few cases of SRS with IC1 LOM [26]. Although the extent of these deletions is similar to those found in BWS families, the sequence and the CTSs involved are different.
In the last twenty years, several mutations have been introduced into the endogenous mIC1 locus [27–35]. This work has been instrumental to demonstrate the fundamental role that H19/IGF2 imprinting has in the aetiology of congenital overgrowth and undergrowth associated with imprinting defects. However, due to its structural differences with the orthologous mouse locus, the mechanism by which hIC1 mutations affect epigenotype and phenotype in both BWS and SRS is still obscure.
To clarify the role of hIC1 mutations in the origin of imprinting defects and in the pathogenesis of BWS and SRS, we generated a knock-in (KI) mouse line in which the endogenous mIC1 was replaced by the orthologous hIC1 allele carrying a mutation (hIC1Δ2.2) that is associated with BWS on maternal transmission [36]. We compared the H19hIC1Δ2.2 line with wildtype mice and the previously described line carrying a humanized H19 allele with the wildtype human ICR (H19hIC1) [37]. The results demonstrate growth and molecular abnormalities of the mice with maternal and paternal transmission of the mutant KI that resemble those of BWS and SRS, respectively, including asymmetric organ growth. Importantly, tissue-specific and mosaic dysregulation of H19/Igf2 imprinting indicates new pathogenetic mechanisms of congenital growth disorders and lateralized/regional over/under-growth associated with imprinting defects.
In order to study the relationship between genotype, epigenotype and phenotype of IDs in the mouse, we replaced the endogenous mIC1 with a mutant hIC1 allele (H19hIC1Δ2.2) previously found in BWS [19, 36], by homologous recombination in mouse embryonic stem (ES) cells (Fig 1A). Chimeras were obtained and germ line transmission was confirmed by Southern blotting (Fig 1B). The transgenic line was then bred to pCX-NLS- Cre transgenic line to remove the NeoR cassette and its excision was confirmed by PCR (Fig 1C. See also Materials and methods).
We crossed KI mice on a C56BL/6 (B6) background to Balb/C and used polymorphisms present between these two strains to distinguish parental alleles. To compare the behaviour of the mutant with that of the wildtype hIC1 allele (H19hIC1) in the same genetic background, the previously described H19hIC1 line [37] was generated anew by targeting ES cells and breeding the mice using similar procedures to what done with the H19hIC1Δ2.2 line. Subsequent experiments were performed on both humanized KI lines. The wildtype +/+ littermates were also assayed as control.
To investigate the presence of kidney asymmetry, a clinical sign often found in BWS [2], we measured the weight of left and right kidneys of the mice carrying the mutant KI. Maternal transmission was first assessed. A significant difference between the two kidneys (with no bias toward the left or right organ) was found in adult and newborn H19hIC1Δ2.2/+ mice, but not in H19+/+ littermates (Fig 8A and 8B). In contrast, no difference was observed in H19hIC1/+ mice (S7 Fig). hIC1 methylation and H19/Igf2 expression were then assessed in the organs of the neonates carrying the mutant KI. While DNA methylation and H19 expression were comparable (Fig 8C and 8D), significant differences of total and allele-specific expression of Igf2 were found between the heavier and lighter kidneys of H19hIC1Δ2.2/+ (with higher expression in the larger organ) mice (Fig 8E and 8F). Next, we investigated the presence of kidney asymmetry in the mice with paternal transmission of the KI. As for maternal transmission, weight differences between the two kidneys in H19+/hIC1Δ2.2 mice were significantly higher than in their H19+/+ littermates, both at neonatal and adult stages (Fig 8G and 8H). Also, comparable hIC1 methylation and global H19 RNA levels were found in left and right kidneys of all tested animals (Fig 8I and 8J). In contrast, paternal H19 expression was relatively more up-regulated in the lighter kidneys of the H19+/hIC1Δ2.2 mice (Fig 8K). Although not statistically significant (P<0.1), a trend toward stronger Igf2 repression was observed in the smaller with respect to the larger kidney (Fig 8I).
Having demonstrated that H19hIC1Δ2.2 is methylated in somatic cells upon both maternal and paternal transmission, we asked if methylation was already present in germ cells. For this purpose, we measured DNA methylation levels in oocytes and sperm by pyrosequencing. As previously demonstrated, H19hIC1 remains properly hypomethylated in female gametes but methylation is inefficiently established on H19hIC1 in male gametes (Fig 9A and Ref. 37). Similarly, H19hIC1Δ2.2 methylation was close to 0% and comparable with the endogenous mIC1 in oocytes (Fig 9A). mIC2 was also analysed as control and to rule out contamination of somatic cells. The expected methylation value close to 100% was found in H19hIC1Δ2.2/+, H19hIC1/+ and H19+/+ oocytes.
In sperm, methylation levels were relatively low (10–14%) on H19hIC1Δ2.2 as well as H19hIC1 (7–22%), while the endogenous mIC1 was almost 100% methylated (Fig 9A). The endogenous mIC2, analysed as control, was correctly unmethylated in all three lines (H19hIC1Δ2.2/+, H19hIC1/+ and H19+/+). The low methylation status of H19hIC1Δ2.2 was confirmed by bisulphite sequencing, in both oocytes and sperm (Fig 9B).
Overall, these results demonstrate that while DNA methylation of mutant hIC1 is normally absent in oocytes, methylation is not efficiently established on the mutant KI as well as the wildtype KI in male germ cells, indicating that paternal hypomethylation of the H19hIC1Δ2.2 allele is seemingly acquired as early as germline development and persists into embryo development, while maternal hypermethylation of hIC1Δ2.2 alleles does not occur until after fertilization.
Gain and loss of IC1 methylation result in H19/IGF2 imprinting defects that are characteristic of BWS and SRS, respectively. In some patients, genetic mutations have been found associated with DNA methylation abnormalities in cis. However, the mechanism by which the genotype affects the epigenotype and phenotype in these cases is unknown. By employing a KI mouse line, this study demonstrates that a human genetic IC1 mutation reproduces several molecular and phenotypic abnormalities of BWS and SRS. The analysis of this mouse model provides mechanistic insights into the origin of prenatal overgrowth and undergrowth associated with H19/Igf2 imprinting defects, which are useful for understanding the aetiopathogenesis of BWS and SRS.
We have previously demonstrated that maternal transmission of hIC1 can functionally replace mIC1 in the mouse by properly regulating IC1 methylation and H19/Igf2 imprinting [37]. We now observe that mice maternally inheriting the mutant KI allele acquire methylation on hIC1Δ2.2 after fertilization, exhibit H19 repression and biallelic Igf2 activation, and pre- and post-natal overgrowth. Upon paternal transmission, hIC1 lacks methylation resulting in complete Igf2 silencing, H19 activation, severe growth restriction particularly in placenta and embryonic lethality [37]. In contrast, paternal hIC1Δ2.2 is partially methylated and results in a more moderate imprinting defect and pre/post-natal undergrowth which is compatible with life.
Methylation is established only in a minority of male germ cells on both hIC1Δ2.2 and hIC1 and appears unstable indicating evolutionarily divergent mechanisms of imprinting establishment between human and mouse [37]. However, while hIC1 is completely unmethylated, hIC1Δ2.2 shows partial methylation on both paternal and maternal chromosomes in embryo and placenta. Thus, methylation is likely acquired de novo and in mosaic form on hIC1Δ2.2 in somatic cells, during development. The functional difference between hIC1Δ2.2 and hIC1 is likely due to the lack of three CTSs in the mutant allele, which shows lower affinity for CTCF in human cells [36]. The results are consistent with hIC1 having an intrinsic propensity to acquire methylation that is inhibited by CTCF binding [32]. The different behaviour of hIC1Δ2.2 and mIC1, which share a similar number of CTSs (4) and ZFP57 binding sites (6), suggests that the CTSs alone are not sufficient to maintain insulator function and that CTS spacing or other transcription factor binding sites contribute to IC1 function. It is possible that one or more of these elements are reduced or missing and this exposes the mutant hIC1 allele to the action of de novo DNA methyl-transferases in pre- and/or post-implantation embryos.
The mechanisms by which hIC1 methylation is altered in BWS and SRS are unknown. Our mouse model indicates that the maternal hIC1Δ2.2 methylation is acquired post-zygotically in BWS, but does not allow to distinguish if the partial IC1 methylation of SRS is due to a primary germ cell imprint establishment defect, or a post-zygotic maintenance defect, or both. However, the observation that the hIC1 methylation status can drastically change from gametes to somatic cells, suggests that maintenance mechanisms have a critical role in the origin of imprinting defects on both maternal and paternal chromosomes.
Molecular analyses show that, although hIC1Δ2.2 is similarly methylated in neonatal liver, kidney and tongue, allele-specific H19/Igf2 expression is differently altered, suggesting that hIC1 enhancer blocking function is regulated in a tissue-specific manner (Fig 10), consistent with what was previously shown for mIC1 [35]. In particular, the insulator activity of hIC1Δ2.2 appears to be robust in liver and placenta, as demonstrated by weak expression of the Igf2 allele in cis with the KI. In contrast, the relatively high Igf2 expression indicates that the insulator activity of hIC1Δ2.2 is weak in tongue. The kidney shows intermediate and mosaic insulator activity resulting in differential Igf2 expression between left and right organs (see below). Tissue-specific differences in insulator activity may result from different post-translational modifications of CTCF [35]. Importantly, the observation that Igf2 expression is properly regulated in specific tissue contexts in the presence of abnormal IC1 methylation paves the way to new exciting avenues for BWS and SRS therapy. Concerning H19, this gene is significantly down-regulated in all three tissues of H19hIC1Δ2.2/+ mice and up-regulated in liver and tongue of H19+/hIC1Δ2.2 mice, with respect to wildtype littermates, consistent with the partial methylation of hIC1Δ2.2 and the demonstrated repressor activity of methylated IC1 [32]. In liver of H19+/hIC1Δ2.2 neonates, a strong derepression of the imprinted allele was not accompanied by a global (mat + pat) increase of the H19 RNA, possibly because of other physiological perturbations limiting its expression.
Overall, gene expression results are consistent with the hypothesis that the organomegaly of H19hIC1Δ2.2/+ mice is associated with autocrine/paracrine effects of IGF2, while defective hepatic and placental IGF2 expression underlie the growth restriction of H19+/hIC1Δ2.2 mice. Opposite deregulation of the growth inhibitory H19 transcript is likely playing an additional role in the growth abnormalities of the mice with maternal and paternal transmission of the mutant KI.
Although in humans this mutation is associated only with BWS, the hIC1Δ2.2 mice display several features of BWS and SRS, on maternal and paternal transmission of the KI, respectively. Growth abnormalities originate prenatally and persist in adulthood. In particular, H19+/hIC1Δ2.2 mice do not catch-up growth during development, as seen in the majority of children with SRS [38]. In addition, nephromegaly and macroglossia that are observed in H19hIC1Δ2.2/+ mice are also distinctive clinical signs of BWS with IC1 molecular defects [39]. Finally, the tissue-specific differences of hIC1 function we observed in the mouse may explain some of the aetiopathogenetic mechanisms of BWS and SRS. The model we propose predicts that maternal IC1 GOM in BWS causes IGF2 activation primarily in peripheral tissues, such as tongue and kidney resulting in macroglossia and organomegaly, while paternal IC1 LOM in SRS causes IGF2 repression primarily in liver and placenta leading to deficient growth stimulation through defective endocrine secretion and placenta function.
Both H19hIC1Δ2.2/+ and H19+/hIC1Δ2.2 mice show differential growth of the kidneys indicating that asymmetric growth of bilateral organs occurs in these animals, as in many cases of BWS with IC1 defects [40]. Derepression of the maternal Igf2 allele and repression of the paternal Igf2 allele are incomplete in the kidney of P1 H19hIC1Δ2.2/+ and P1 H19+/hIC1Δ2.2 mice, respectively (see Figs 3 and 6), indicating mosaic expression of H19/Igf2 in these tissues. Further analyses showed that, although IC1 methylation and H19 RNA levels were similar, Igf2 expression was higher in the larger than in the smaller kidney due to imprinting defects, indicating that Igf2 is mostly responsible for the asymmetric kidney growth. These results raise the hypothesis that mosaic IGF2 expression may also cause lateralized somatic overgrowth in humans. Our results are consistent with the findings of Ginart et al [41] demonstrating that the incomplete derepression of the paternal H19 allele in mutant mice can result from an epigenetic mosaicism at a single cell level.
Overall, our findings show that a mutant human IC1 sequence can reproduce the opposite growth and molecular phenotypes of BWS and SRS in mouse, when introduced at the orthologous locus. Several mouse lines carrying mutations in the H19/Igf2 locus have been described so far. Some of these mice, including a 1.3 kb deletion of mIC1 (Δ2,3) that results in tissue-specific loss of Igf2 imprinting, show similarities with our H19hIC1Δ2.2 model [28, 34–35]. However, the methylation defects, growth abnormalities and H19/Igf2 dysregulation of H19hIC1Δ2.2 more closely reproduce the phenotypic features and contribute better in understanding the molecular pathogenesis of BWS and SRS. Such humanized mouse models will be useful for more accurately unravelling pathogenetic mechanisms and for developing new therapeutic strategies in these rare congenital growth disorders.
To generate the H19hIC Δ2.2 mouse line, we performed gene targeting by homologous recombination in E14 embryonic stem (ES) cells [42] to target the endogenous mIC1 with a plasmid containing the H19hIC1Δ2.2 allele and neomycin resistance cassette (NeoR; Fig 1A). Briefly, a PciI–MluI restriction fragment of 620 bp spanning the break-point of the 2.2 kb deletion found in a BWS family [18] was extracted from the EΔ2.2 (B5/b1)pL vector [36] and subcloned in the Δ3.8kb-5’ pre-targeting vector containing the wildtype hIC1 region [28]. Sanger sequencing of the fragment was performed and no variant was found in respect to the reference human genome hg19. The subsequent steps to obtain the targeting vector were performed as previously described [37]. To compare H19hIC1 and H19hIC1Δ2.2 in the same strain background and avoid animal transfer from US to Italy, the hIC1 KI line was generated anew by performing gene targeting in E14 ES cells with the original vector used in the previous study [37].
Injection into B6 blastocysts of the H19hIC1-neo and H19hIC1Δ2.2-neo KI ES clones and generation of chimeras were performed by Cogentech Facility S.c.a.r.l. (Milan, Italy). Chimeras were crossed to B6 mice and germline transmission of the KI was confirmed in the agouti progeny by PCR-genotyping using primers flanking the deletion break point (hDMDB5SeqF: 5’–GGTAGTGAGGGATAGAACAC– 3’; hDMDB1RepR: 5’–GAGTGTCCTATTCCCAGATGAC– 3’) (Fig 1B). The NeoR cassette was excised by crossing heterozygous KI with pCX-NLS- Cre transgenic mice [43] on a B6 background. Excision was tested by PCR using primers flanking the NeoR cassette (NeoEXL3: 5’–ACAGAATCGGTTGTGGCTGT– 3’ H19SeqR1: 5’–CCACAGAGTCAGCATCCAC– 3’) (Fig 1C). KI mice without the NeoR cassette were crossed with B6 mice and only the progeny carrying the KI and lacking the Cre gene were selected to expand and assay the KI lines. All animal experimentation was conducted in accordance with the guidelines of the Animal Care and Use Committee of Campania University “Luigi Vanvitelli” (Naples, Italy) and was authorized by the Italian Ministry of Health.
Liver, kidneys and tongue were collected from mice at birth and at 14 weeks of age. Placenta and whole body excluding the head were recovered from conceptuses of 15.5 days post coitum (E15.5). Genomic DNA was isolated from tissues following the standard protocol of proteinase K digestion and phenol-chloroform extraction. Total RNA was extracted using TRI Reagent (Sigma-Aldrich Italia, Milan) and following the manufacturer’s protocol. Concentrations of nucleic acids were determined with Nanodrop spectrophotometer. Sperm was isolated from adult mice and DNA was extracted as described previously [44]. Unfertilized oocytes were collected from 4–5 superovulated females of 8 weeks and resuspended in 0.03% SDS, 10 mg glycogen, 10 mg proteinase K and 1 x PBS to a final volume of 20μl. Suspension was incubated at 37°C for 90 min followed by 15 min at 95°C before sodium bisulphite treatment for DNA methylation analysis (see below).
About 1 μg of total RNA was treated with RNase-free DNase, and first-strand cDNA was synthesized using the QuantiTech Reverse Transcription Kit (Qiagen Italy, Milan), according to the manufacturer’s protocol. Total expression of H19 and Igf2 was measured by SYBR Green quantitative real-time RT-PCR (Applied Biosystems Italy, Milan). Reactions were set up in triplicate and run on ABI PRISM 7500 using the default cycling conditions. Relative expression was determined using the ΔΔCT method, and the gene expression values were normalized to the expression of the Gapdh, Arpp0 and beta-actin reference genes. Primer sequences are available on request. Allele-specific expression analysis was performed by typing for the polymorphisms present in the F1 progeny between the C57BL/6 (B6) and Balb/C mouse strains. The MspI Restriction Fragment Length Polymorphism (RFLP) of H19 (GRCm38/mm10 chr 7:142,577,609; sequenced region: chr7:142,577,530–142,577,732) and the (CA)n repeat of Igf2 (GRCm38/mm10 chr 7:142,652,936–142,652,973; sequenced region: chr7:142,652,821–142,653,091) were analysed as described by Pedone et al, [45]. The forward primer of Igf2 was labelled with FAM and the PCR products were run on the ABI 3130XL fluorescent capillary system (Applied Biosystems Italy, Milan).
The methylation status of cytosines in gDNA was determined by bisulphite treatment followed by pyrosequencing or cloning and sequencing. About 1 μg of genomic DNA was treated by sodium bisulphite using the Epitech Kit (Qiagen Italy, Milan), according to the manufacturer’s instructions. For the pyrosequencing, converted DNA was amplified with primers of which the reverse primer was biotinylated. The PCR products were run on the PyroMark Q24 platform, using PyromMark Gold Q96 Reagent [Qiagen Italy, Milan]. For bisulphite cloning and sequencing, converted DNA was amplified, the PCR products were cloned in Topo pCR2.1 vector (Topo-TA cloning kit, Termo Fisher Scientific Italy, Milan) and the clones were sequenced by Sanger method at Microtech Sequencing Core (Naples, Italy). Primers and PCR conditions are reported in S1 Table.
Unless otherwise indicated, data are expressed as the mean ± standard error of the mean (SEM). The significance of the difference between two groups (KI and wildtype) was determined with a two-tailed Student’s t-test with two-sample unequal variance. The number of samples, animals, biological or technical replicates are indicated in the respective figure legends. Differences with P-values ≤ 0.05 were considered significant.
|
10.1371/journal.pbio.0060318 | PKMζ Maintains Spatial, Instrumental, and Classically Conditioned Long-Term Memories | How long-term memories are stored is a fundamental question in neuroscience. The first molecular mechanism for long-term memory storage in the brain was recently identified as the persistent action of protein kinase Mzeta (PKMζ), an autonomously active atypical protein kinase C (PKC) isoform critical for the maintenance of long-term potentiation (LTP). PKMζ maintains aversively conditioned associations, but what general form of information the kinase encodes in the brain is unknown. We first confirmed the specificity of the action of zeta inhibitory peptide (ZIP) by disrupting long-term memory for active place avoidance with chelerythrine, a second inhibitor of PKMζ activity. We then examined, using ZIP, the effect of PKMζ inhibition in dorsal hippocampus (DH) and basolateral amygdala (BLA) on retention of 1-d-old information acquired in the radial arm maze, water maze, inhibitory avoidance, and contextual and cued fear conditioning paradigms. In the DH, PKMζ inhibition selectively disrupted retention of information for spatial reference, but not spatial working memory in the radial arm maze, and precise, but not coarse spatial information in the water maze. Thus retention of accurate spatial, but not procedural and contextual information required PKMζ activity. Similarly, PKMζ inhibition in the hippocampus did not affect contextual information after fear conditioning. In contrast, PKMζ inhibition in the BLA impaired retention of classical conditioned stimulus–unconditioned stimulus (CS-US) associations for both contextual and auditory fear, as well as instrumentally conditioned inhibitory avoidance. PKMζ inhibition had no effect on postshock freezing, indicating fear expression mediated by the BLA remained intact. Thus, persistent PKMζ activity is a general mechanism for both appetitively and aversively motivated retention of specific, accurate learned information, but is not required for processing contextual, imprecise, or procedural information.
| How long-term memories are stored as physical traces in the brain is a fundamental question in neuroscience. Recently, we discovered the first molecular mechanism of long-term memory storage. We showed that unpleasant memories are stored by the persistent action of an enzyme, a form of protein kinase C, termed PKMζ, because these memories can be rapidly erased by injecting a PKMζ inhibitor into the brain. But are all forms of memory and information in the brain stored by PKMζ? Here, we first confirmed with a second inhibitor of PKMζ that unpleasant long-term memories in the hippocampus, a region of the brain critical for storing spatial information, are rapidly erased. We then examined other memories stored in the hippocampus and the basolateral amygdala, another region critical for emotional memories. We tested memories for specific places, both unpleasant and rewarding, memories for general background information, associations between a sound and a fearful event, like that studied by Pavlov, and memories for performing a specific action. We found that PKMζ stores specific associations, both unpleasant and rewarding, for places, events, and actions, and is thus a general mechanism for memory storage in the brain.
| Although the molecular mechanisms of initial memory consolidation have been extensively studied, little is known about the mechanism of persistent memory storage [1]. Recently, however, the persistent phosphorylation by the autonomously active protein kinase C (PKC) isoform, protein kinase Mzeta (PKMζ), has been shown to be critical for the maintenance of aversive long-term memories, specifically, place avoidance in the hippocampus [2] and conditioned taste aversion in the neocortex [3].
PKMζ was initially identified as a persistently active kinase that is both necessary and sufficient for the maintenance of long-term potentiation (LTP) [4,5]. PKMζ is a persistently active kinase because of its unique structure [4]. Most PKC isoforms consist of an N-terminal regulatory domain, which contains second messenger-binding sites and an autoinhibitory pseudosubstrate sequence, and a C-terminal catalytic domain [6]. Under basal conditions, the pseudosubstrate interacts with the catalytic domain and maintains the enzyme in an autoinhibited resting state. Second messengers, such as diacylglycerol or Ca2+, can then activate full-length PKCs by binding to the regulatory domain, causing a conformational change that releases the autoinhibition. PKMζ, in contrast, is an independent PKCζ catalytic domain, which, lacking autoinhibition from a regulatory domain, is autonomously active. In the brain, PKMζ is generated by an internal promoter within the PKCζ gene, which produces a PKMζ mRNA that encodes only the ζ catalytic domain [7]. During LTP, tetanic stimulation induces de novo synthesis of PKMζ, increasing the amount of the persistently active kinase [7,8]. The persistent PKMζ activity is critical for maintaining enhanced synaptic transmission, because inhibition by the cell-permeable zeta inhibitory peptide (ZIP), which mimics the pseudosubstrate of the missing PKCζ regulatory domain, reverses synaptic potentiation in the hippocampus when applied up to 1 d after LTP induction [2]. The effect is specific to potentiated synapses because the same dose of ZIP does not affect baseline synaptic transmission [2].
In parallel studies, this dose of ZIP eliminated the retention of long-term memory, but not short-term memory, for place avoidance in the hippocampus [2] and disrupted the storage, but not acquisition, of conditioned taste aversion in the insular neocortex [3]. Despite affecting retention of these aversively conditioned long-term memories, PKMζ inhibition had no effect on taste familiarity, although it is not known whether memory supporting taste familiarity is stored in the insula [3]. Thus, whether the persistent activity of PKMζ maintains all information in a brain region is a critical open question.
To address this issue, we studied a battery of conditioned behaviors that require either the dorsal hippocampus (DH) or the basolateral amygdala (BLA) for memory retention, as previously determined by posttraining ablation studies. To compare results across a range of types of long-term memory induced by widely different behavioral paradigms, 1 d after the completion of training to acquire long-term memory, we injected the standard dose of ZIP that locally reverses 1-d-old in vivo LTP without affecting baseline synaptic transmission [2].
Previous studies in the DH of place avoidance memory had shown that the PKMζ inhibitor ZIP, but not the conventional and novel PKC isoform inhibitor staurosporine, which does not effectively inhibit PKMζ, caused a selective loss of long-term memory retention [2]. The only other potent PKMζ inhibitor to have been characterized is chelerythrine, a benzophenanthridine alkaloid rather than a pseudosubstrate peptide like ZIP, and a general inhibitor of the catalytic domain of PKCs that strongly inhibits PKM forms [5]. We therefore examined the effect of chelerythrine on long-term memory retention of active place avoidance. On the first trial, rats entered the shock zone within seconds, but with training, the animals learned to avoid the shock zone for several minutes (Figure 1). Twenty-two hours later, chelerythrine or vehicle was injected into both hippocampi, and retention of the avoidance memory was tested 2 h after the injection, as previously described for ZIP [2]. Rats showed excellent retention following injections of the control solution, avoiding the shock zone for several minutes (Figure 1). However, after injections of chelerythrine, the rats once again rapidly entered the shock zone within seconds (Figure 1).
Chelerythrine did not prevent acquiring or expressing short-term memory for active place avoidance. Immediately after the 24-h retention test, the shock was turned on for 10 min and then turned off [2]. Turning on the shock improved avoidance in the vehicle-injected rats even further. Thus, during the long-term memory testing, the animals spent only 7.9 ± 0.7% of their time in the shock zone, significantly less than 16.7%, the level of chance (t3 = 12.0; p = 0.001), and then after the single training session, they spent even less time in the shock zone (0.44 ± 0.44%; t3 = 37; p < 0.0001 compared to chance). During the long-term memory testing, the animals injected with chelerythrine showed no long-term memory retention, as expected, spending time in the shock zone at the level of chance (16.4 ± 1.1%, t4 = 0.3; p = 0.7), but then these animals avoided the shock zone after the single training session, spending only 9.3 ± 1.1% of the session in the shock zone after the shock was turned off (t4 = 6.5; p = 0.003 compared to chance). Thus, the action of chelerythrine on both long-term and short-term memory retention was indistinguishable from the action of ZIP [2]. In subsequent experiments, we used ZIP, the more specific of the two drugs, and controlled for nonspecific effects of the peptide in each of the conditioned behaviors with the scrambled, inactive version of ZIP (scr-ZIP) [2].
We then examined whether appetitively conditioned spatial information was maintained by the same mechanism as aversively conditioned spatial information in the DH by examining the effect of ZIP on conditioned behavior in the eight-arm radial maze. Rats learned the task, and after six trial blocks (3 d), performance was asymptotic and optimal for an additional six trial blocks (days 4–6; Figure 2A–2C). On day 7, bilateral DH injections of the control compounds, saline or scr-ZIP, did not alter performance during testing that began 2 h later. In contrast, injections of ZIP caused the number of correct choices to drop to the level of naive rats (Figure 2A–2C). The deficit could not be attributed to an increase in working memory errors (Figure 2B; p = 0.33), but was due to a specific increase in spatial reference memory errors (Figure 2C; p = 0.001).
To further characterize these reference memory errors, we examined the errors made by the ZIP-injected rats when there was only one correct choice remaining, i.e., after three of the four baited arms had been chosen. Of 21 such errors, ten were to an arm adjacent to the correct arm, and six, three, and two were to arms that were two, three, or four arms away from the correct arm, respectively. After accounting for the fact that there was only a single arm four arms away from the correct arm, and two arms for the other categories, there was a significant effect of where the errors were distributed (F3,28 = 3.1; p = 0.04). Thus, ZIP impaired spatial reference memory, possibly by impairing spatial accuracy, but because the rats foraged appropriately on the maze and continued to use the win-shift strategy that requires working memory, the memory for the general contextual and procedural aspects of the task appeared unaffected.
In the active place avoidance task, ZIP had a persistent effect on long-term memory retention [2]. To examine whether impairment in the radial arm maze was also persistent, 2 wk after the ZIP injection, the animals were reexamined with a single training trial. The rats that had been injected with saline 2 wk earlier showed excellent memory retention (96.7 ± 6.5% correct choices), whereas the rats that had been injected with ZIP made fewer correct choices (59.1 ± 6.7%), indicating they were still impaired (t11 = 8.6; p < 0.0001).
We next examined whether spatial reference memory in the water maze is also maintained by PKMζ in the hippocampus. Rats learned the location of the escape platform during 5 d of training (Figure 2D). After injections with the control solutions, saline or scr-ZIP, 2 h before the probe test on day 6, the rats repeatedly crossed the platform location and concentrated their search in the correct quadrant (Figure 2E–2G). In contrast, ZIP injections diminished the accuracy of searching. Although the ZIP-injected rats concentrated their swim time in the correct quadrant of the pool (Figure 2E and 2G; p = 0.15), they crossed the platform location fewer times than the rats injected with saline or the control compound (Figure 2F and 2G; p = 0.03). These differences were not due to changes in the total distance the rats swam (saline [sal] = 15.2 ± 0.4 m; scr-ZIP = 16.3 ± 0.5 m; and ZIP = 16.3 ± 0.5 m; F2,21 = 1.56; p = 0.2) or the average swim speed calculated every 2 s (sal = 25.3 ± 0.7 cm/s; scr-ZIP = 27.2 ± 0.8 cm/s; and ZIP = 27.2 ± 0.9 cm/s; F2,21 = 1.56; p = 0.2). The amnesia induced by ZIP persisted, because on day 29, the rats that had been injected with scr-ZIP 23 d earlier, took 18.4 ± 3.5 s to find the platform on the first trial, in contrast to the rats that had previously received ZIP injections, which took significantly longer, 43.4 ± 9.4 s (n's = 4; t7 = 2.6; one-tailed p = 0.04). These results indicate that PKMζ in DH maintains the precise spatial information that is needed for accurate localization, but not the global spatial information or the contextual information that is necessary for the spatial search strategy.
We then tested whether persistent PKMζ activity in the hippocampus maintains conditioned-fear responses to context. ZIP injections into the DH 22 h after context/tone-shock pairing failed to alter contextual freezing tested 1 d later (Figure 3A; p = 0.86). Although the DH may have a role in some tone-fear paradigms [9,10], its role in long-term storage of tone-fear associations is uncertain [10–12]. ZIP in the DH also did not impair tone-associated fear tested in a novel chamber 3 d after the infusions (unpublished data). In additional experiments, decreasing the number of shocks from five to one, eliminating the tone during conditioning with a single shock, and bilaterally injecting into both dorsal and ventral hippocampi 1 d after conditioning failed to reveal an effect of ZIP on contextual fear (sal, 86 ± 11% freezing; ZIP, 83 ± 9% freezing; F1,14 = 0.32; p = 0.9).
Together, these results in the DH indicate that PKMζ selectively maintains precise learned associations for locations, but not associations to imprecise spatial or background contextual stimuli. We therefore tested whether PKMζ maintains the fear-mediated CS-US associations thought to be stored in the BLA. Rats received a single tone-shock pairing trial and 22 h later were injected with saline, scr-ZIP, or ZIP. The saline- and scr-ZIP–treated rats expressed normal conditioned fear 2 h and 24 h after the injection, but the ZIP-injected rats showed impaired conditioned freezing at both retention delays. The results of the two retention delays were indistinguishable and therefore analyzed together. The effect of ZIP was different from that of the control solutions (Figure 3B; p = 0.01). In separate animals (n = 8 for each group), the ZIP injections into the BLA 24 h after context/tone-shock pairing attenuated both contextual freezing tested 1 d later and tone-associated fear tested in a novel chamber 3 h after the context test (freezing to context: sal, 82 ± 4.5%; ZIP, 48 ± 6.7%; F1,14 =18.2, p < 0.001; freezing to tone: sal, 61 ± 12%; ZIP, 8 ± 3%; F1,14 = 20.3, p < 0.001). In parallel locomotion experiments, ZIP infusion did not induce hyperactivity measured by beam crossings during 1 h (scr-ZIP, average = 146.4 ± 15.7, n = 4; ZIP, average = 197.2 ± 35.8, n = 5; F1,7 = 1.7, p = 0.23). Thus, in contrast to the DH, ZIP injection into the BLA impairs retention of conditioned-fear behavior.
Ablation of the BLA as well as the adjacent central nucleus of the amygdala attenuates freezing to the shock itself [13–15], which may confound the interpretation of whether information is stored in the BLA, or instead, whether the BLA is required for the expression of a fear association that is stored elsewhere. We therefore tested whether injecting ZIP into the BLA affected the expression of fear immediately after a shock. ZIP or saline was infused into the BLA 5 min or 2 h prior to testing immediate postshock freezing. ZIP did not affect immediate postshock freezing at either time point (Figure 3C; data from both time points combined, p = 0.46). Because ZIP in the BLA did not alter the ability to express fear, but attenuated conditioned fear, we conclude that persistent PKMζ activity in the BLA maintains the information that is required for fear associations, but not the function of the BLA in expressing fear.
We then tested whether other forms of memory that depend on the BLA also require persistent PKMζ activity for maintenance. Injecting ZIP into the BLA 22 h after inhibitory avoidance conditioning impaired retention of the conditioned response that was tested 2 h later (Figure 4; p < 0.01). Two weeks later, the impairment persisted (latency to enter the dark compartment: n's = 4; scr-ZIP = 297 ± 104 s; ZIP = 79 ± 50 s; t7 = 2.5; one-tailed p < 0.05). Thus, long-term memories for both classically conditioned fear and instrumentally conditioned inhibitory avoidance depend upon persistent PKMζ activity in the BLA.
We find that site-specific inhibition of PKMζ impairs the retention of specific, accurate associations in multiple tasks in different brain regions, regardless of positive or negative reinforcement, and thus the persistence of PKMζ activity is a general molecular mechanism for the maintenance of memory. This mechanism is specific for sustaining accurate learned associations because inhibition of the kinase did not affect the expression of imprecise, contextual, and procedural information that depends upon the functioning of the brain regions in which the associations were stored. PKMζ inhibition thus contrasts with permanent or temporary lesions, which affect both types of information. Because ZIP specifically reverses information stored in synapses by late-LTP maintenance and does not affect baseline synaptic transmission in the hippocampus [2,5,16,17], these results suggest that the physiological function of late-LTP–like plasticity may also be selectively important for storing specific accurate information. Although future work will be required to examine whether PKMζ maintains late LTP in the BLA as well the hippocampus, in all the tasks, ZIP produced a persistent loss of long-term memory, consistent with previous results of an effect on memory storage [2,3], although the possibility of an as yet undiscovered role in information retrieval cannot be ruled out.
In our study, we first replicated the main finding of Pastalkova et al. [2], using a second inhibitor of PKMζ, chelerythrine. We found that the drug, the only other potent inhibitor of PKMζ activity known, produces the identical rapid impairment of long-term memory retention but sparing of short-term memory. Although chelerythrine affects all PKC isoforms at high doses [5], (1) ZIP does not affect conventional and novel PKCs [5], and inhibition of these other PKCs does not affect memory retention [2]; and (2) the common target of the two inhibitors is PKMζ, and both agents cause the same pattern of amnesia. The possibility that ZIP might inhibit another as yet unidentified protein kinase or some other process cannot be excluded [18]; however, such an effect would require the specific sequence of amino acids in ZIP that inhibits PKMζ activity, because the effect of the scrambled version of the peptide was indistinguishable from saline in all of the behaviors examined in this study.
In the DH, persistent PKMζ activity was shown to specifically maintain memories for precise spatial locations, but not imprecise spatial, contextual, or procedural information. Thus, the same injections of ZIP resulted in the loss of information supporting accurate spatial reference memory in the eight-arm radial maze task (Figure 2C), but no effect on the ability to use working memory to do the win-shift foraging procedure (Figure 2B). Likewise, the effect of ZIP injection in the water maze task was the elimination of information supporting accurate spatial navigation (Figure 2F), whereas information needed for the general place response to search in the platform quadrant of the pool was spared (Figure 2E and 2G). Thus, the general recognition of the contextual and the procedural aspects of these tasks appeared unaffected by ZIP. Indeed, the equivalent ZIP injection in the DH did not impair context-associated fear at all (Figure 3A), whereas injection of the inhibitor in the BLA disrupted the conditioned response. It is possible that regions of the hippocampus not affected by the ZIP injections contributed to the sparing of contextual aspects of memory [19]; however, we consider this unlikely because simultaneous ZIP injections in dorsal and ventral hippocampi also did not affect contextual fear. Thus, long-term information stored within the DH by PKMζ activity appears to be required for fine, accurate spatial reckoning or precise discrimination between related memories of location, as between the arms in the radial maze. These findings are consistent with the complete loss of the ability to perform active place avoidance on a rotating disk following PKMζ inhibition (Figure 1 and [2]). In this task, the ability to discriminate between specific memories of shock locations with respect to the room and shock locations on the rotating disk is essential for avoiding the stationary shock zone [20,21].
The ability of ZIP to disrupt specific types of stored information in the DH while leaving other information intact may be due to its ability to specifically reverse late-LTP maintenance in the hippocampus, but not other forms of neural plasticity that can store information [22]. For example, PKMζ activity in the DH is not required for working memory in a radial arm maze, which appears to be mediated by transient early LTP [23,24] that is not maintained by PKMζ [16,17]. Long-term memories encoding coarser-grained spatial positions or context may be mediated by forms of long-term synaptic plasticity that might not be maintained by PKMζ, such as long-term depression [17,22] or perhaps, changes in neural excitability [25–27].
Alternatively, memories unaffected by ZIP in the DH may be stored elsewhere but require an intact hippocampus for processing or retrieving the information rather than storing it. Indeed, the specific effect of ZIP on fine, but not coarse-grained spatial information in the DH is consistent with the recent discovery that grid cells of the medial entorhinal cortex, which is the main input to the hippocampus, provide sufficient contextual and place information for spatial navigation based on distal landmarks [28–30]. Thus extrahippocampal regions appear to encode sufficient spatial information for the rat to recognize its environment and general location. Prior work with ablation or inactivation of the hippocampus would have interrupted the projection loops of this spatial information from the superficial layers of the entorhinal cortex through the subfields of the hippocampus and back out to the deep layers of the entorhinal cortex. In contrast, transmission of information through these circuits may not have been disrupted by PKMζ inhibition because ZIP has no effect on baseline synaptic transmission in hippocampal slices or in vivo [2,5,16,17].
In the BLA, both specific instrumentally conditioned associations for inhibitory avoidance and classically conditioned associations for fear were impaired by PKMζ inhibition. The impairment was specific to long-term memory because the PKMζ inhibition did not induce hyperlocomotion or disrupt the expression of fear to a recent shock, as observed, for example, with ablations of the BLA [13–15, 31]. Thus, specific fear memories are maintained in the BLA by persistent PKMζ activity, which may be distinct from the BLA's role in modulating aversively motivated information stored elsewhere [32,33].
Lastly, the characterization of the forms of information stored in the brain by PKMζ may have clinical implications for disorders thought to be mediated by excessive memory retention or LTP-like plasticity, such as posttraumatic stress, phobias, and addictions. Previous studies have indicated that PKMζ inhibition by ZIP in the hippocampus and neocortex erases long-term memories encoded even weeks prior to injection [2,3]; therefore, further study will ultimately be required to identify a method by which PKMζ inhibition might target specific memories, perhaps by examining the role of PKMζ during reconsolidation after memory reactivation [34,35]. As an enabling first step, however, our current findings suggest that PKMζ inhibition disrupts the retention of specific, precise information stored in a brain region, but spares the region's processing functions such as relaying information or performing computations on information stored elsewhere. Thus, not all memories and functions previously ascribed to a brain region will be lost by site-specific PKMζ inhibition, but discrete pathophysiological associations induced by both fearful and rewarding experiences may.
Adult male rats were implanted with cannulae for intracranial infusion in the DH or BLA (Figure 5). The rats were first trained to acquire a robust long-term memory in a particular behavioral paradigm. A day later, we tested whether bilateral inhibition of PKMζ impaired memory retention, by injecting a PKMζ inhibitor or a control solution [2]. All experimental methods have been published and are briefly described here. The procedures were approved by the local institutional animal care and use committee in compliance with the National Institutes of Health and federal guidelines.
Each rat was deeply anesthetized (Nembutal >50 mg/kg intraperitoneally [i.p.] or ketamine/xylazine mixture 100/10 mg/kg i.p.) and mounted in a stereotaxic frame to drill bilateral holes in the skull for a pair of 22-ga infusion guide cannulae. The guide tips were at least 1 mm above the infusion target. Bone screws and dental cement secured the cannulae to the skull. Each rat was handled for at least 5 d to habituate it to the experimenter prior to training. Training began at least a week after surgery. The rats were habituated to the infusion procedure by mock saline infusions at least a day before the experimental solutions were infused. After the task was well learned, each rat received a bilateral intracranial infusion (1 μl/side) of one of three experimental solutions: the myristoylated peptide PKMζ inhibitor ZIP (10 nmol/μl saline; QCB and University Wisconsin Biotech peptide synthesis facility), the control myristoylated peptide, scr-ZIP, comprising a scrambled sequence of the same amino acids as ZIP (10 nmol/μl saline; QCB), or saline [2]. Memory retention was tested 2 h after the infusion, unless stated otherwise. The infusion target in the DH was 3.8 mm posterior, 2.5 mm lateral, and 3.5 mm ventral to bregma. In the experiment to examine the effect of simultaneous PKMζ inhibition of both the dorsal and the ventral hippocampi, the dorsal coordinates were 2.5 mm posterior, 2.4 mm lateral, and 3.0 mm ventral to bregma, and the ventral coordinates were 5.6 mm posterior, 5.0 mm lateral, and 6.0 mm ventral to bregma. The coordinates for bilateral infusion into the BLA were 2.8 mm posterior, 4.8 mm lateral, and 5.8 mm ventral to bregma. After behavioral testing, the rats were sacrificed by anesthetic overdose, then transcardially perfused with saline followed by 10% formalin. The brains were removed, postfixed in 10% sucrose-10% formalin solution, sectioned, then stained with cresyl violet, and examined by light microscopy to estimate the injection site (Figure 5).
Active place avoidance memory is a rapidly acquired form of spatial memory, the long-term retention of which is disrupted by bilateral hippocampal inactivation [36]. The first demonstrations that PKMζ maintains long-term memory used ZIP in DH to eliminate 1-d-old place avoidance memories [2]. We used the same training protocol to test whether chelerythrine, another PKMζ inhibitor, also eliminates long-term place avoidance memory.
The place avoidance procedures have been described in detail [2,20]. Briefly, the rat is placed on an 82-cm–diameter metal disk that is elevated 78 cm from the floor and rotates at 1 rpm within a room with numerous visual landmarks off of the disk. Prior to training, the rat is implanted with a subcutaneous shock electrode between the shoulders, through which a constant current (0.3 mA, 60 Hz, 500 ms) electrical foot shock is delivered whenever the rat enters an unmarked shock zone. The impedance between the shock electrode and the skin is approximately 1,000 times less than the impedance between the rat's feet and the metal disk, which is grounded, so the major voltage drop is across the feet. The shock zone is an unmarked 60° sector that is defined by distal visual landmarks in the room. The location of the rat is determined from an overhead television camera each 33 ms by a PC-controlled tracking system (Bio-Signal Group). When the system detects the rat in the shock zone, the shock is delivered and repeated every 1,500 ms until the rat leaves the shock zone.
Place avoidance training begins with a pretraining trial. The rat is placed on the rotating disk to explore the environment with the shock turned off for 10 min. After resting in the home cage for 10 min, the rat receives eight training trials with the shock turned on. There is a 10-min rest in the home cage between trials. Twenty-two hours after training, the rat was injected in both hippocampi with either chelerythrine (n = 5; 10 nmol in 50% DMSO-saline) or the vehicle (n = 4), and 2 h later, retention of the 24-h place avoidance memory was tested by returning the rat to the rotating disk with the shock off. The time to first enter 9th shock zone and the percent time in the shock zone estimated retention of memory. Immediately following the retention test for long-term memory, short-term memory was assessed. Short-term memory is established by turning on the shock for 10 min, and then retention is tested during a 10-min test period with the shock off.
Spatial reference memory is distinguished from spatial working memory in the eight-arm radial maze because in reference memory, information about which arm locations are consistently baited is valid across trials, whereas working memory requires spatial information for which arm locations were visited within a trial, information that is only useful for the specific trial. We used the standard four-arms baited, four-arms unbaited task variant [37]. In this task, lesions of the hippocampus increase working memory errors, but not reference memory errors [37]. This basic result [38,39] contrasts, however, with many studies indicating that the hippocampus is critical for spatial reference memory in water maze tasks and other tests of spatial reference memory [40].
Detailed surgical methods are reported in [20,21]. The rats were food deprived to 85%–90% of their free-feeding weight prior to training on the eight-arm radial maze. The maze was 220 cm in diameter with a 60-cm–diameter central platform. Each arm was 16-cm wide and radiated 80 cm from the center. The maze was wiped with 70% ethanol between trials and rotated 90° every day to discourage the use of internal maze cues. The day before formal training began, each rat received two 10-min shaping trials with all arms baited by placing approximately 0.05 g of a sweetened oatmeal cereal mash (Maypo; International Home Foods) in the sunken food well at the end of each arm. Two rats were on the maze for the first shaping trial; and 1 h later, each rat received a second shaping trail by itself. On training trials, four arms were baited, and the food cups at the ends of the unbaited arms had inaccessible mash to control for odor cues. The locations of baited and unbaited arms were constant for a subject and balanced across subjects. There were ten training trials on each day. The rat was confined to the center of the maze by a large, overturned transparent bowl prior to each trial. Once released, the rat was free to forage until it consumed all the accessible food, or until 3 min had elapsed. Entry to an arm was scored when the rat crossed the halfway point of an arm. A trial was scored for correct entries, reference memory errors (visits to unbaited arms), and working memory errors (return visits to an arm) [41]. Training continued for 6 d (60 trials) to establish a strong memory. The next day, 24 h after training ceased, a single reinforced trial was given to test memory. Two hours prior to the memory test, each rat received a bilateral DH infusion of saline, scr-ZIP, or ZIP.
We used a standard version of the water maze to assess spatial reference memory. The training protocol establishes a hippocampus-dependent memory, which can be demonstrated on day 6 by an inability to localize searching for the platform on a probe trial following hippocampal inactivation [20].
The consensus is that the DH is important for spatial reference memory in the water maze, but whether the DH is crucial for storing this spatial reference memory is controversial. Although tetanic stimulation to saturate potentiation of synaptic transmission, pharmacological blockade of N-methyl-D-aspartate receptors, and permanent and functional lesions of the DH all impair learning and memory of the escape location [20,42,43], the impairment is absent in rats that had learned the water maze procedure, but not the particular escape location, prior to the amnestic intervention [44–47].
Detailed methods were reported [20]. The rats were trained in a 1.83-m diameter circular pool filled with 40-cm deep 21–22 °C opaque water. The rats were trained to find a circular, 10-cm–diameter clear Perspex platform that was submerged 1 cm below the water surface halfway between the center of the pool and the wall along the 0° radius. A rat was released in pseudorandom order from one of four equally separated locations along the wall. Each release location was used once in a four-trial training block. The path of the rat was automatically tracked with an overhead television camera (iTrack; Bio-Signal Group). Software (TrackAnalysis and TrackExplorer; Bio-Signal Group) calculated the latency to escape onto the platform, the time spent in each quadrant centered at 0°, 90°, 180°, and 270°, the time in each 11.5-cm–square region of the pool, and each 2 s, the rat's swim speed. Acquisition training lasted 5 d (two blocks/day). On day 6, 22 h after the last training, the rats were injected with saline, scr-ZIP, or ZIP, and 2 h after that, retention of the long-term place memory was tested by a probe trial with the escape platform removed. On this probe, the rat was placed in the center of the pool, and where it spent its time was measured for 1 min.
Contextual fear. Lesions of the DH disrupt the general, contextual component of the learned fear response in contextual conditioning paradigms. We conditioned rats in a standard, combined context and tone-conditioned fear protocol in which the rat received five shocks [9]. Following this training, long-term retention of contextual, but not tone fear is impaired by posttraining DH lesions [9,48]. We also examined two other context-conditioning protocols in which only a single shock was administered. In one protocol, the shock was paired with a tone; in the other, the tone was not presented.
The procedures have been described in detail [49]. Aluminum and Plexiglas conditioning chambers housed in sound-attenuating cabinets were used. The floor of each chamber was made of parallel rods that could deliver pole-scrambled, constant current foot shock (1.0 mA, 2 s). The chambers were contextually distinct. “Context A” (used for conditioning and context retention testing), had working ventilation fans that produced background noise (65 dB). The chamber lights and room lights illuminated the space because the sound-attenuating chest doors were open. The chambers were cleaned with a 1% ammonium hydroxide solution, which covered the surface under the shock floor. The chest doors were closed for “Context B” (used for tone retention testing), and fluorescent red light provided illumination. The ventilation fans were inactive, and the chambers were cleaned with a 1% acetic acid solution, which covered the pan below the shock floor.
Movement in each chamber was monitored using load cell inputs to a 5-Hz analog-to-digital converter calibrated to activity values between 0 and 100. Freezing behavior was automatically detected as inactivity (activity value <10) during at least 1 s (Threshold Activity software; Med-Associates).
Each rat was placed in a conditioning chamber for training, and after 3 min, five tone (2 kHz, 80 dB, 10 s) shock (1.0 mA, 2 s) pairings were delivered (70-s intertrial interval). The tone coterminated with the shock. Twenty-two hours after training, both dorsal hippocampi were infused with the saline, scr-ZIP, or ZIP solutions. Twenty-six hours after the infusions, long-term retention of contextual fear memory was assessed by measuring freezing behavior during a 10-min extinction session in Context A. Seventy-four hours after the infusions, long-term memory for the tone-shock association was assessed by measuring freezing during an extinction test in the novel Context B. In this test, an 8-min continuous tone was presented 2 min after a rat was placed in the chamber.
Tone fear. Specific associations between cues and fear are formed in the BLA [11]. We used a standard tone-fear conditioning protocol, in which a single tone cue is paired with shock. Detailed surgical procedures were described [50]. Training and testing sessions were conducted in two contextually distinct, aluminum and Plexiglas conditioning chambers, similar to the ones used for contextual fear conditioning. Freezing was automatically measured for 2 s of every 5 s for 10 min (Med PC version 4; Med Associates).
After 3 d of habituation to the two contexts, the rats were placed in Chamber A, and after 4 min, a 30-s, 90-dB, 5-kHz tone was played. The tone coterminated with a 1.5-mA, 1-s foot shock. Thirty seconds later, the rat was returned to its home cage. One hour after exposure to shock in Chamber A, the rat was placed in Chamber B for 5 min without the tone or shock. Twenty-two hours after training, the rats were infused with saline, scr-ZIP, or ZIP. Two or 24 hours later, long-term memory retention was tested by placing the rat in Chamber B, exposing the rats to the tone, and then measuring freezing in response to the tone. Six weeks after tone-shock conditioning, immediate postshock freezing was assessed in a counter-balanced subset of rats (n = 18). ZIP or saline was infused into both BLA sites, 5 min or 2 h prior to testing. An animal was then placed in Chamber A, and after 4.5 min, it was shocked once (1.5 mA, 1 s). Freezing behavior was scored during the subsequent 10 min.
The BLA is important for inhibitory avoidance, because it modulates the strength of information as it is stored at extra-amygdala sites during a posttraining consolidation window lasting several hours [32,33]. Whether the BLA also stores associations for maintaining inhibitory avoidance beyond this time window, however, is controversial. We used a standard inhibitory avoidance protocol in which the rat is shocked once when it enters from the brightly lit side to the dark compartment of the conditioning environment [51].
Experiments were performed as previously described [51]. Briefly, the inhibitory avoidance training apparatus consisted of a rectangular box comprising two compartments, a safe (brightly lit) one and a shock (dark) one, separated by a vertically sliding door (Med Associates). During training, each rat was placed in the safe compartment with its head facing away from the door. After 10 s, the door automatically opened, allowing access to the shock compartment, and latency to enter was taken as a measure of inhibitory avoidance acquisition. The door closed after the rat completely entered the shock compartment, and 2 s later, a brief foot shock (0.9 mA, 2 s) was delivered. The rat was then removed from the apparatus and returned to its home cage. Twenty-two hours after training, the rats were infused with saline, scr-ZIP, or ZIP. Long-term inhibitory avoidance memory was then tested 2 h later, by placing the rat back in the safe compartment and measuring the latency to enter the shock compartment. A foot shock was not delivered during the retention test. For animals that did not enter the shock compartment, the test was terminated at 9 min.
Group comparisons were made by ANOVA. Significance was accepted for p < 0.05. When appropriate, Newman-Keuls post hoc tests were performed. Student t-tests were used to compare group performance during retention of place avoidance against the chance value.
|
10.1371/journal.pcbi.1005899 | Method for the simulation of blood platelet shape and its evolution during activation | We present a simple physically based quantitative model of blood platelet shape and its evolution during agonist-induced activation. The model is based on the consideration of two major cytoskeletal elements: the marginal band of microtubules and the submembrane cortex. Mathematically, we consider the problem of minimization of surface area constrained to confine the marginal band and a certain cellular volume. For resting platelets, the marginal band appears as a peripheral ring, allowing for the analytical solution of the minimization problem. Upon activation, the marginal band coils out of plane and forms 3D convoluted structure. We show that its shape is well approximated by an overcurved circle, a mathematical concept of closed curve with constant excessive curvature. Possible mechanisms leading to such marginal band coiling are discussed, resulting in simple parametric expression for the marginal band shape during platelet activation. The excessive curvature of marginal band is a convenient state variable which tracks the progress of activation. The cell surface is determined using numerical optimization. The shapes are strictly mathematically defined by only three parameters and show good agreement with literature data. They can be utilized in simulation of platelets interaction with different physical fields, e.g. for the description of hydrodynamic and mechanical properties of platelets, leading to better understanding of platelets margination and adhesion and thrombus formation in blood flow. It would also facilitate precise characterization of platelets in clinical diagnosis, where a novel optical model is needed for the correct solution of inverse light-scattering problem.
| Blood platelets are the second most numerous component of blood after red blood cells. Their main function is to stop bleeding upon vessel wall injury. Contact with foreign substances, normally absent inside the bloodstream, leads to platelet activation. After this, platelets adhere to the damaged surface and to each other, finally forming a mechanical plug. Activation is very fast and complex process, and is accompanied by dramatic change of cell shape. This morphological alteration may be vital for the prevention of blood loss. However, there is no complete understanding of its significance, maybe because the quantitative description of the platelet shape during activation is absent. Here, we describe the shape change based on the physical consideration of the platelet cytoskeleton. We propose the formulation of mathematical problem and its solution, which allows one to easily simulate a wide class of shapes for both resting and activated cells. These models are in a good agreement with platelet images observed in numerous experiments published elsewere. They may be used to simulate the shape change and hence the influence of activation on hydrodynamic, mechanical and optical properties of platelets. As a result, novel diagnostic and therapeuthic strategies and clinical implications may be obtained.
| Human blood platelets are the paramount element of hemostasis and also contribute to a variety of other normal and pathological processes, including thrombosis, inflammation, and tumor development. These corpuscles attracted attention of early microscopists by the outstanding capacity to change physical properties in response to vessel wall injury or foreign substances [1]. The first step in reaction to such stimuli is platelet activation, which comprises a series of prothrombotic events, triggered by the increase of intracellular calcium [2–4]. Among them, morphological “disk-to-sphere” transformation is the earliest and the most notable effect. Dramatic shape change of platelet from discoid (resting) to rounded with pseudopodia (activated) occurs within seconds and leads to the alteration of optical, hydrodynamic, and mechanical properties of cells.
Quantitative description of the platelet shape and its change during activation is of great importance for various related fields. For instance, the platelet shape change affect the traces of light transmission aggregometry, a “gold standard” technique in clinical laboratories [5], and should be considered for the correct interpretation of test results [6,7]. Novel optical methods for single-cell morphological analysis [8–11] rely on the appropriate optical model for the solution of the inverse light-scattering problem. Correct description of platelet geometry is also vital for the simulation of blood cells motion in vessels [12–16], especially platelets margination and adhesion under flow [17,18]. An oblate spheroid is commonly used to approximate the platelet shape in optical and hydrodynamic simulations. This model accurately captures cell properties in a limited number of cases. Even if the influence of pseudopodia is not significant, which is the case for optical computations [19], the cell body itself may have a complicated shape. Electron microscopy images of platelets reveal the presence of cells lacking the axial symmetry [20]. While some of them only slightly deviate from the classic discoid shape, the others show large distortions, such as an overall bending. These curved cells have been thought of as an artifact for decades, but their existence is consistent with recent studies of platelet cytoskeleton [21]. There is an urgent need for a shape model which accounts for platelet microanatomical structures and captures a wide class of morphologies, including non-axisymmetric ones. Consideration of cytoskeletal elements underlying the platelet morphology is necessary for the understanding of mechanical properties of platelets (e.g., deformability), which is a topic of active experimental and computational research [22–24]. However, the model should be simple (controlled by a few tunable parameters) to facilitate the advancements in the above-mentioned fields.
Resting platelet possesses the peripheral ring (marginal band) of microtubules, which supports cell discoid shape [25–27]. Upon activation, it coils to form a saddle-like structure, which was recently observed and regarded as the principal mechanistic feature underlying cellular shape change [28]. However, the specific cause of this coiling may be related to different mechanisms (which we discuss in a specific section). Given this uncertainty, direct simulation of marginal band coiling seems to be impractical.
Instead, in this paper we propose to model coiled marginal band using a mathematical concept of the overcurved circle [29]. Such structures approximate a wide class of coiled rings formed in a variety of physical processes. Its geometry is controlled by a single parameter called the overcurvature, which increases from unity as the circle coils out of plane. We show a remarkable similarity of experimental images (published elsewhere) and the overcurved circles, which allows us to consider the marginal band overcurvature as a state variable of activation. Next, we model the outer cell membrane and the submembrane cortex as an elastic surface with constant mean curvature (or, equivalently, minimal surface area), which rests on the marginal-band skeleton and embraces a prescribed cell volume. The particular surface configuration is determined by the numerical optimization. The major advantage of this platelet shape model is that it is defined by only four parameters (or three up to the scale): length, overcurvature, thickness of the marginal band, and the cell volume. Small number of parameters is vital for massive simulations and especialy for inverse problems solutions. At the same time the model encompasses both classical discoid morphology (overcurvature = 1), and novel class of non-spheroidal, highly curved or concave shapes. We compare the simulation results with previously published platelets images and show that the new model constitutes a major improvement over the spheroidal model, still preserving reasonable simplicity. Finally, we discuss important biological and biophysical insights into platelets behavior and predict cell morphologies that have not yet been reported.
Mouthuy et al [29] showed that overcurved circles are formed in a variety of physical conditions, from curved origami to foldable tents. Surprisingly, these structures approximate marginal bands of activated platelets as well. In Fig 1A, the images of coiled marginal bands of platelets under different stimuli are reproduced from [28]. We show direct one-to-one comparison of experimental images with the overcurved circles, which manifests the remarkable agreement. An overcurvature Op is listed near each plotted circle.
The overcurvature is the ratio between the circle’s curvature and the curvature of the same-length planar ring. As Op increases from 1, the circle comes out of plane and adopts more and more twisted three-dimensional shape (Fig 1B). A great advantage of the overcurved circles is a simple one-parametric analytical representation. They are also invariant to the 90° rotation around the z-axis combined with the reflection with respect to the xy-plane (D2d symmetry). Possible mechanisms leading to the formation of this particular structure in platelets are discussed in the special section, together with the mechanisms causing deviations (including the tensioned submembrane cortex).
In case of blood platelets, especially discoid cells, the thickness of marginal band can not be neglected. Further we model it as the tube of circular cross-section, whose central line forms an overcurved circle. Circular cross-section is consistent with published experimental data, although there may be reasons for compression and extension in certain directions. Two-parametric expressions for the tube surface are derived below, and several examples are shown in Fig 1C–1E. The geometry (relative shape) of platelet marginal band is therefore controlled by two parameters—the overcurvature Op and the relative tube radius r/R (thickness); the radius of the ring in the initial (planar) state R is a conventional scaling parameter.
First we note that the minimization of surface energy has been used for a long time to describe the biconcave shape of red blood cells (RBCs), the most studied element of blood [30,31]. Generally, this energy consists of three parts: expansion, shear, and bending [32]. The RBC surface comprises relatively weak spectrin cytoskeleton and the lipid bilayer, whose expansion or compression requires unrealistically high energy. Minimization of membrane bending energy alone at a constant area provides the shape which agrees well with experiments, at least in case of native, non-deformed RBCs [33].
By contrast, blood platelets have strong actomyosin submembrane cortex [27]. The filaments pull together nearby points on cell membrane and make it wrinkled at small scales [34]. In other words, the expansion energy is associated with the cortex, which has equilibrium surface area significantly lower than that allowed by the fixed volume. And this expansion energy is assumed to be much larger than bending energy of both the lipid bilayer (including wrinkles) and the cortex. Therefore, the description of a platelet outer shell (membrane + submembrane cortex) as a surface which tends to minimize its area is a reasonable approximation, that have been used in biophysical modeling of blood platelets [35]. We also do not account for the rigid cytoplasmic actin network, which stabilizes resting platelet shape and rapidly disassembles during activation [36]. Thus, the surface should cover the coiled marginal band and confine certain cellular volume, while minimizing the area. Such mathematical problem has not been considered elsewhere; however, one would expect that one part of surface would be attached to the marginal band and the other would be of constant mean curvature. The only difficulty is to find a boundary where the surface detaches from the marginal band. In the section “Cell surface of resting platelets”, we give a solution for axially symmetrical problem (Op = 1) and provide the boundary conditions for the general case. We show that the contact between attached and free parts of the surface should be smooth. These considerations were used for the following simulation of platelets morphology, details of which are described in the subsequent section. Further we show attached parts of the surface in green and free surface in red.
Marginal band in resting platelets appears as a planar ring of microtubules. In our model, it corresponds to a ring torus, characterized by Op = 1, ring radius R, and the relative tube radius r/R < 1. The overall morphology preserves axial symmetry; in this case, the analytic solution of the surface optimization problem is possible. The resting platelet is described as the torus (marginal band) with two spherical caps covering the torus hole. Possible shapes for R = 1 and r/R = 0.2 are shown in Area I of Fig 2.
In this figure a phase diagram of possible platelets morphologies is shown in the coordinates of overcurvature Op and the dimensioness volume v = V/R3, while marginal band thickness r/R is fixed to 0.2. Dark gray area on the bottom represents volumes less than that of torus; if the cell volume v is slightly above, it adopts biconcave shape, which is indeed rarely seen for platelets (see e.g. Fig. 2A in [37]). As the volume increases, caps change from inward to outward, and the cell becomes discoid and then biconvex, which corresponds to the classical platelets appearance. Further cell inflation leads to the formation of nearly spherical shapes, that are not characteristic for intact platelets, but correspond to cells after osmotic swelling [38]. Finally, the cell volume hits the boundary of area III, where it is larger than that of the sphere circumscribing the marginal band. In this case, the marginal band no longer stretches the membrane, and the cell adopts spherical shape. Hypothetically, the uneven distribution of submembrane microfilaments may cause certain regions of surface to bulge and form protrusions like pseudopodia or lamellopodia.
During activation, marginal band of platelets may alter its length and thickness, according to Diagouraga et al. [28]. For the simplicity, we assume that the relative thickness of marginal band r/R = 0.2 holds in this process. Activation begins with the rapid out-of-plane buckling of the marginal band, resulting in jump from area I to II in Fig 2. Area II consists of platelet shapes with Op > 1. Note the unusual D2d symmetry of shapes, inherited from the overcurved circles. The rightward movement on the phase diagram reflects the degree of marginal band coiling and hence the progression of activation. The cell volume may also alter upon this movement, but there is no agreement on this topic in the literature. In the region of small v cell surfaces preserve concave appearance, and such morphologies indeed present in experiments, see e.g. Fig. 2a in [20] and Fig. 2c in [21]. Larger cells with convex of intermediate appearance can be found in a number of recent studies, see e.g. Fig. 2A in [28] and Fig. 7A in [39]. Structures with larger values of Op, characterized by self-contact or self-intersection, have not been encountered among platelets. While self-intersection may take place for the mathematical concept of overcurved ring, it is physically impossible for the marginal band, given its non-negligible thickness. Therefore, the case of Op > 2.2–2.4 ceases to describe platelet structure, and we do not consider such values in the present paper. However, self-intersecting structures were observed in invertebrate erythrocytes, which also have marginal band and change morphology from discoid to rounded upon blood withdrawal [40]. These cells are much larger than human platelets and have much smaller relative thickness of marginal band. Note also that the mathematical upper limit of Op = 3 corresponds to the planar triple-folded ring, which can be coiled again to make a new overcurved circle.
The trajectories of platelets on Fig 2 during activation are not necessarily horizontal since the cell volume (and marginal band length) can vary. Nevertheless, the increase of overcurvature leads to rapid decrease of marginal band dimensions, i.e., of the volume of the circumscribing sphere. Eventually it becomes smaller than the cell volume V, and platelet proceeds to the zone III. The marginal band no longer stretches the membrane, which may lead to the detachment and disorganization of microtubules constituting the irreversible activation. In the case of weak stimulus, platelet trajectory reverses before the area III, following the return of intracellular calcium concentration to the baseline. This simple physical framework for revesible/irreversible activation may shed light on the platelet decision-making mechanisms [41]. However, the relationship between the intracellular calcium concentration and the overcurvature of marginal band is an important topic for future research. Our model of platelet activation is animated in the S1 Movie including the stages of marginal band bending and cell spherizing, after which pseudopodia extrusion can be additionally simulated (please see below for detalis). In this video the cell volume is constant.
The marginal band coiling implies the action of forces. Mechanical work of these forces is stored as a bending energy of microtubules in a bundle, which is proportional to the local squared curvature. As a natural first approximation, we assume that the amount of stored energy per unit length should not vary significantly, which implies nearly constant curvature of the coiled marginal band. However, only detailed considerations of possible mechanisms can prove this hypothesis.
The platelet cytoskeleton comprises the following major elements: marginal band of microtubules, submembrane cortex, actin filaments and the molecular motors associated with microtubules. It is widely accepted to attribute marginal band coiling solely to the influence of cortical tension. However, platelet activation affects every cytoskeletal subsystem in various ways, and each of them can contribute to the shape change:
The roles of each mechanism are to be studied; however, all of them lead to approximately the same morphology of the marginal band, which we discuss in detail in the next section.
The concept of the excessive curvature naturally arises in the first mentioned mechanism. In platelets, the marginal band consists of multiple microtubules with different polarity [52]. Molecular motors exert sliding forces on the microtubules towards the plus-end (in case of dynein). Two microtubules with opposite polarity may be joined by dynein [53], which is inactive (or counterbalanced by kinesin) in the resting state. Activation shifts the motor balance [28], which leads to microtubule sliding and formation of elongating and shortening regions in the marginal band (Fig 3).
Additionally, these regions are cross-linked by protein bridges [35]. This resembles the mechanism of flagellar beating, also driven by dynein-mediated sliding of cross-linked microtubules, which leads to constant-curvature bending [54]. Another analogy is the bimetallic strip, which bends during heating due to the different thermal expansions of its joined metallic constituents. If the initial structure is the closed planar ring, formation of excessive curvature induces out-of-plane coiling. Experiment of Mouthuy et al [29] indeed showed that bimetallic ring forms the overcurved circle after temperature change.
The described interplay between microtubules and molecular motors in the marginal band may constitute the major or at least significant factor of its coiling in platelets. Nevertheless, different mechanisms of the excessive curvature formation are possible. For instance, dyneins or kinesins can be anchored in the submembrane cortex. Cortex influence on the microtubules bundle are discussed in the following section.
The influence of cell surface tension, which is implied in the majority of above-mentioned mechanisms, may lead to the coiling of marginal band with non-constant curvature. Direct simulation of filamentous ring growth in flexible confinements shows distinct packing patterns depending upon elastic properties of the filament and the shell [55]. However, in each case the ring coils to a saddle-like structure at the initial step, very similar to the marginal band coiling in platelets. It is not clear whether these structures obey the constant curvature condition or the similarity is only visual. From physical point of view, in the limit of large surface tension the shell would adopt spherical shape, irrespectively to the internal cytoskeletal elements. It is useful to consider the case when microtubules lie on spherical surface, at the same time minimizing the bending energy. The solution of the corresponding mathematical problem was obtained by Guven et al [56]. The only stable configuration possesses the same D2d symmetry as an overcurved circles used in the present paper. However, while the normal component of its curvature vector is constant, geodesic component oscillates around zero with non-negligible amplitude. The total curvature can be found from these two components by the Pythagorean theorem; it would also be the sum of some constant value and oscillations (generally non-symmetrical). For this point, we conclude that the overcurved circle is suitable as the first approximation in the limit of large surface tension, as well as in other discussed cases. It allows us to capture symmetry relations and the overall structure of platelets shapes. However, for the studies which rely on platelets microstructural properties our method may not be accurate enough. For instance, Zhang et al [57] had to directly model the molecular-scale intra-platelet constituents to account for the influence of hemodynamic stress on the cells (i.e., shear-induced activation). Another experimental study [24] deals with platelet deformability at nanoscale level and the ability to exert contraction force. This force was successfully used for the targeted drug delivery using platelet-hybridized capsules, which rupture upon activation [58]. Generally, in such constraint conditions, one could expect the marginal band to deform and adopt complex shape, dependent on the applied force. It is an interesting research direction to consider perturbations of overcurved circle, such as curvature oscillations and unidirectional compression.
We model the marginal band of platelets as an overcurved ring (tube) with circular cross-section. The central line consists of 4 lobes, each of them being a part of the Salkowski curve γm(t) = (xm(t), ym(t), zm(t)) [59], where t is a parameter and the internal variable m can be uniquely (numerically) determined from overcurvature. The only modification is that we normalize the overcurved circle so that its length equals 2π for all values of Op, denoting the normalization coefficient as 1/R. Tubular surface is defined as follows: for each point on the γm(t)/R, we take a vector along the curve normal with the length r/R, and rotate it around the tangential direction. End of the vector gives the circle, which would be the cross-section of the tube at this particular point. In the Frenet frame (tangent, normal, binormal) the vector is defined by
r=(0,rRcosφ,rRsinφ)
(1)
where φ is the angle of rotation. Now we should transform the coordinates to the laboratory reference frame. For this purpose, one can use the transition matrix to the Frenet frame from [59] and its inverse. We obtain
T=(−costcoscosnt−nsintsinnt−cosntsint+ncostsinnt−nmsinntnmsint−nmcost−1ncosntsint−costsinnt−ncostcosnt−sintsinntnmcost)−1==1n(1+m2)(−(1+m2)ncostcosnt−m2sintsinntmsintm2sintcosnt−n(1+m2)costsinnt−(1+m2)ncosntsint+m2costsinnt−mcostm2costcosnt−n(1+m2)sintsinnt−msinnt−m2mcosnt),
(2)
where n=m1+m2. Finally, two-parametric expression of the tube surface (for one lobe) is as follows:
M(t,φ)=γm(t)R+rR(T12cosφ+T13sinφ,T22cosφ+T23sinφ,zm(t)+T32cosφ+T33sinφ)
(3)
Four copies of (3) are then rotated, translated, and joined using the procedure described in [29] to form an overcurved ring with a circular cross-section.
Inside resting platelets microtubules are organized in a circumferential ring, which, under our assumption of circular marginal band cross-section, corresponds to the torus. Its radii are R and r (Fig 4).
The surface adopts configuration with minimal area, while it must contain the torus and confine certain cellular volume V. This problem possesses axial symmetry, and we assume that the sought surface is also axisymmetric. So we further consider a sectional plane passing through the axis of revolution (defined to be the y-axis). Another coordinate axis in this plane is the x-axis, which belongs to the torus symmetry plane. Additionally we assume that the platelet membrane is a closed bounded surface with zero genus, i.e. has no holes, in accordance with all existing platelet images.
We seek for the parametric profile x(t) ≥ 0, y(t), t ∈ [0,T], for which the surface of revolution has minimal area:
S=∫0T2πxx′2+y′2dt→min
(4a)
subject to
V=∫0Tπx2y′dt=const,
(4b)
(x−R)2+y2≥r2,
(4c)
x(0)=x(T)=0,
(4d)
where the prime denotes differentiation over t, and condition (4d) follows from the zero-genus assumption above.
Condition (4c) makes the problem distinct from that of finding constant mean curvature surface with free boundary. However, generally the profile consists of two parts: one is attached to the marginal band and follows its exterior and the other is free (both of them may consist of several disjoint components). From physical considerations, the free part should obey the constant-mean-curvature condition, which (in the case of axial symmetry) leads to the family of Delaunay surfaces [60]. The only Delaunay surface which satisfies condition (4d) is a sphere, so one can conclude that free-surface parts intersecting the symmetry axis are necessarily spherical caps. Moreover, those caps have the same radius so that the whole free surface has the same mean curvature (including sign), which physically corresponds to constant pressure of the surface-tension forces. The mean curvature of the attached part equals that of the marginal band, which is always larger than that of the free part, since the latter spans the whole diameter of the torus. And it is the reaction force of the rigid marginal band which compensates for the excessive surface tenstion. If the free part would attach to only one side of the torus (when considered inside the section of Fig 4), then it would have even larger mean curvature than that of the marginal band; hence, such free parts are impossible.
In the following we obtain the same solution formally from the problem statement (4). By introducing the slack variable f 2 = (x−R)2 + y2 − r2, we transform condition (4c) into equality. Application of Euler-Lagrange multiplier theorem leads to the minimization of functional
∫0bL(t,x,x′,y,y′,f,f′,λ2,λ2′)dt,L=2πxx′2+y′2−2λ1πx2y′−λ2[(x−R2)+y2−r2−f2],
(5)
where 2λ1 and λ2(t) are Lagrange multipliers (factor of 2 is introduced for simplicity). The set of equations follows:
{2πx′2+y′2−4πλ1xy′−2λ2(x−R)−ddt(2πxx′x′2+y′2)=0(6a)−2λ2y−ddt(2πxy′x′2+y′2−2λ1πx2)=0(6b)(x−R)2+y2−r2−f2=0(6c)2λ2f=0(6d)
Let us first consider the segments for which inequality (4b) is not active, i.e. when f 2(t) > 0. Then Eq (6d) gives λ2(t) = 0, and Eq (6b) yields:
xy′x′2+y′2−λ1x2=C,
(7)
where C is constant for each continuous segment (interval of t), but may differ between them. It can be shown that general solutions of Eq (7) are Delaunay surfaces with constant mean curvature. However, as discussed above, there can be at most two such free segments, containing t = 0 and t = T, respectively. Then condition (4d) implies C = 0 for all of these segments. Hence, Eq (6a) and (6b) are transformed into
x′2+y′2−2λ1xy′=ddt(xx′x′2+y′2),
(8a)
y′=λ1xx′2+y′2.
(8b)
Eq (8b) implies
x2+(y−b)2=λ1−2,sgn(y′)=sgn(λ1),
(9)
which is a circle with the center on the y-axis and radius 1/|λ1|. Further, we additionally assume that our parametrization is counter clockwise, i.e. y(0) < y(T), then Eq (9) corresponds to a circle with signed outward curvature equal to λ1. Interestingly, Eq (8a) is automatically satisfied for any such solution. Considering two possible free segments, we note that integration constant b may differ, but λ1 is the same. The latter corresponds to mean curvature being constant over all free segments.
If the attached segment has zero length, then the profile is a single semicircle, and the whole platelet surface is a sphere containing the marginal band inside (potentially touching it). Otherwise we additionally need the boundary conditions between the attached and free segments, e.g. at the point t = u. Naturally, both x(t) and y(t) are continuous at t = u. Further, we apply Weierstrass–Erdmann conditions:
∂L∂x′|t=u−0t=u+0=0⇒x′x′2+y′2|t=u−0t=u+0=0,∂L∂y′|t=u−0t=u+0=0⇒y′x′2+y′2|t=u−0t=u+0=0,
(10)
noting that the rest of those conditions are trivial:
∂L∂f′=∂L∂λ2′=0,L−x′∂L∂x′−y′∂L∂y′−f′∂L∂f′−λ2′∂L∂λ2′=−λ2[(x−R)2+y2−r2−f2]=0,
(11)
where the last equality is due to Eq (6c). While Eq (10) does not guarantee continuity of x′ and y′ at t = u, it implies the continuity of tangent direction (i.e. x′/y′ or y′/x′) at t = u. Physically, it is also justified if there are no adhesion forces (at least near the contact line) between the marginal band and the cell membrane. In this case only tension forces are present, which are directed along each surface segment. For these forces to cancel out, they must be antiparallel, so the segments should join smoothly. The equivivalent boundary condition is the zero contact angle between the cell surface and the marginal band (i.e., “superhydrophilic” [61] marginal band exterior). Note that this physical interpretation remains valid in the general case of activated platelets (no axial symmetry).
In the case of resting platelets, the condition (10) together with f = 0 and Eq (9) implies that the marginal-band circle is tangent to the semicircle of Eq (9), dividing the latter into two parts, one of which belongs to the sought profile. The centers of two circles and contact point lie on a line, which leads to the following equation for a = x(u):
λ1a=a−Rr,|a−R|≤r,
(12)
for which y(u) can take one of the two values with opposite signs. Note that positive and negative signs of λ1 corresponds to internal and external tangency, respectively. The solution of Eq (12) is given by:
a=R1−λ1r,−1R−r≤λ1≤1R+r.
(13)
If the radius of free-segment circle is smaller than the torus hole radius R − r (|λ1| is too large), joining of two curves is not possible at all. While a small semicircle can minimize the functional (5), it does not surround the marginal band. For larger free-segment radius but negative λ1, there are two circles externally tangent to the marginal band. There are two ways to combine the parts of these circles with parts of the marginal band into a smooth profile, but only one of them, corresponding to a biconcave disk, surrounds the marginal band. Similarly, for positive λ1 there are two cirles, to which the marginal band is internally tangent. In principle, the segments of these circles can be combined smoothly with the marginal band in three ways, but two of such profiles would only touch the marginal band at a single contact point corresponding to the semicircle solutions described above. The remaining solution is a biconvex one. Importantly, all solutions except a semicircle are symmetric with respect to the x-axis and are uniquely determined by the value of λ1 [cf. Eq (13)].
The only additional complication is that the segments of the biconcave shape must not intersect, i.e. each concave segment should not cross the x-axis (zero-genus assumption). The minimum possible curvature is thus −2r/(R2 –r2). For smaller (more negative) curvatures there exist surfaces with flat double layer in the center, that potentially may be obtained for real platelets by removing sufficient amount of volume. Still, we do not consider them here, as they require modification of the functional (5) to explicitly include the double-layer part. Finally, all symmetric solutions can be expressed with the following formula:
y(x)=±{1λ1(1−(λ1x)2−(1−λ1r)2−(λ1R)2),0≤x≤R1−λ1rr2−(x−R)2,R1−λ1r<x≤R+r},−2rR2−r2≤λ1≤1R+r
(14)
It can be shown that y(x) is smooth with respect to λ1 in the whole specified range, including λ1 = 0. It is straightforward to obtain the cell volume V from Eq (13), but the final expression is rather cumbersome. Importantly, V monotonically increases with λ1, which also follows from not-crossing of profiles for different λ1, i.e. for any two profiles, one completely surrounds the other (potentially touching it).
We summarize all the obtained morphologies of resting platelets, and provide the characteristic values of λ1 and V in a table presented in Fig 5.
Volume increases from top to bottom and λ1 is a single-valued function of V for all ranges of shapes. Thus, the variational problem (4) has unique solution (except for vertical translations of a semicircle profiles). If only symmetric attached profiles are considered (bounded V), curvature λ1 is a convenient internal parameter, bijectively related to V.
The numerical optimization algorithm of platelets shape for arbitrary overcurvature of the marginal band was briefly described above and summarized in Fig 6.
It is implemented as the Surface Evolver [62] script, available as S1 File. If cell parameters correspond to non-sphered shape (volume is not too large; cf. Fig 5), the optimization procedure begins. At the first step we construct the surface of overcurved ring (3) from given Op and r/R with φ varying from φd to 2π−φd. For this purpose, we define facets that are constrained to lie on boundary and refine them several times using intrinsic Surface Evolver procedure (r command). The obtained surface (Fig 5A) simulates the region where cell membrane sticks to the marginal band. It is excluded from the following optimization (fixed). We also define two mobile parts of the surface: first is bounded by the space curve φ = φd and the second by φ = 2π−φd, corresponding to the upper and lower cell halves (Fig 5B). These parts are refined and optimized to have minimal area with the requirement that the overall surface should confine the prescribed volume V. Standard procedure (g and r commands) is used in this step.
Next, we optimize the interface at the border between mobile and fixed parts. The aim is to make the surface smooth, i.e., minimize the contact angle, analogously to the condition (10) in the axisymmetric case. So we attenuate the contact position φc(t) along the border. For this purpose we implemented special procedure bopt, which computes the new positions of vertices on the overcurved circle (simple one-step gradient optimization). Next the surface is optimized, which again leads to the non-zero contact angle. After several iterations, we obtain reasonably converged shape. The typical process is shown in S2 Movie. We stress that the physical reasoning behind our method requires both contact angles and the free-surface to be simultaneously optimized (as is in Fig 5E). The two-level iterative optimization is just one of the possible options to achieve this goal. Note also that we do not consider the influence of the membrane tension on the marginal band. Instead, we assume that it is an overcurved ring with already known shape.
Characteristic feature of the activated platelets is the presence of pseudopodia, which influence the physical properties of cells. For instance, they can increase the hydrodynamic radius of cells. Light-scattering patterns also differ for platelet models with or without pseudopodia [19]. However, the main effect in this case is the decrease of volume of the central cell body, because protrusions with thickness less than the wavelength are weak scatterers. Anyway, we implemented the possibility to add simple membrane protrusions (resembling pseudopodia) by the following procedure. First, we fix the obtained surface except several vertices, randomly selected on the mobile surface parts. Next, we increase the target cell volume by a small amount, multiplying it by (1+ ppv), where ppv is a user-controlled parameter representing the relative volume of pseudopodia. On the next step, we perform a single evolver iteration (g), which forces the unfixed vertices to shoot out (Fig 7).
It should be stressed that this is a zero-order approximation of pseudopodia, mainly to give the shape model better visual resemblance to fully-activated platelet. It may correspond physically to the extrusion of feeble sites of platelet surface (i.e., membrane loci where no or little filaments are attached) by an intracellular pressure, but we do not consider microtubules or actin polymerization, which are the main mechanistical feature of the process.
In order to demonstrate the relevance of the proposed shape model for platelets research, we performed simulation of light scattering by these shapes using the discrete-dipole approximation (DDA). These simulations can be implemented in single script using the surface evolver’s ability to run system commands. The workflow consists of:
We performed simulation for the particle oriented as shown in Fig 8A, where the incident wave propagates from below. The refractive index of the platelet model was set to 1.03 (relative to the medium), the volume to 11.5 fL, and the wavelength in vacuum—to 405 nm, while medium refractive index was 1.337. In Fig 8B, the intensity of scattered light (more precisely, the Mueller matrix element S11) is shown as a function of polar and azimuthal scattering angles. The distribution of intensity shows complex features and is distinctive from smooth pattern of oblate spheroid described in [66]. Note also that the presented model of activated platelet not only possess one or two additional morphological parameter in comparison with models of resting platelets, but also lacks axial symmetry, which brings about the influence of Euler angle γ on the light-scattering patterns. These three additional parameters complicate the solution of the inverse light scattering problem and hinder the incorporation of the presented model into the existing algorithms [9,10]. Further theoretical and experimental research concerning optical properties of blood platelets is needed to extract new information and develop novel diagnostic techniques.
|
10.1371/journal.pgen.1007579 | The temporal landscape of recursive splicing during Pol II transcription elongation in human cells | Recursive splicing (RS) is an evolutionarily conserved process of removing long introns via multiple steps of splicing. It was first discovered in Drosophila and recently proven to occur also in humans. The detailed mechanism of recursive splicing is not well understood, in particular, whether it is kinetically coupled with transcription. To investigate the dynamic process that underlies recursive splicing, we systematically characterized 342 RS sites in three human cell types using published time-series data that monitored synchronized Pol II elongation and nascent RNA production with 4-thiouridine labeling. We found that half of the RS events occurred post-transcriptionally with long delays. For at least 18–47% RS introns, we detected RS junction reads only after detecting canonical splicing junction reads, supporting the notion that these introns were removed by both recursive splicing and canonical splicing. Furthermore, the choice of which splicing mechanism was used showed cell type specificity. Our results suggest that recursive splicing supplements, rather than replaces, canonical splicing for removing long introns.
| Nascent precursor RNAs newly transcribed from a DNA template are processed into mature RNAs by splicing out introns and ligating exons. Recursive splicing (RS) is a special kind of splicing whereby long introns are removed piecewise in multiple consecutive steps. Only nine human introns are known to undergo recursive splicing, and in this paper, we have developed a computational pipeline to search for RS introns in the human transcriptome using labeled and synchronized RNA sequencing data. We identified 342 RS sites in three human cell types and characterized their sequence and epigenetic features. We further show that half of the RS events occur post-transcriptionally and RS introns are removed by both recursive splicing and canonical splicing.
| Recursive splicing (RS)—a multi-step process to excise long introns from pre-mRNAs—was first identified in the Ultrabithorax (Ubx) gene of Drosophila melanogaster [1]. Each recursive splice site (RS site; also called a ratchet site in Drosophila) contains a pair of juxtaposed 3′ and 5′ splice sites, dividing a long intron into two segments to be spliced out sequentially. The RS site first functions as a 3′ splice site, pairing with the upstream 5′ splice site to remove the upstream intronic segment. The reconstituted 5′ splice site interacts with the downstream 3′ splice site for the subsequent removal of the remaining intronic segment.
Genome-wide studies revealed that hundreds of long fly introns perform recursive splicing [2, 3] and nine long introns of human neuronal genes contain high-confidence RS sites [4]. RS sites are generally followed by an “RS exon” that is crucial for the recognition of the RS site [3, 4]. The RS exons are retained in low-abundance transcript isoforms, resulting from the competition between the two 5′ splice sites—one at the RS site and the other one at the 3′ end of the RS exon [3, 4]. Some annotated cassette exons could also function as RS exons [4]. These findings bridge the gap between recursive splicing and alternative splicing in humans, suggesting that recursive splicing can, in principle, use the machinery of alternative splicing. However, the features of RS exons that differ from cassette exons have not been thoroughly investigated and the mechanism of conversion between them remains elusive.
Abundant evidence indicates that most human splicing occurs co-transcriptionally, and the transcriptional elongation by RNA polymerase II (Pol II) plays an essential role in regulating splicing by kinetic coupling [5, 6]. Recursive splicing, which is a special case of splicing, may also be regulated co-transcriptionally. Indeed, the Pol II elongation rate has been shown to exert an impact on the efficiency of recursive splicing of the fly Ubx gene [7]. However, the extent of coupling between recursive splicing and Pol II elongation in humans has not yet been studied. Moreover, it was proposed that recurrent splicing was constitutive in flies [2], but it is not known whether this is also the case for humans.
Half of the human genes contain at least one intron longer than 10 kb, and 14% of all human introns are over 10 kb (S1A Fig), much more common than fly introns (4%). To obtain a better understanding of the role of recursive splicing in long intron excision, we developed an integrative RS sites identification pipeline and applied it to previously published RNA sequencing datasets with 4-thiouridine (4sU) tagging of nascent RNAs upon synchronization of Pol II elongation with 5,6-dichlorobenzimidazole 1-ß-D-ribofuranoside (DRB) [8]. We identified 342 candidate RS sites in three human cell lines: embryonic carcinoma PA1 cells, embryonic stem cells (H9), and differentiated forebrain neuron progenitor cells. We measured the extent of kinetic coupling between recursive splicing and Pol II elongation and found half of the RS events occur post-transcriptionally with long delays. We also looked for evidence for or against constitutive recursive splicing in these human cells and found that many recursive splicing events co-occurred with canonical splicing. Furthermore, the choice of recursive splicing versus canonical splicing appears to be cell-type specific. Together, our results indicate that recursive splicing is part of the cell’s arsenal for removing long introns.
RS sites have several genomic features that can facilitate their identification. First, RS sites match the AGGT motif, i.e., the 3′ splice site consensus (AG) immediately followed by the 5′ splice site consensus (GT), and they are evolutionarily conserved [2, 4]. Second, RS sites are enriched in long introns because short introns are too spatially constrained to perform multiple-step splicing [9]. Third, the 3′ splice sites of RS sites need to be sufficiently strong to be recognized by the spliceosome at the first step of recursive splicing [10]. We systematically evaluate these features on the nine reported human RS sites [4] and canonical splice sites (S1A–S1C Fig). The results were then used to set cutoffs in the first step of our RS site identification pipeline (Fig 1A, step 1; S2 Fig, step 1; Materials and methods).
We first identified all potential sites that matched the AGGT motif in long introns. There are 65,019 long introns in the human genome, defined as being longer than 5,000 nucleotides (nt). We then eliminated the sites that overlapped with any splice sites in the UCSC Known Genes annotation, retaining those with strong 3′ splice sites (strength ≥ 5 computed with MaxEntScan [11]) and high evolutionary conservation (PhastCons score ≥ 0.5 [12]). We found 36,870 putative RS sites in 21,202 introns (Fig 1B, left; S1 Table), covering 33% of long introns, with an average of 1.7 putative RS sites per intron.
Another feature that we used to detect RS sites is the splicing intermediate that joins the 3′ splice site of each RS site with the 5′ splice site of the upstream exon. The second step of our pipeline identifies such RS junction reads by generating a set of putative 5′-exon–RS site junctions and realigning all sequencing reads against them. This strategy was used to map RS junctions in flies and increased the specificity of the method [2]. Sites with at least three unique supporting junction reads were identified and designated candidate RS sites for the remaining analyses (Fig 1A, step 2; S2 Fig, step 2; Materials and methods).
The last feature of RS sites implemented in our pipeline is the "sawtooth" pattern in the steady-state total RNA-seq data (with ribosomal RNAs removed). All nine previously reported human RS sites show strong evidence of a sawtooth pattern in the steady-state RNA-seq data in human forebrain cells (S1D Fig). However, the sawtooth pattern is difficult to detect for several reasons. First, the RNA-seq profile is noisy; thus, the sawtooth pattern is only detectable at a sufficiently high sequencing depth for genes that are expressed at high levels. Second, the sawtooth pattern varies by intron length and shows strong tissue specificity [13]. Third, the sawtooth pattern is based on the assumption that recursive splicing is a rapid process [2, 4, 13], and to what extent that is the case has not been evaluated. Consequently, we used the sawtooth pattern as the most stringent criterion in the third step of our pipeline, and candidate RS sites with the sawtooth pattern are called sawtooth RS sites (Fig 1A, step 3; S2 Fig, step 3; Materials and methods).
In summary, we have established a computational pipeline that integrates various known genomic or expression features of recursive splicing and uses stringent thresholding for RS site detection (Fig 1A; S2 Fig; Materials and methods).
The previous study identified recursive splicing in humans using steady-state total RNA-seq [4]. We postulated that time-course 4sUDRB-seq data might be more effective than steady-state RNA-seq data for detecting RS sites because the former could capture splicing intermediates that would exist only for a limited time after transcription. Thus, we analyzed the published 4sUDRB-seq data in human ovarian carcinoma PA1 cells, human H9 embryonic stem cells, and forebrain (FB) neuronal progenitor cells differentiated from H9 cells, previously generated for studying the biogenesis of circular RNAs [8]. Upon transcription synchronization by reversible DRB inhibition, nascent RNAs tagged by 4sU were sequenced at five time points from 10 minutes to 2 hours for all three cell types, and additionally at 4 and 16 hours for PA1 cells (Fig 1C). In addition to providing data for determining Pol II elongation rates genome-wide, 4sUDRB-seq data at such long durations capture complete transcription and splicing of genes with long introns. The sequencing depths of these datasets are particularly high, with the five deepest datasets ranging from 260 to 387 million reads (S3A Fig). These data are particularly suitable for studying recursive splicing.
Using our RS site prediction pipeline, we identified 342 candidate RS sites in PA1, H9, or FB cells (S2 Table), and 108 of them (31.6%) were in two or more cell lines, significantly enriched over random chance (permutation test p-value < 1.00×10−4; Fig 1B). Of these, 16 RS sites show a significant sawtooth pattern (S3 Table), including six of the nine known human RS sites [4] (Fig 1D). The other three known RS sites also showed the sawtooth pattern (S1D Fig), but we did not detect sufficient junction reads for two sites, and the third site overlapped with an annotated cassette exon and was filtered out by our pipeline. Because we could identify only the 5′ splice site of RS introns that corresponds to each RS site but not the 3′ splice site, we included all potential 3′ splice sites according to gene annotations. As a result, 342 candidate RS sites and 16 sawtooth RS sites correspond to 378 and 17 introns, respectively (Fig 1B).
We identified more RS sites in the FB neuronal cells than in the undifferentiated H9 cells at the 30 and 120 min time points, even after we downsampled the datasets to the same sequencing depth (S3B Fig). This result is consistent with the previous finding that genes that were highly expressed in neurons tended to have long introns [4, 14]. We performed Gene Ontology enrichment analysis on the genes harboring RS sites and found that they were enriched in ontology terms such as neuron projection guidance and axon guidance (S3C Fig), also consistent with previous findings [4].
We asked whether our pipeline could rediscover the 435 putative RS sites by Sibley et al., which were based on the RS site motif (AGGT) and RS junction reads in their total RNA-seq data [4]. S4A Fig shows the breakdown of these 435 sites: 53% of them are in introns longer than 5 kb, 26% of them pass our evolutionary conservation cutoff, and only twenty sites passed all the criteria in Step 1 of our pipeline, indicating that our pipeline is stricter than that of Sibley et al. Among these 20 putative RS sites, twelve also passed Step 2 of our pipeline and were classified as candidate RS sites. We detected the sawtooth pattern for nine of these ten sites, including six sites for which Sibley et al. also detected the sawtooth pattern (red sites in S1D Fig). Thus, among the 36,870 putative RS sites, 342 candidate RS sites, and 16 sawtooth RS sites identified by our pipeline, 36,850 (99.9%) putative RS sites, 330 (96.5%) candidate RS sites, and 7 (43.8%) sawtooth RS sites are reported for the first time.
We proceeded to computationally validate our candidate RS sites in four ways. First, we asked if our RS sites were supported by junction reads. Searching for junction reads in 806 RNA-seq datasets from the ENCODE project totaling more than 70 billion reads, we found RS junction reads supporting 60% of our putative RS sites and 89% of our candidate RS sites (S4B Fig). Second, we asked whether we could detect enrichment of the branch point motif upstream of the candidate RS sites. Indeed, there is an enrichment for the branch point motif at 20–50 nt upstream the RS sites, but not for randomly selected intronic AGGT sites (Fig 2A). Third, we searched for lariat reads from the two introns flanking each RS site, as such lariat reads provide direct evidence of the recursive splicing (S4C Fig, top panel). Typically, lariat reads are rare, but we did find 266 reads supporting 23 RS intron lariats of 15 RS sites in publicly available RNA-seq data (S4 Table). Nineteen of the 23 identified branch points reside 20–50 nt upstream of the RS sites or their downstream 3′ splice sites, and an A is evident at the branch point (S4C Fig, bottom left panel). We also found 153 lariat reads in these RNA-seq datasets that supported 12 of the 230 putative RS sites in introns (≥ 5k nt) by Sibley et al. (S4C Fig, bottom right panel). Fourth, we asked whether our RS sites could also reconstitute the extended 5′ splice site motif, given that our pipeline only required a strong match to the 3′ splice site motif and the GT dinucleotide in the reconstituted 5′ splice site. Indeed, the reconstituted 5′ splice sites of our candidate RS sites (logo shown in S4D Fig, top) are significantly stronger than the 5′ splice sites of randomly selected intronic sites that start with a GT dinucleotide (median strengths 1.2 vs.−7.6; p-value = 1.42×10−48), albeit weaker than the 5′ splice sites of cassette exons and constitutive exons (Fig 2B). The reconstituted 5′ splice sites of our sawtooth RS sites (logo shown in S4D Fig, bottom) are slightly stronger than the 5′ splice sites of cassette exons and constitutive exons (p-value = 5.71×10−4; Fig 2B). These four analyses provide support for the validity of our predicted RS sites.
We performed RT-PCR and Sanger sequencing experiments to validate a novel RS site in the first intron of the MAGI1 gene, which could be detected in all three cell lines with a clear sawtooth pattern (Fig 2C, top). Using a pair of primers specific for the intermediate product of recursive splicing, we detected a product at the correct size (a 136-bp band in the lane 1 of Fig 2C) and further verified it with Sanger sequencing. This intermediate product was not detected by two primers that annealed to the flanking exons (only a 135-bp band but not a 262-bp band in the lane 2 of Fig 2C).
We compared 4sUDRB-seq with total RNA-seq for their efficacy in finding RS sites. Among the candidate RS sites, we identified in PA1, H9, and FB cells using 4sUDRB-seq data, only 15%, 38%, and 42% could be detected using the steady-state RNA-seq data in the corresponding cell types, which appeared to support our hypothesis that 4sUDRB-seq data were more effective than RNA-seq data for identifying RS sites. However, this result was partly due to that the 4sUDRB-seq datasets totaled more reads than the RNA-seq datasets; thus, we must perform a fair comparison at equal sequencing depth. Given that the Pol II elongation rate is on average 2,520 nt/min, 97% of human genes would have been completely transcribed after 120 mins [8]. Accordingly, we performed detailed comparison by downsampling the 120-min 4sUDRB-seq and steady-state RNA-seq datasets to 80 M reads each, the lowest sequencing depth among all datasets (S3A Fig). Averaged over these three cell lines, we detected roughly the same number of RS sites using the 4sUDRB-seq datasets as using the steady-state RNA-seq datasets (15.3 vs. 14; S5A Fig). We also compared the 4-hr and 16-hr 4sUDRB-seq data with the steady-state RNA-seq data in PA1 cells at the sequencing depth of 80 M. We detected the most RS sites using the 16-hr 4sUDRB-seq data, 1.6 times as many as using the steady-state RNA-seq data (S5B Fig). The majority of the RS junctions detected using the steady-state RNA-seq data were also detected using the 4sUDRB-seq data, and the junctions were supported by comparable numbers of reads in the two types of data (S5C Fig). Taken together, our results indicate that 4sUDRB-seq data with long durations can capture more RS events than steady-state RNA-seq data.
A distinct advantage of time-series 4sUDRB-seq data is that they can provide quantitative information on the temporal progression of recursive splicing. For example, we found three RS sites in the PDE4D gene, two of which (RS1 and RS3) were also reported by Sibley et al. All three RS sites show the sawtooth pattern in PA1 cells, but the pattern emerges at different time points (S5D Fig, left). Pol II transcribes from two independent promoters of PDE4D, and recursive splicing could be detected at as early as 60 min for the promoter-proximal RS1 and RS2 sites, while only after 120 mins for the promoter-distal RS3. Once the sawtooth pattern starts, it persists throughout later time points, supported by junction reads that correspond to the recursive splicing intermediates at all these time points (S5D Fig, right). The sawtooth pattern is weaker at 16 hr than at 2 hr and 4 hr for most of the sawtooth RS sites in PA1 cells (7 out of 12, including PDE4D), suggesting that the optimal time points for detecting the sawtooth pattern are 2 hr and 4 hr.
To characterize the genomic features of human RS sites identified by our pipeline (Fig 1B), we analyzed intron length, evidence for RS exons, GC content, and evolutionary conservation.
We first tested whether RS introns tended to be longer than non-RS introns. A direct comparison is not appropriate because we only searched for RS sites in long introns (≥ 5k nt). As specified in our pipeline (S2 Fig), candidate RS sites are those putative RS sites supported by RS junction reads and sawtooth RS sites are those candidate RS sites that show a sawtooth pattern, and neither of these additional requirements favors longer introns over shorter introns. Therefore, we compare the lengths of candidate RS introns with or without the sawtooth pattern with the lengths of putative introns. We found that candidate RS introns with the sawtooth pattern were significantly longer than the introns without the sawtooth pattern, which were in turn significantly longer than introns with putative RS sites (median lengths 328.9 k, 32.9 k, and 29.7 k; Wilcoxon rank-sum test p-values = 1.38×10−8 and 4.37×10−3 respectively; Fig 3A). Furthermore, 82% of the introns with sawtooth RS sites correspond to the longest introns of their genes, compared with 62% for candidate RS introns without the sawtooth pattern and 55% for putative RS introns (Fig 3B). Although most introns contain just one RS site, introns with two or more RS sites are significantly longer than introns with one RS site (S6A Fig, bottom). These results support the notion that long introns are enriched for RS sites, and recursive splicing is a mechanism aiding the removal of long introns.
RS exons were reported to be essential for the recognition of human RS sites according to the exon definition mechanism [4]; therefore, we next examined whether our predicted RS sites preceded an RS exon. Splice junction analysis (Fig 3C, top; Materials and methods) predicted that 69% of non-sawtooth and 50% of sawtooth RS sites were followed by an RS exon, which was significantly more frequent than those putative RS sites that were not candidate RS sites (14%; Fisher’s exact test p-value = 5.97×10−63; Fig 3C). Furthermore, the regions around the 5′ splice sites immediately downstream of the predicted RS exons are evolutionarily conserved, albeit not to the same extent as canonical exons (S6B Fig), lending support for these RS exons. We also detected multiple potential 5′ splice sites following RS exons, significantly more frequently than following random intronic sites (Wilcoxon rank-sum test p-value = 3.65×10−2; S6C Fig). These additional 5′ splice sites may provide further opportunities to facilitate the splicing of the RS exons.
Most previously identified RS exons are spliced out of the final transcripts, and a competition model was proposed to explain this result: the reconstituted 5′ splice site at an RS site outcompetes the 5′ splice site of the RS exon [4]. Indeed, the median Percent-Spliced-In (PSI) value [15] for our predicted non-sawtooth and sawtooth RS exons are only 3.45 and 10.73, respectively, much lower than that of cassette exons (58.90; Wilcoxon rank-sum test p-value = 2.43×10−37; Fig 3D). Accordingly, 67% of the reconstituted 5′ splice sites at the RS sites are stronger than the 5′ splice site of the corresponding RS exons (Wilcoxon signed-rank test p-value = 8.06×10−10; Fig 3E), supporting the competition model. Both of these two groups of 5′ splice sites are significantly weaker than the cassette exons that match the AGGT RS site motif, indicating that most cassette exons are not RS exons (Wilcoxon signed-rank test p-values = 2.91×10−16, respectively; Fig 3E).
To further investigate the low efficiency of RS exon inclusion, we analyzed the GC contents around RS exons. Short human exons tend to have higher GC% than their long flanking introns. The exon definition model suggests that this differential GC signal assists the spliceosome to recognize the short exon in the vast intronic landscape and accordingly, exons with greater GC differences from their flanking introns are included at higher rates despite the often weaker 5′ and 3′ splice sites surrounding these exons [16]. In sharp contrast to constitutive exons which have substantially higher GC% than their downstream introns, RS exons exhibit the same GC% as their downstream introns, while the exons upstream of RS exons show even more elevated GC% over their reconstituted introns than constitutive exons over their downstream introns (Fig 3F). This substantial difference in GC% between RS exons and their upstream exons (median 2% vs. 8%; Wilcoxon signed-rank p-value = 3.55×10−9) further favors the upstream exons in the competition and accounts for the inefficiency of RS exon inclusion.
Few minor isoforms, if any, can still be detected with RS exons (Fig 3C and 3D). We asked whether these isoforms would be translated into mature proteins or introduce premature termination codons (PTCs), which might result in their rapid degradation via nonsense-mediated decay (NMD) [4]. NMD depends on the exon-exon junction complex (EJC), and an efficient NMD requires a PTC to be located at least 50 nucleotides upstream of the 3′ most exon-exon junction (EEJ) [17, 18]. Indeed, 68% of RS exons would introduce a PTC more than 50 nucleotides upstream of the 3′ most EEJ (S6D Fig), suggesting that NMD is another factor that contributes to the low abundance of RS exons in mature transcripts.
Ample evidence supports the co-occurrence of and interaction between transcription and splicing [7, 19–21]. It is not known whether recursive splicing is kinetically coupled with Pol II elongation, although this is possible because recursive splicing is one of the splicing strategies; thus, we used the time-course 4sUDRB-seq data to test this possibility.
Historically, co-transcriptional splicing is interpreted as splicing that takes place after an intron is transcribed but before the transcription of the gene is completed and the transcript is released, while post-transcriptional splicing takes place after transcription termination. Here, we follow this tradition and define co-transcriptional recursive splicing as the detection of a junction read that spans the recursive splice site (the R1 read in Fig 4A) before the gene is completely transcribed (Fig 4B). Because the 4sUDRB-seq data were sampled at specific time points, to determine whether an RS event was co-transcriptional, we first identified the earliest time point that the RS event could be detected and then compared this time point with the amount of time needed by Pol II to transcribe the entire gene. We used the 4sUDRB-seq data to estimate the transcription rate for each gene, and the averages were 2.52 kb/min for PA1 cells, 2.40 kb/min for FB cells, and 2.28 kb/min for H9 cells, in good agreement with previous estimates [22, 23].
To quantify an RS event, we further defined PRS (Percent Recursive Splicing; Materials and methods) as a function of the junction reads R1 and R2 to evaluate the percentage of transcripts that completed the first step of recursive splicing (Fig 4A). We clustered the 134 RS events in PA1 cells by their earliest time points of onset (indicated by a positive PRS): within 1 hour (class I), 1–2 hours (class II), 2–4 hours (class III), or after 4 hours (class IV; Fig 4C, left panel). For each class, we determined the percentage of RS events that commenced after the gene was completely transcribed (S7A Fig; Materials and methods). Most RS events (121 of 134 in PA1 cells) belonged to classes II to IV, i.e., they commenced at least one hour after Pol II elongation. Surprisingly, for many of these RS introns, the recursive splicing is post-transcriptional (Fig 4C, right panel). For example, for Class II RS events, the RS junction reads started to appear at 60 min, but the transcription of 44% of the corresponding genes was completed in less than 30 mins (Fig 4C, right), indicating that at least 44% of Class II events in PA1 cells were post-transcriptional. Similar results were also observed in Class III: 56% of the events in this class started recursive splicing after the gene was completely transcribed. We applied the same analysis to H9 and FB cells, which had a shorter time course of 4sU labeling than PA1 cells (2 hr versus 16 hr), and observed that at least one-third of RS events underwent post-transcriptional recursive splicing (S7B Fig, bottom).
The sawtooth pattern in RS introns is predicated on the assumption that recursive splicing rapidly follows transcription. However, if the intron is spliced with a long delay after transcription, the sawtooth pattern would be diminished. Thus, we asked whether we were more likely to detect the sawtooth pattern for co-transcriptional RS introns than for post-transcriptional RS introns. Among the 12 sawtooth RS introns in PA1 cells, we classify eight as co-transcriptional and none as post-transcriptional, and we did not have the time resolution to classify the remaining four. Overall, we could detect the sawtooth pattern for 26% of the 31 PA1 co-transcriptional introns, much higher than for the post-transcriptional introns (0%, 0 out of 77, Fisher’s exact test p-value = 2.24×10−5; S7C Fig). These results indicate that the sawtooth pattern is only observed at a subset of RS sites partly because recursive splicing is often a post-transcriptional event.
Given that the onset of recursive splicing lags behind the transcription of RS introns, we asked whether recursive splicing precedes slowly as well. To answer this question, we compared the PRS at each time point for the RS introns in Class II and III. We observed a sharp initial decrease of PRS followed by a slower decline, with the median PRS of Class II RS events halving by the end of the first hour and quartering after two more hours (Fig 4D). The decrease in RPS between the first two time points is significantly higher than the decrease between later time points (S7D Fig). In summary, these results suggested that although most RS events start after the completion of transcription with a considerable delay, they proceed in a timely fashion henceforth.
RS genes have significantly higher Pol II elongation rates than non-RS genes in all three cell lines (S8A Fig; S5 Table). It has been reported that long genes tend to be transcribed more rapidly [24] and RS introns are long (Fig 3A), so we asked whether RS genes have other features of genes with fast Pol II elongation rates.
Multiple histone modifications have been reported to correlate with Pol II elongation. For humans, genes with higher elongation rates show higher densities of H3K79me2 and H4K20me1 throughout their bodies than genes with lower elongating rates [24], while for mouse, H3K79me2 and H3K4me1 are enriched only in the proximal regions around the TSS of fast elongating genes [25]. We observed similar results for RS genes. Among the histone marks assayed in H9-derived neural progenitor cells, H3K79me2, H4K20me1 and H3K4me1 show higher signals at the first 20 kb of RS genes compared with non-RS genes (S8B Fig). We did not observe a higher signal of H3K36me3 in RS genes than non-RS genes (S8C Fig), consistent with previous studies which reported that gene-body H3K36me3 level was correlated with fast Pol II elongation in mice but not in humans [24].
We examined several genomic features known to be correlated with fast Pol II elongation in human [24] and mouse [25]—gene length and first intron length are positively correlated with fast Pol II, while exon density shows the most prominent negative correlation (S8D Fig). Compared with non-RS genes, RS genes in all three cell lines contain significantly longer first intron, lower exon density and more simple repeats (S8D Fig). Furthermore, RS genes have significantly lower GC content (S8D Fig), features shown to slow down Pol II elongation [24]. In summary, our results indicate that RS genes have high Pol II elongation rates and many genomic and epigenetic characteristics that are associated with fast elongation.
It has been proposed that fly RS introns are constitutively spliced out via their RS sites, evidenced by the lack of junction reads that correspond to the lariats of the canonical splicing for these introns [2, 26]. However, it is not known whether recursive splicing is constitutive in humans and conflicting data have been reported: individually mutating three of seven RS sites in the first intron of human SAMD4A gene substantially reduced the splicing efficiency of this intron [26], while blockage of RS site usage in human CADM1 and ANK3 genes did not affect their expression levels [4]. We used the time-series 4sUDRB-seq data to test whether human RS introns were constitutively spliced by the recursive mechanism.
We hypothesized that if recursive splicing were constitutive, we would detect junction reads corresponding to the first step of recursive splicing (R1, Fig 5A) before we detect junction reads corresponding to canonical splicing (R2). In PA1 cells, we detected R1 in an earlier time point than R2 for 22% of the RS events, R1 and R2 at the same time point for 45% of the RS events, and R1 after R2 for 33% of the RS events (Fig 5A). We obtained similar results for H9 and FB cells, detecting R1 at a later time point than R2 for 47% and 18% of the RS events respectively (S9A Fig). For RS events in which R1 and R2 appeared at the same time point, we detected more R2 reads than R1 reads (darker blue in the R2 heatmap than the R1 heatmap; Fig 5A and S9A Fig), suggesting that recursive and canonical splicing occur concurrently for these RS introns although we do not have the time resolution to conclude definitively.
We further searched for lariat reads that could support recursive splicing or canonical splicing of RS introns using publicly available RNA-seq data. Among the 378 RS introns, we could detect reads that traverse the predicted branch point in a lariat configuration (full-length or recursive introns; see Materials and methods) for 17% introns (63 out of 378). For 79% of these 63 introns with lariat-supporting reads, the lariats corresponded to canonical splicing, i.e., they spanned the entire length of the intron subsuming the RS site (Fig 5B), indicating that these introns are removed via the canonical splicing mechanism, i.e., recursive splicing is not constitutive for these introns. We further observed an example of cell-type-specific switching between recursive splicing and canonical splicing. The second intron of the GLI3 gene was recursively spliced in PA1 cells, as supported by both junction reads for recursive splicing and lariat reads for recursive splicing at the 16-hr time point, whereas in FB neurons we could only find junction and lariat reads to support the full-length intron at the 60-min time point, suggesting that this intron is primarily spliced via the canonical mechanism in FB neurons (Fig 5C). Overall, we detected lariat reads that support the full-length intron in other cell lines for 34 PA1, 9 H9, and 15 FB RS introns in cells (S9B Fig). Collectively, these results suggested recursive splicing in humans, unlike in flies, is not mutually exclusive with canonical splicing and the choice between them may be under cell-type-specific regulation.
Recursive splicing is a non-canonical splicing process which removes an intron in multiple steps. It was first identified for a 73-kb intron of the fly Ubx gene [1] and recently shown to act on additional fly [2] and human introns [4]. In particular, Sibley et al. identified 435 putative human RS sites, including 9 RS sites that showed the sawtooth pattern in their steady-state RNA-seq data. We hypothesized that time-course 4sUDRB-seq data might be more effective than steady-state RNA-seq data for identifying RS sites and studying the temporal progression of recursive splicing. Thus, we built a computational pipeline to reanalyze the published 4sUDRB-seq data on three human cell lines [8]. We used the six RS sites by Sibley et al. that had sufficient expression in the 4sUDRB-seq data and did not overlap annotated cassette exons to set strict cutoffs in our pipeline. Analyzing only long introns (≥ 5k nt), we identified 342 candidate RS sites and 16 sawtooth RS sites. These include all 20 of the 435 putative RS sites by Sibley et al. that passed the stringent filtering steps of our pipeline. Thus, our pipeline substantially increased the number of candidate RS sites in long introns. If these criteria were relaxed, we would have predicted even more RS sites in humans. We performed detailed comparison and concluded that indeed 4sUDRB-seq data are more effective than steady-state RNA-seq data in finding RS sites.
We only searched for highly conserved RS sites in long introns that did not overlap cassette exons. If we had not excluded RS sites that overlapped cassette exons, we would have had 3,140 more putative RS sites. Some cassette exons were reported to also function as RS exons [4], but we found that cassette exons tended to have significantly stronger 5′ splice sites and much higher PSI values than RS exons (Fig 3D and 3E). Thus, we conclude that most cassette exons do not also function as RS exons.
In both human and fly, RS exons are crucial to the recognition of RS site [3, 4]. RS sites followed by an RS exon are subsequently spliced out via a competition mechanism[3, 4]. We found several lines of evidence that supported human RS exons and the competition between 5′ splice sites for skipping these RS exons. One feature that we report here for the first time is that the GC% of RS exons are typically as low as the GC% of their downstream introns while their upstream exons are even more GC-rich than canonical exons. Thus, the enlarged GC% difference between an RS exon and its upstream exon promotes the skipping of the RS exon.
Even with time-series 4sUDRB-seq data, we observed the sawtooth pattern for only 16 out of our 342 candidate RS sites, although the sawtooth pattern was consistently observed at multiple time points. This observation prompted us to examine the hypothesis underlying the sawtooth pattern, i.e., recursive splicing rapidly follows transcription. We found that for 57% of the PA1 RS introns, we could detect their junction reads that supported the first step of the recursive splicing only after the entire gene was transcribed, suggesting that these RS introns were recursively spliced post-transcriptionally. This long delay partly explains why we could detect the sawtooth pattern for only a small fraction of the RS introns. Accordingly, among the 12 sawtooth RS introns with the sawtooth pattern in PA1 cells, eight are co-transcriptional, and the other four undetermined due to the lack of time resolution. Thus, unlike canonical splicing, recursive splicing often occurs post-transcriptionally in humans.
Another difference between the fly and humans with regards to recursive splicing is that it has been suggested that recursive splicing is constitutive in the fly [2], but it is not known whether it is so in humans. Again, we searched for RS junction reads that could support recursive or canonical splicing of RS introns in the 4sUDRB-seq data. We also searched for lariat reads that could support recursive or canonical splicing in RNase-R-treated RNA-seq data. Notwithstanding the limited time resolution of the 4sUDRB-seq data, we could detect canonical splicing before recursive splicing for 18–47% of RS introns in the three human cell lines, indicating human RS introns are spliced by both recursive and canonical splicing mechanisms.
Our results raised the question of whether recursive splicing and canonical splicing of long human RS introns are co-regulated. We also showed that genes with RS introns had high Pol II elongation rates. It is tempting to speculate that when canonical splicing cannot catch up with the rapid transcription, recursive splicing is employed as a supplementary mechanism for removing these introns.
In summary, we used time-series 4sUDRB-seq data to expand the collection of long RS introns in humans. We also deduced that many of these RS introns are spliced post-transcriptionally and by both canonical and recursive splicing mechanisms. Our results reveal a complex landscape of RNA splicing and its regulation in the human transcriptome.
We downloaded the 4sUDRB-seq (nascent transcripts) and steady-state ribo− RNA-seq (total RNA after ribosomal RNA removal) datasets derived from PA1, H9 and H9-differentiated forebrain (FB) neuron progenitor cells from the Gene Expression Omnibus (GEO) with the accession GSE73325 [8]. To generate these 4sUDRB-seq data, Zhang et al. first incubated the cells with DRB for three hours to block Pol II elongation, and then labeled the newly transcribed RNAs with 4sU upon DRB release. At each time point after the synchronized start of transcription elongation, Zhang et al. extracted 4sU-labeled RNAs to perform RNA sequencing. Total RNA (100–140 μg) was used for biotinylation and purification of 4sU-labeled nascent RNAs. Corresponding 4sUDRB-seq samples with RNase R treatment were also downloaded from the same GEO entry to facilitate intron lariat analysis. The 4sUDRB-seq datasets of each cell line contain five time points of 4sU exposure (10, 15, 30, 60, and 120 min). Also, the PA1 4sUDRB-seq datasets include two more 4sU exposure durations (4 and 16 hr). As previously reported [8], these datasets have a low bias for the collection of nascent RNAs and high efficiency in measuring transcription elongation rates. We mapped the sequence reads from each sample using the HISAT2 tool [27] (Version 2.0.4) against the GRCh37/hg19 human reference genome and annotated the mapped reads with the UCSC Known Genes annotation (hg19 knownGene.txt updated on June 30, 2013).
To compare the performance of 4sUDRB-seq and ribo− RNA-seq data in identifying recursive splicing (RS) sites without being affected by sequencing depth, we randomly selected the same number of reads for each dataset and ran our RS site identification pipeline and downstream analysis with the same parameters (S3B and S5 Figs).
Previous studies revealed that RS sites have four genomic features [2, 4]. First, they tend to be in long introns. Second, they contain a sequence motif (AGGT), which is like the concatenated 3′ and 5′ splice site consensus motifs. Third, they are recognized by the spliceosome as a 3′ splice site in the first step of recursive splicing, so RS sites typically score well as 3′ splice sites. Fourth, known human and fly RS sites have high evolutionary conservation. To accurately identify RS sites genome-wide, we first compared the nine known human RS sites [4] and canonical splice sites on these four genomic features to determine the suitable cutoffs for our pipeline. Half of the 22,144 human annotated genes (hg19 knownGene.txt updated on June 30, 2013) have their median-length introns more than 1,928 nt long (S1A Fig, left) and their longest introns more than 10,090 nt long (S1A Fig, right). Thus, we chose 5 kb as the cutoff for defining long introns to focus our analyses on. Splicing strength analysis revealed that all the nine reported human RS sites have strong (≥ 5) splicing strength when evaluated as 3′ splice sites, which are even stronger than constitutive exons and cassette exons (S1B Fig). Also, these nine human RS sites are as conserved as canonical splice sites (conservation score ≥ 0.5) (S1C Fig). Taking into account these pieces of information, our pipeline contains three steps (S2 Fig), described as follows.
We compared introns harboring RS sites (called RS introns) with other introns (≥ 10k nt) that do not harbor RS sites (non-RS introns) in terms of length (Fig 3A).
We predicted RS exons based on junction reads detected in the 4sUDRB-seq datasets with the following steps. First, we used 4sUDRB-seq reads to identify all possible exon-exon junctions that could result in an RS exon up to 500-nt long (the 500-nt parameter was chosen based on the knowledge that 87% annotated human exons are shorter than 500 nt). We further required that there were at least two unique reads that mapped to the junction. If there were more than one putative 5′ splice site, the 5′ splice site with the most junction reads was defined as the 5′ splice site of the RS exon (Fig 3C). We estimated the Percent-Spliced-In (PSI) value [15] for each predicted RS exon, i.e., the frequency that the RS exon could be included in mature transcripts, using the steady-state RNA-seq data (Fig 3D).
When we computed the differential GC content between neighboring exons and introns (Fig 3F), we excluded the ±5 bp window around each splice site from the calculation, because we did not want to be biased by the splice site motifs.
We identified reads that map to intron lariats and pass over the branch point using a strategy similar to previously described [28]. Briefly, we first mapped all RNA-seq reads to the human genome (hg19) using STAR [29] and only retained the unmapped reads. We then built an index using the first 23 nt sequence of each annotated intron (GENCODE release 19 supplemented with the introns flanking our RS sites) and mapped the unmapped reads to these 23-nt intron prefixes using Bowtie2 [30]. The portion of a read that maps to an intron prefix is defined as the A portion, and its upstream sequence is defined as the B portion. In the case of a lariat read, the B portion should map to the sequence upstream of the branch point of the intron that A portion belongs to. We required the B portion to be > 23 nt and mapped them to the human genome using Bowtie2, and only kept alignments with mapping qualities (MAPQ) ≥ 10. We then required that the A and B portions of the same read to be aligned to the same intron. The 3′ end of each B portion indicated the putative branch point. To filter out spurious alignments that might have risen from sequence errors or repetitive regions of the genome, we then reconstructed 200-nt putative lariat-template sequences (100 nt upstream of each putative branch point and the first 100 nt of its intron). We mapped these putative lariat-template sequences to the human genome and only retained those with ≤ 80% sequence identity for further analyses. To obtain the final annotation of branch points and their supporting reads, we aligned all reads (unmappable to the genome) to such lariat-template sequences and retained alignments that had at most one mismatch and ≥ 23 nt overhangs flanking the branch point. Finally, we only kept those branch points that were near (<100 nt) 3′ splice sites or RS sites (S4C Fig; Fig 5B).
Splice site strength was calculated using MaxEntScan [11]. To estimate the splicing strength of reconstituted 5′ splice site upon recursive splicing (Figs 2B and 3E), we concatenated the last three nucleotides of the upstream exon with the six nucleotides downstream a candidate RS site and used the resulting sequence for the calculation. RS site motifs (S4D Fig) were created using WebLogo3 [31]. Branch points (Fig 2A) were predicted using SVM-BPfinder [32]. Consensus 5′ splice sites (S6C Fig) were the sites that matched GTAAG, GTGAG, GTAGG, GTATG, GTAAA, GTAAT, GTGGG, GTAAC, GTCAG, or GTACG.
Gene Ontology (GO) analysis was performed using the DAVID tool (https://david.ncifcrf.gov/, [33]), and GO terms from GOTERM_BP_FAT and GOTERM_MF_FAT were used. GO terms with multiple-testing-corrected p-values [34] lower than 0.05 were deemed enriched (S3C Fig).
A549 cells were cultured as previously described [35]. Total RNA was purified using the RNeasy Mini Kit (Qiagen). cDNA was synthesized using SuperScript (ABI) and used as the template in RT-PCR. LA-Taq (Clontech) PCR kit was used for RT-PCR. PCR bands were gel purified (Qiagen), TOPO cloned (Invitrogen TOPO kit), and sent for sequencing (Genewiz). PCR primers are listed in S6 Table.
To facilitate quantitative analysis of the recursive splicing process during Pol II transcription elongation, we define the PRS metric (Percent Recursive Splicing; Fig 4A) as the ratio between the number of RS junction reads (R1) and the total junction reads (R1 + R2). Defined as such, PRS quantifies how complete the first step of recursive splicing is. PRS first increases, indicating that the first step of recursive splicing starts to occur, and then it decreases, indicating that recursive splicing transitions from the first step to the second step. To simplify our analysis of the temporal progression of recursive splicing and decrease false positives (Fig 4 and S7 Fig), we only included the RS introns that had only one RS site and further met the following criteria:
We estimated the Pol ll transcription elongation rate (TER) for each gene in each sample using the TERate tool [8] (https://github.com/YangLab/TERate). We estimated the rates from the two datasets with 10 or 15 min of 4sU labeling after DRB removal. For those genes without an estimated elongation rate at both time points due to being too short or having too few reads, an average Pol II elongation rate of 2,520 nt/min was used.
To determine whether a recursive splicing event occurred co-transcriptionally or post-transcriptionally, we compared the first time point of detecting a junction read that spanned the RS site with the time required for the complete transcription of the entire gene, using the estimated elongation rate for the gene (Fig 4B). Note that we can only score the final ligation product of recursive splicing and this ligation step may be delayed relative to the determination of splice site pairs, we may underestimate the number of co-transcriptional splicing. Due to the limited time resolution of the 4sUDRB-seq experiments (sampled at 10, 15, 30, 60, 120, 240, 960 mins), we could not classify all RS events into the co- or post-transcriptional classes. For example, if the gene is completely transcribed in 80 mins and we detected the onset of recursive splicing (indicated by a positive PRS) in the 120-min sample (i.e., not in any of the earlier samples), this recursive splicing event could have occurred before or after the entire gene was transcribed at 80 mins (we only know that it was after 60 mins and before 120 mins). To classify RS events into co- and post-transcriptional categories unambiguously, we tabulated the precise relationship between the onset time point of a positive PRS and the time point when the entire gene was transcribed (S7A Fig).
Genomic features, including exon density, length of the first intron, the fraction of simple repeats, and GC content (S8D Fig), were calculated based on the UCSC Known Genes annotation (hg19 knownGene.txt updated on 2013/6/30) and RepeatMasker annotation (hg19 rmsk.txt.gz updated on 2009/4/27). The ChIP-seq data of several histone modifications on neural progenitor cells derived from the H9 embryonic stem cells were downloaded from the ENCODE Portal (https://www.encodeproject.org/; accessions for H3K79me2: ENCFF905UHS, H4K20me1: ENCFF560PIQ, H3K4me1: ENCFF637CXU, and H3K36me3: ENCFF555VCB). We plotted normalized ChIP-seq signals, i.e., fold change over control, for RS introns in S8B and S8C Fig.
The source code for our RS site identification pipeline and downstream analyses can be accessed from the GitHub (https://github.com/kepbod/rs).
|
10.1371/journal.ppat.1005500 | PSGL-1 on Leukocytes is a Critical Component of the Host Immune Response against Invasive Pneumococcal Disease | Bacterial uptake by phagocytic cells is a vital event in the clearance of invading pathogens such as Streptococcus pneumoniae. A major role of the P-selectin glycoprotein ligand-1 (PSGL-1) on leukocytes against invasive pneumococcal disease is described in this study. Phagocytosis experiments using different serotypes demonstrated that PSGL-1 is involved in the recognition, uptake and killing of S. pneumoniae. Co-localization of several clinical isolates of S. pneumoniae with PSGL-1 was demonstrated, observing a rapid and active phagocytosis in the presence of PSGL-1. Furthermore, the pneumococcal capsular polysaccharide and the main autolysin of the bacterium ―the amidase LytA― were identified as bacterial ligands for PSGL-1. Experimental models of pneumococcal disease including invasive pneumonia and systemic infection showed that bacterial levels were markedly increased in the blood of PSGL-1−/− mice. During pneumonia, PSGL-1 controls the severity of pneumococcal dissemination from the lung to the bloodstream. In systemic infection, a major role of PSGL-1 in host defense is to clear the bacteria in the systemic circulation controlling bacterial replication. These results confirmed the importance of this receptor in the recognition and clearance of S. pneumoniae during invasive pneumococcal disease. Histological and cellular analysis demonstrated that PSGL-1−/− mice have increased levels of T cells migrating to the lung than the corresponding wild-type mice. In contrast, during systemic infection, PSGL-1−/− mice had increased numbers of neutrophils and macrophages in blood, but were less effective controlling the infection process due to the lack of this functional receptor. Overall, this study demonstrates that PSGL-1 is a novel receptor for S. pneumoniae that contributes to protection against invasive pneumococcal disease.
| S. pneumoniae is one of the most important and devastating human pathogens worldwide, mainly affecting young children, elderly people and immunocompromised patients. In terms of host immune defense against invasive pneumococcal isolates, professional phagocytes require receptor-mediated recognition of certain ligands on the bacterial surface for the uptake and clearance of the microorganism. In this study, we demonstrate that the P-selectin glycoprotein ligand-1 (PSGL-1) on leukocytes is involved in the phagocytosis process of S. pneumoniae by targeting the capsule and the surface protein LytA as pathogen-associated molecular patterns. To explore this process in more detail, we have used wild-type mice and mice deficient in PSGL-1 demonstrating that lack of PSGL-1 is detrimental for the host by increasing the susceptibility to the infection and the severity of the pneumococcal invasive disease. Overall, these data show the importance of PSGL-1 on leukocytes in host defense against S. pneumoniae and confirm that PSGL-1 plays a critical protective role against invasive bacterial disease.
| Streptococcus pneumoniae (pneumococcus) is one of the major causes of invasive disease accounting for more deaths than any other vaccine-preventable bacterial infection. This microorganism colonizes the human nasopharynx, being one of the leading causes of acute otitis media, community-acquired pneumonia and invasive pneumococcal disease (IPD) including sepsis and meningitis [1]. The World Health Organization estimates that nearly 14 million episodes of serious pneumococcal disease occur every year with a critical impact in childhood population as pneumonia kills more children than AIDS, malaria and measles combined [1, 2].
Resolution of pneumococcal disease is regulated by the efficient recognition and clearance of the invading pathogen by professional phagocytes [3]. Leukocytes play an important role in inflammatory and immune responses against bacterial infection and bacterial clearance depends on the efficacy of different receptors on phagocytic cells to recognize, internalize and kill the pathogen [4–7]. Phagocytic receptors on the cell surface trigger phagocytosis following direct recognition of particulate targets. Interaction between selectins and selectin-ligand molecules is essential for the host-pathogen encounter due to its crucial role in leukocyte extravasation [8]. In this sense, expression of P-selectin and E-selectin by the endothelium provides protection against invading pathogens such as S. pneumoniae [9, 10]. P-selectin glycoprotein ligand-1 (PSGL-1) on leukocytes mediates interactions with P-selectin and E-selectin expressed by endothelial cells [11]. PSGL-1 is a homodimeric mucin-like glycoprotein expressed on the surface of almost all circulating leukocytes with a great importance in leukocyte adhesion and transmigration as it is responsible for the initial steps of the extravasation cascade [8, 12]. However, certain intracellular pathogens have developed sophisticated strategies exploiting specific receptors for their own benefit to enter eukaryotic cells and replicate intracellularly [13]. This is the case of the obliged intracellular pathogens Anaplasma phagocytophilum, Ehrlichia sp., and enterovirus 71 that get access inside the cell by binding PSGL-1, causing granulocytic anaplasmosis/ehrlichiosis and hand-foot-mouth disease respectively [14–16]. However, there is no experimental evidence indicating that PSGL-1 could act as a receptor on leukocytes participating in the recognition and clearance of extracellular invading pathogens such as S. pneumoniae. In this sense, the main goal of this study was to investigate the protective contribution of PSGL-1 in host defense against IPD.
The plasma membrane of phagocytes expresses an array of receptors that interact with specific microbial ligands promoting the internalization and clearance of the potential pathogen. To evaluate the impact of PSGL-1 in pneumococcal phagocytosis, HL-60 cells differentiated to granulocytes were used as target cells because they express the same phagocytic receptors as peripheral blood neutrophils including PSGL-1 [14–17] (S1 Fig). To assess the role of PSGL-1 in the phagocytosis of S. pneumoniae, the receptor function was blocked using the specific monoclonal antibody KPL-1. This is an accepted method to assess the impact of PSGL-1 in microbial interaction [14–16]. To determine the generic role of PSGL-1 in host defense against this important pathogen, clinical isolates of S. pneumoniae belonging to different serotypes were assessed. Phagocytosis was significantly impaired when PSGL-1 was blocked, indicating that pneumococcal phagocytosis is more efficient when this receptor is fully active (Fig 1A and 1B). The contribution of FCγ-receptors was evaluated indicating that the effect of PSGL-1 in phagocytosis is independent of FCγ-receptors activity (S1 Fig). In addition, bacterial killing mediated by PSGL-1 was examined using three different clinical isolates. Our results showed that pneumococcal survival increased when PSGL-1 in phagocytic cells was blocked, demonstrating that this receptor is involved in the clearance of S. pneumoniae (Fig 1C). Finally, phagocytosis experiments using neutrophils obtained from the spleen of wild-type and PSGL-1–/–mice by cell sorting, confirmed that PSGL-1 is involved in the phagocytosis of S. pneumoniae (Fig 1D and 1E).
To further analyze the kinetics of pneumococcal phagocytosis, cells with PSGL-1 ―either active or antibody-blocked― were infected with the D39 (serotype 2) strain expressing the green fluorescent protein (GFP), and the phagocytosis process was monitored using live imaging confocal microscopy (Fig 2A and 2B, S1 and S2 Movies). When PSGL-1 was active, a rapid and active phagocytosis was observed, with the majority of the cells containing pneumococcal cells by the end of the process (Fig 2A and S1 Movie). However, when PSGL-1 receptor was blocked, the recognition and engulfment of S. pneumoniae was impaired, which confirmed the importance of PSGL-1 in pneumococcal phagocytosis (Fig 2B and S2 Movie). To confirm the interaction of S. pneumoniae with PSGL-1, fluorescently-labeled pneumococcal isolates of serotypes 2, 3, 6B, 11A, 23F and 19A were used to observe co-localization with PSGL-1 (Fig 3 and S2 Fig). Hence, our findings show that PSGL-1 is involved in the recognition and phagocytosis of a major human pathogen such as S. pneumoniae contributing therefore, to the variety of receptors on professional phagocytes that are needed to efficiently identify and destroy invading pathogens [4–6, 18].
Phagocytosis requires receptor-mediated recognition of microbial ligands that are usually expressed in the surface of invading pathogens. These ligands are frequently known as pathogen-associated molecular patterns (PAMPS) which are recognized by specific receptors of the innate immune system [19]. As LytA ―the main cell wall hydrolase of S. pneumoniae― is located on the bacterial surface and it is essential to interact with critical components of the host immune response including neutrophils and macrophages [20, 21], we explored the possible interaction of PSGL-1 with LytA. Pneumococcal recognition by PSGL-1 was hindered in the absence of LytA, suggesting that LytA might be a bacterial ligand for PSGL-1 (Fig 4A and 4B). Direct interaction between purified LytA and PSGL-1 molecules was observed confirming that this receptor recognizes LytA (Fig 4C). This interaction was dependent on the concentration of PSGL-1, suggesting that recognition of S. pneumoniae is enhanced when PSGL-1 levels are increased. Preincubation of HL-60 cells with purified LytA before infection reduced the phagocytosis in a similar way than KPL-1 antibody, supporting additional evidence that LytA interacts with PSGL-1 (S3 Fig). To confirm the interaction of LytA and PSGL-1 we included the non-capsulated strain (M11) and an isogenic lytA mutant strain (Fig 4D). Our results demonstrated that in the absence of LytA, the binding of S. pneumoniae to PSGL-1 is impaired. Although the capsular polysaccharide (CPS) is one of the major virulence factors of S. pneumoniae, resistance to phagocytosis can vary with the capsular type, which might explain differences of invasiveness among strains [22, 23]. Capsule recognition by PSGL-1 was investigated using a non-capsulated strain (M11) and several isogenic transformants of M11 expressing different CPSs (Fig 4E). The absence of CPS caused impaired recognition by PSGL-1, in comparison to the corresponding encapsulated transformants, confirming that PSGL-1 recognizes the pneumococcal CPS (Fig 4E). However, different levels of binding were observed depending on the CPS with the weakest recognition for the pneumococcal strain expressing serotype 19A (Fig 4E). Experiments including purified CPS of type 3 and different concentrations of PSGL-1 were included to investigate the binding of PSGL-1 to the pneumococcal capsule (Fig 4F). This CPS was assessed because is included in the current PCV-13 vaccine and clinical isolates of serotype 3 are a major cause of IPD [24]. Direct recognition of purified CPS by PSGL-1 was observed showing a concentration-dependent pattern, which confirms that PSGL-1 recognizes the capsule of S. pneumoniae (Fig 4F).
The development of IPD depends of the complex interplay of many factors including virulence determinants of the pathogen and the efficacy of the host immune response to clear the infection process. A failure to efficiently detect and destroy S. pneumoniae in the lower respiratory tract or the systemic circulation lead to severe pneumonia or disseminated infection which are associated to increased mortality rates [3]. Next, the protective role of PSGL-1 against IPD was investigated using pneumonia and sepsis models of infection (Fig 5). In pneumonia, bacterial counts were slightly higher in the bronchoalveolar lavage fluid (BALF) and lung of PSGL-1−/− mice, and much higher in the blood of KO mice (Fig 5A). These bacterial levels were markedly elevated at 24 h in the blood of PSGL-1−/− mice (with the progression of the infection), confirming that PSGL-1 contributes to control bacterial load by reducing the severity of pneumococcal dissemination (Fig 5A). In the sepsis model, PSGL-1−/− mice had greater levels of bacteria in blood (Fig 5B). In addition, lethal infection developed faster in PSGL-1−/− mice than in wild-type mice indicating that PSGL-1 plays a critical role in host defense against IPD by controlling bacterial infection in the systemic circulation (Fig 5B and5C). To confirm this hypothesis and exclude the contribution of cellular migration mediated by PSGL-1, mice were directly infected by the intravenous route (Fig 5D). Mice lacking PSGL-1 had increased levels of bacteria in blood than wild-type mice both at 6 and 24 h confirming that a major function of PSGL-1 in host defense is to clear the bacteria in the bloodstream controlling the dissemination (Fig 5D). To extend the importance of PSGL-1 in the clearance of S. pneumoniae from the systemic infection, a sepsis model was repeated using a lethal dose of a different serotype such as TIGR4 strain. Our findings corroborated the results obtained with the D39 strain demonstrating that lack of PSGL-1 was associated with increased bacterial counts in blood and a more severe infection compared to wild-type mice (Fig 5E and 5F). Overall, our findings demonstrate that PSGL-1 plays an important role against IPD.
The inflammatory response to infection with S. pneumoniae in PSGL-1−/− and wild-type mice was characterized in BALF and serum. The pattern of the major cytokines associated to infection was similar in BALF (Fig 6A), although in serum of PSGL-1−/− mice there were significant increased levels of IL-5, IL-6, and IFN-γ (P <0.05) which is compatible with the higher bacterial levels found in the blood of these mice (Fig 6B).
Neutrophils predominate within cellular infiltrates in pneumococcal pneumonia and the consequences of the neutrophil influx for the host can be advantageous or detrimental, depending on the degree of cellular influx and the ability of the pathogen to successfully avoid the immune response [3]. Leukocyte infiltration into lungs and circulating leukocytes in blood, were measured by flow cytometry, using wild-type and PSGL-1−/−mice infected with S. pneumoniae D39 strain (Fig 7A and 7B). In the pneumonia model (intranasal inoculation), the number of neutrophils and macrophages were similar in the lungs of both types of mice whereas T cell counts were higher in PSGL-1−/−mice (Fig 7A). In a systemic model of infection (intravenous inoculation), however, the number of T cells were similar, although the proportion of neutrophils and macrophages were significantly higher in PSGL-1−/−mice (Fig 7B), which is compatible with the higher levels of bacteria in the blood of these mice (Fig 5D). Overall, these results indicate that PSGL-1−/−mice, despite having greater numbers of leukocytes in blood, had an impaired ability to clear the bacteria from the bloodstream confirming the importance of PSGL-1 in the recognition and killing of S. pneumoniae in the systemic circulation.
Immunohistochemical characterization of thin sections from lung tissues confirmed that mice deficient in PSGL-1 had greater infiltration of T cells and neutrophils compared to wild-type mice, which are consistent with the severity of the infection process developed in PSGL-1−/− mice (Fig 7C).
To demonstrate that pneumococcal LytA is involved in the physiological effects mediated by PSGL-1, phagocytic assays were performed using HL-60 cells exposed or not to the KPL-1 antibody and a pneumococcal strain lacking LytA (Fig 8A and 8B). Hence, our findings confirmed that phagocytosis of a LytA deficient strain is not dependent on PSGL-1 on HL-60 cells (Fig 8A and 8B). To confirm the relevance of LytA in the interaction with PSGL-1, pneumonia and sepsis models of infection were repeated in wild-type and PSGL-1−/− mice using a D39 lytA deficient strain (Fig 8C and 8D). In contrast to mice infected with wild-type D39, there were no differences in lung or BALF CFU between PSGL-1−/− and wild-type mice infected intranasally with the D39 lytA mutant strain. Similarly, in the sepsis model there were no differences in recovered D39 lytA deficient strain CFU between PSGL-1−/− and wild-type mice. These results confirm that the interaction of LytA with PSGL-1 is important for innate immunity against S. pneumoniae. (Fig 8C and 8D).
Neutrophils are key players in the innate and adaptive immune responses to microbial cells, since they are critical for rapid clearance of invading bacteria [25, 26]. For this process, neutrophils must first detect the microorganisms using surface receptors that are essential to engulf and kill the pathogen [18, 27]. PSGL-1 is a ligand of P-, E- and L-selectins, and is able to mediate the tethering and rolling of circulating leukocytes on the activated endothelium prior to their extravasation [8, 11, 12]. The role of P, E and L-selectin as well as certain integrins against pneumococcal infection has been previously characterized [9, 10, 28, 29]. However, the direct role of PSGL-1 on leukocytes in host defense against S. pneumoniae including its contribution to the recognition and clearance of this microorganism is unknown. As this bacterium is highly variable with up to 96 serotypes described so far [30], we included different serotypes of S. pneumoniae to investigate the role of PSGL-1 in phagocytosis. Hence, our findings showed that PSGL-1 is involved in the recognition and phagocytosis of a major human pathogen such as S. pneumoniae contributing therefore to the variety of receptors on professional phagocytes that are needed to efficiently identify and destroy invading pathogens [4–6, 18]. For the detection of the pathogen it is necessary the interaction of phagocytic receptors with bacterial ligands that are usually exposed on the surface of the microorganism [19]. Using LytA-deficient mutants, we have recently demonstrated that this surface-exposed protein is a major determinant in the virulence of S. pneumoniae by interacting with essential components of the immune system including phagocytes [20]. Our results show now that the LytA autolysin is recognized by PSGL-1 and this effect is dependent on the level of PSGL-1 suggesting that variations in the expression of PSGL-1 on leukocytes might contribute to the efficiency of this interaction. One of the major concerns in the outcome of the infection is that pneumococcal disease can be produced by clinical isolates of a wide variety of polysaccharide capsules, the main virulence determinant of S. pneumoniae [22, 23]. In this sense, using pneumococcal transformants expressing the same genetic background but different CPS, we have demonstrated that PSGL-1 recognizes the capsule of S. pneumoniae. This effect was variable depending on the CPS expressed, with the lowest binding related to the strain expressing serotype 19A. In this sense, incidence of IPD cases caused by this serotype has dramatically increased in the last few years and it has been linked to the emergence of vaccine escape variants that arise by switching the capsular locus from serotype 4 to 19A [31, 32]. These results are important in terms of pathogenesis because differences in the recognition of pneumococcal CPS by receptors such as PSGL-1, might explain why certain serotypes of S. pneumoniae are more associated to IPD and dissemination worldwide than others [31–33].
The repertoire of host receptors involved in the binding, uptake, signaling and response to invading pathogens is critical for the outcome of the infection. PSGL-1 is the main selectin receptor involved in neutrophil adhesion and migration [8]. Although the importance of leukocyte extravasation is relevant in pneumococcal infection [9, 10, 29], the contribution of PSGL-1 to the resolution of IPD was previously unknown. In this study we have used PSGL-1−/−mice to investigate the in vivo role of this receptor in host defense against pneumococcal infection. Our findings confirm that PSGL-1 may act as a pathogen-recognition receptor of the immune system [34]. In this sense, PSGL-1 acts in host defense controlling bacterial proliferation, dissemination and tissue injury which are critical aspects of IPD. This is of great relevance from the respiratory perspective as invasive pneumonia caused by S. pneumoniae is one of the major causes of mortality in children and adults [1, 3]. Lack of PSGL-1 has been linked to a greater susceptibility against the intracellular pathogens Salmonella typhimurium or Citrobacter rodentium, which is associated with dramatically increased levels of pro-inflammatory cytokines after intestinal infection [35, 36]. In the case of S. pneumoniae, bacterial recognition by immune cells generates an array of cytokines which may play a significant role in host defense. Increase of IL-5, IL-6 and IFN-γ was observed in PSGL-1–/–mice sera. According to our findings, it has been described that increased serum levels of IL-5 and IL-6 were associated with reduced microbial clearance and higher mortality rates in sepsis [37, 38]. In addition, the increased levels of IL-12 and IFN-γ in the lungs of wild-type mice might be protective as IFN-γ is involved in the IL-12 regulation of neutrophil-mediated host defense against pneumococcal pneumonia [39].
Bacterial infections of the lower respiratory tract are characterized by massive accumulation of neutrophils in the alveolar spaces [3, 40]. Our results show that similar numbers of neutrophils and macrophages were observed in the lungs of PSGL-1−/− mice. Our results might be unexpected as PSGL-1 is involved in cellular migration of these cells and therefore, increased numbers should be present in the lungs of wild-type mice [8]. Hence, our findings can be explained in the context of a bacterial infection as certain pathogens ―including S. pneumoniae― have the ability to impair neutrophil migration to the site of infection by cleaving PSGL-1 [41, 42]. Interestingly, this is in line with previous findings confirming that the absence of endothelial selectins E, L and P is not associated with impairment of leukocyte emigration to infection sites after pneumococcal disease [10, 43]. In addition, PSGL-1 negatively regulates CD4+T cell immune responses in vivo which can explain the increased levels of T cells observed in the lungs of PSGL-1−/− infected mice [44].
Using an intravenous infection model when no collateral migration effects were expected, higher numbers of neutrophils and macrophages but not T cells, were observed in PSGL-1−/− mice. This is consistent with previous observations showing that blood of PSGL-1−/− mice contained similar numbers of lymphocytes although they had a significant increase in the proportion of leukocytes, with enhanced levels of granulocytes and monocytes in comparison to wild-type mice [45]. This is relevant from the phagocytosis perspective as PSGL-1−/− mice, despite having increased numbers of these phagocytic cells in the bloodstream, were unable to control bacterial replication in the blood, leading to the rapid development of fatal infection. Overall, these results confirm that PSGL-1 on leukocytes plays a critical role in host defense against pneumococcal infection. As a consequence of pathogen-recognition by PSGL-1, S. pneumoniae is efficiently engulfed and destroyed, reducing bacterial replication and dissemination in the host, contributing to control the severity of IPD.
All the experiments involving the use of animals have been performed following the guidelines of the Bioethical and Animal Welfare Committee of Instituto de Salud Carlos III (ISCIII) that reviewed and approved protocol CBA PA 52-2011-v2, to be performed at the National Centre for Microbiology of ISCIII. Animals were bred at Universidad Autónoma de Madrid animal facility following institutional guidelines for animal use and care. Infection experiments conformed to the Spanish government legislation (RD 1201/2005) and European Community regulations (86/609/EEC).
The S. pneumoniae strains used were D39 (NCTC 07466, serotype 2), TIGR4 (ATCC BAA-334, serotype 4), and clinical isolates of different serotypes obtained from the Spanish Pneumococcal Reference Laboratory; 957/12 (serotype 3), 1515/97 (serotype 6B), 450/12 (serotype 11A), 3347/12 (serotype 19A), 69 (serotype 19F) and 48 (serotype 23F). The non-encapsulated S. pneumoniae strain M11 and their isogenic transformants expressing CPS of serotypes 3, 6B and 19A were also included in this study [46]. S. pneumoniae D39 strain expressing the GFP was constructed by transformation with plasmid pMV158GFP [47] and used for confocal microscopy experiments. S. pneumoniae strains were cultured on blood agar plates at 37°C in a 5% CO2 atmosphere, or in Todd-Hewitt broth supplemented with 0.5% yeast extract, to an optical density at 550 nm (OD550) of 0.5, and stored at −70°C in 10% glycerol as single use aliquots.
HL-60 cells (CCL-240; ATCC) differentiated to granulocytes were used and the general conditions of the assay were based on those described previously [17, 48]. Briefly, S. pneumoniae strains were fluorescently labeled by incubation with FAM-SE (Molecular Probes) in 0.1 M sodium bicarbonate buffer for 1 h at 37°C, washed five times with Hanks balanced salt solution (HBSS) and stored at −70°C in 10% glycerol as aliquots for further assays. HL-60 cells were harvested by centrifugation and washed twice with HBSS and once with HBSS in the presence of calcium and magnesium ions. Infection assays were performed in the absence of serum to avoid complement-dependent phagocytosis and 105 HL-60 cells were infected with 2×106 colony forming units (CFU) of viable FAM-SE labeled bacteria. To block PSGL-1, HL-60 cells were incubated for 1 h at room temperature with 25 μg/ml of the KPL-1 antibody (mouse anti-human PSGL-1; MBL) or IgG isotype control (mouse anti-human IgG; Novus Biologicals) as previously described [15, 16]. A similar approach was performed using purified LytA. A minimum of 6,000 cells were analyzed using a FACS Calibur flow cytometer (BD Biosciencies). Using cytochalasin D, an inhibitor of actin polymerization, we have previously shown that the majority of the effect on the association of fluorescent S. pneumoniae with HL-60 cells is due to phagocytosis rather than to adhesion of bacteria to the cell surface [49]. Results were expressed as a fluorescence index defined as the proportion of positive cells for fluorescent bacteria multiplied by the geometric mean of fluorescence intensity which correlates with the amount of bacteria phagocytosed per cell [48].
Opsonophagocytosis killing assays were performed in the absence of serum using 105 HL-60 cells with the PSGL-1 receptor either active or blocked as mentioned above and 2.5 × 104 CFU/ml of S. pneumoniae as previously described [50]. Serial dilutions of culture supernatants were plated on blood agar plates for bacterial counts determination and results were expressed as bacterial survival after 45 min incubation of the pneumococcal strains with HL-60 cells expressing (or not) PSGL-1.
Phagocytosis assays were repeated using neutrophils purified from the spleen of wild-type and PSGL-1−/− mice by FACS using a FACSAria I (BD Biosciences) dispositive with DIVA version 6.1 software as previously described [51]. Briefly, single-cell suspensions were prepared in staining buffer (2% fetal calf serum in PBS), and non-specific binding was blocked with Fc block (BD Biosciences). Staining was performed using standard protocols with the following antibodies diluted 1/200 in staining buffer including propidium iodide (rat anti-mouse CD11b-allophycocyanin (APC) and rat anti-mouse GR-1-phycoerythrin; Biolegend).
S. pneumoniae strains labeled with FAM-SE were used for immunofluorescence microscopy. HL-60 cells previously infected as described above were seeded on 12-mm circular coverslips for immunofluorescence staining. As HL-60 cells are in suspension, cells were cytofuged at 70 × g for 2 min using a Cytospin centrifuge (Thermo Electron, Pittsburgh, PA), as described elsewhere [16]. For the detection of PSGL-1 in HL-60 cells differentiated to granulocytes, cells were fixed with 3% paraformaldehyde (PFA) for 10 min at room temperature and after two washes with PBS, coverlips were kept in a solution 1 M NH4Cl-PBS solution. Coverslips containing the infected cells were washed twice in PBS containing 0.1% saponin and once in PBS and incubated for 30 min with the primary antibody. Staining was performed in PBS containing 10% horse serum, 0.1% saponin and the primary antibody using a mouse anti-human PSGL-1 monoclonal antibody (KPL-1; MBL) diluted 1/300. Cellular DNA was stained with Hoechst (Invitrogen) diluted 1/2500. After 30 min incubation with the primary antibody at room temperature, coverlips were washed twice with PBS-saponin 0.1%, and once with PBS pH 7.0 before incubation during 30 min at room temperature with a dilution 1/200 of the secondary antibody (goat anti-mouse Texas Red; Serotec). Finally, coverslips were washed twice in PBS containing 0.1% saponin, once in PBS, and once in H2O, mounted with Aqua Poly/Mount (Polysciences), and analyzed with a Leica spectral SP5 confocal microscope using the Leica software (LAS-AF).
Binding of PSGL-1 to S. pneumoniae, purified LytA or CPS was analyzed by ELISA as previously described [20]. Briefly, whole cell ELISA was performed by coating 96-well plates with 200 μl of exponentially growing bacteria and resuspended in PBS to an OD550 of 1.0. Plates were air dried at room temperature and blocked with 200 μl of PBS-0.5% BSA-NaN3 for 1 h before 50 μl of different concentrations of PSGL-1 (R&D systems, USA) were added to each well. After overnight incubation at 4°C, plates were washed 5 times with PBS-Tween 0.1% and incubated overnight at 4°C with 50 μl of mouse anti-human PSGL-1 (KPL-1; MBL) diluted 1/4000. After 5 washes with PBS-Tween 0.1%, plates were incubated with goat anti-mouse IgG HRP (Southern Biotech) for 30 min at room temperature and developed with o-phenylenediamine (Sigma-Aldrich). Plates were measured at OD492 using a microtiter plate reader (Anthos 2020). Direct binding of PSGL-1 to purified LytA protein or type 3 CPS (ATCC 169-X, Merck Sharp & Dohme) was performed as described above except that the 96-well plates were coated with 50 μg of purified LytA protein or CPS per well. Purified LytA protein was obtained by overexpression in Escherichia coli [52].
Wild-type C57BL/6 mice and PSGL-1−/− mice were bred in a conventional animal facility at the School of Medicine, Universidad Autónoma de Madrid (UAM). PSGL-1−/− mice were kindly provided by Dr. D Vestweber and Dr. MK Wild (Max Plank Institute for Molecular Biomedicine, Münster, Germany). Wild-type C57BL/6 mice obtained from the Jackson Laboratory and PSGL-1−/− mice were backcrossed and the wild-type and PSGL-1−/− littermates obtained from crosses of the resulting heterozygous mice were used to breed our wild-type and PSGL-1−/− colonies used in this study. Animal procedures were approved by the Animal Care and Use Committee of ISCIII. All mice used were 8–16 weeks old, and within each experiment, groups of mice were matched for age and sex. Studies investigating pneumococcal sepsis or pneumonia were performed using groups of at least 5 mice and infected as previously described [48]. Briefly, for sepsis, mice were challenged with 5 × 106 CFU/ml for D39 strain or 3 × 104 CFU/ml for TIGR4 strain (in a volume of 200 µl) by the intraperitoneal route, whereas for pneumonia, mice under anesthesia with isofluorane were inoculated intranasally with 50 µl containing 107 CFU/mouse of D39 strain. For intravenous inoculation, mice were infected with 2 × 107 CFU/mouse of D39 strain through the tail vein. At 6 h and 24 h after challenge, a lethal dose of pentobarbital was administered and bacterial counts were determined from samples recovered from BALF, lung and blood. Experiments were repeated twice using 5 mice in each group and results were expressed as Log10 CFU/ml of bacteria recovered from the different sites. Cytokines were measured from BALF and blood of wild-type mice and PSGL-1−/− mice infected with D39 strain by the intranasal route as explained above. Cytokines levels (IL-2, IL-4, IL-5, IL-6, IL-10, IL-12p70, GM-CSF, TNF-α and IFN-γ) were determined by using a Luminex magnetic bead array assay (Bio-Rad) according to manufacturer protocols.
Experiments investigating cellular populations in lungs and blood after pneumococcal infection were performed in wild-type and PSGL-1−/− mice infected as explained above using a FACS assay as previously described [51, 53]. Briefly, single-cell suspensions were prepared in staining buffer (2% fetal calf serum in PBS), and non-specific binding was blocked with Fc block (BD Biosciences). Staining was performed using standard protocols with the following antibodies diluted 1/400 in staining buffer (rat anti-mouse CD11b-allophycocyanin (APC), Biolegend; rat anti-mouse CD4-FITC, Biolegend; rat anti-mouse CD8-APC, Biolegend; rat anti-mouse GR-1-phycoerythrin; Biolegend) [51, 53]. Cells were analyzed on a FACSCanto I using the FlowJo version 6.3.4 software package.
Mice were euthanized with pentobarbital and lungs were inflated and fixed with 4% PFA in PBS. Lungs were paraffin-embedded and 5-μm sections were obtained. Infiltrates of granulocytes and T cells were measured by staining with anti-Ly-6G/6C (antigen retrieval pH 6.5, 1/400, Abcam ab2557) and anti-CD3 (antigen retrieval pH 6.0, 1/200, Santa Cruz Biotechnology sc-1127) antibodies respectively. Immunohistochemistry was performed with the Dako LSAB+ System-HRP following manufacturer’s instructions.
Data are representative of results obtained from repeated independent experiments, and each data point represents the mean and standard deviations (SD) for 3 to 5 replicates. Statistical analysis was performed by using two-tailed Student’s t test (for two groups), whereas analysis of variance (ANOVA) followed by a Dunnett’s post hoc test was chosen for multiple comparisons. Survival was analyzed by the log-rank test. GraphPad InStat version 5.0 (GraphPad Software, San Diego, CA) was used for statistical analysis. Differences were considered statistically significant with P <0.05 (*) and highly significant with P <0.01 (**) and P <0.001 (***).
|
10.1371/journal.pcbi.1003793 | Sparsity and Compressed Coding in Sensory Systems | Considering that many natural stimuli are sparse, can a sensory system evolve to take advantage of this sparsity? We explore this question and show that significant downstream reductions in the numbers of neurons transmitting stimuli observed in early sensory pathways might be a consequence of this sparsity. First, we model an early sensory pathway using an idealized neuronal network comprised of receptors and downstream sensory neurons. Then, by revealing a linear structure intrinsic to neuronal network dynamics, our work points to a potential mechanism for transmitting sparse stimuli, related to compressed-sensing (CS) type data acquisition. Through simulation, we examine the characteristics of networks that are optimal in sparsity encoding, and the impact of localized receptive fields beyond conventional CS theory. The results of this work suggest a new network framework of signal sparsity, freeing the notion from any dependence on specific component-space representations. We expect our CS network mechanism to provide guidance for studying sparse stimulus transmission along realistic sensory pathways as well as engineering network designs that utilize sparsity encoding.
| In forming a mental percept of the surrounding world, sensory information is processed and transmitted through a wide array of neuronal networks of various sizes and functionalities. Despite, and perhaps because of, this, sensory systems are able to render highly accurate representations of stimuli. In the retina, for example, photoreceptors transform light into electric signals, which are later processed by a significantly smaller network of ganglion cells before entering the optic nerve. How then is sensory information preserved along such a pathway? In this work, we put forth a possible answer to this question using compressed sensing, a recent advance in the field of signal processing that demonstrates how sparse signals can be reconstructed using very few samples. Through model simulation, we discover that stimuli can be recovered from ganglion-cell dynamics, and demonstrate how localized receptive fields improve stimulus encoding. We hypothesize that organisms have evolved to utilize the sparsity of stimuli, demonstrating that compressed sensing may be a universal information-processing framework underlying both information acquisition and retention in sensory systems.
| It is well known that natural stimuli, such as visual images, are sparse in the sense that they can be well represented by a small number of dominant components, typically in an appropriate frequency space [1]. We may thus naturally expect that organisms' sensing has evolved to be adapted to such sparsity. One sign of this adaptation may be the great reduction in numbers between the receptor cells and the sensory neurons in the immediate downstream layers along the early stages of sensory pathways [2], [3]. For example, in the retina, the stimuli received by ∼150 million rods and cones are transmitted through only ∼1.5 million retinal ganglion cells [2]. More generally, it is important to know how the network topology of early sensory pathways reflects this type of adaptation. How have the networks along these pathways evolved so that they can best transmit sparse stimuli and the least amount of information is lost through network dynamics [4], [5]?
Theoretically, the above question translates into the search for a class of neuronal networks that takes advantage of stimulus sparsity and thus best encodes such stimuli. Naturally, such networks should need relatively few downstream neurons to sample the input from the receptors. An instructive technological analog is provided by the compressed sensing (CS) theory [6], [7]. When using sufficiently random sampling of sparse images, this theory allows us to dramatically reduce the sampling rate as compared to that expected for the uniform sampling of finite-bandwidth stimuli [8], without degrading the image reconstruction. Greatly improving the fidelity of high dimensional data reconstructions and developing efficient sampling algorithms, applications of CS have emerged in numerous fields, including physics, biology, and imaging [9]–[13].
In the context of neuroscience, it has been conjectured that the information processing in the brain may be related to the existence of an efficient coding scheme, such as compressed sensing [14], [15]. Using adaptive CS, for example, sparse representations of sets of sub-sampled inputs can be developed through unsupervised learning without knowledge of either the sampling protocol or the sparse basis of the measured signal, revealing that CS may, in theory, help to explain signal interpretation and transmission in the brain [16]. Following the CS mathematical structure, it has also been suggested that linear, discrete-time network dynamics can be used to encode sparse temporal sequences of information in their current activity and therefore neuronal networks may possess a greater theoretical memory capacity than previously hypothesized [17]. In this work, we take a new direction by constructing a spiking-neuron network model of an early sensory pathway and demonstrating how the firing rates of a relatively small set of sensory neurons with nonlinear dynamics can successfully encode network inputs. Deriving a linear mapping embedded in the network dynamics, we use CS theory and the dynamics of our model network over a biologically realistic time-scale to reconstruct visual stimuli, which are known to be sparse in frequency space [1]. We also find that the performance of this model can be greatly improved by incorporating the biologically realistic property of localized receptive fields [18], [19]. Unlike previous work [14], [15], the derived input-output relationship is not constructed through learning, and is instead intrinsic to the network dynamics, suggesting a possible way sensory information is transmitted downstream via sparse coding of stimuli through network dynamics.
Even before the discovery of CS, sparse coding was hypothesized as a feature fundamental to optimally representing sensory information, thus possibly leading to the emergence of spatial receptive-field properties of simple cells in the primary visual cortex [20], [21]. Instead of using the framework of optimization [20], [21], we consider how the time-evolving output of populations of firing neurons encodes stimulus information and examine the key characteristics of a CS neuronal network best evolved to transmit sparse stimuli. Underlining a novel notion of sparsity in terms of network dynamics, our results suggest a stimulus may be considered sparse if it can be accurately encoded by networks in which the number of downstream neurons is much lower than the number of input components, separating the notion of sparsity from any dependence on a particular component-space transform choice.
To study sparse stimulus transmission along early stages of sensory pathways, we have constructed our conceptual network model to consist of two layers, an input layer and a processing layer, representing the receptors and sensory neurons (sensory cells), e.g., retinal ganglion cells in the retina. Invoking the fact that the receptor neurons in the retina exhibit graded-potential rather than the usual action-potential responses [22], we represent the input layer by currents injected into the sensory neurons in the processing layer. Each current represents the stimulus intensity in the receptive field of a given receptor, which relays this intensity downstream to a number of sensory neurons. We describe these neurons using the pulse-coupled, integrate-and-fire (I&F) model [23]–[30]. Our model is intentionally idealized, so that only the most general features of early sensory pathways are incorporated. In this way, we aim to emphasize the possible role of the CS mechanism in a broad class of sensory systems.
In our model, the membrane-potential dynamics of the sensory neuron is governed by the differential equation (1)and evolves from the reset potential until it reaches the threshold potential . At the time this occurs, , we say that this neuron has fired (or spiked), reset to , and inject the currents into all the other sensory neurons post-connected to it, with being the Dirac delta function. Here, is the membrane-potential time-scale, and are the numbers of the receptors and sensory neurons, respectively, is the number of sensory-neuron connections, are the stimulus strengths transmitted by the receptors, and are connection matrices between the receptors and sensory neurons and between sensory neuron pairs, respectively, and and are the respective overall strengths of those connections. Stimulus components, , take on integer values between 0 and 255, indicating the light intensity of the stimulus. These components will typically be fixed over time, since we primarily consider stationary stimuli. We simulate this model for a run-time of 200 ms using an event-driven algorithm in which we analytically solve for sensory neuron voltages and spike times, choosing parameters ms, the dimensionless potential values and , , , , and [31], [32].
We first assume that every sensory neuron samples the stimulus randomly from the entire receptor pool, and choose the numbers of the receptors and of the sensory neurons to be such that (In most of our computations, due to the limitations imposed by our computational power, we take the ratio n∶m to be 10∶1 instead of the 100∶1 observed in early sensory pathways [2], [3].) While this assumption of random sampling is fundamental to conventional CS theory, we later consider the more realistic case in which photoreceptors are sampled locally by sensory neurons, which yields a significant improvement in stimulus encoding [18], [19]. Moreover, the sensory neurons are also initially assumed to be connected to each other randomly, but, as we will subsequently demonstrate, we can also assume that the sensory neurons are uncoupled without affecting the results of this work. While retinal ganglion cells, for example, are in some cases not thought to be connected to each other, there are also other cases in which connectivity is observed, and we therefore address this by considering both scenarios [33]–[35]. Although ganglion-cell connections are typically gap junctions [34], we model these connections with pulse-coupling to maintain model idealization and simplicity. Therefore, we first take the elements and of both connectivity matrices to be Bernoulli-distributed, and later consider several realistic alternative assumptions, such as the localized properties of receptive fields [36]. The inputs into the sensory neurons are assumed noiseless in our preliminary discussion, but we will address the impact of more biological noisy processing, due to fluctuations in photon absorption for example, in the upcoming section [37].
We emphasize that we are modeling a general early sensory pathway, rather than incorporating details specific to the retina, and therefore omit several detailed properties in order to accentuate the underlying CS mechanism. For example, compared to the actual retinal network, we only consider “on,” rod-like receptors, neglecting any time-course details of the graded potentials the receptors produce [38] and any crosstalk among the receptors [39]. In addition, we also neglect the rich variety of the neuron types in the retina [40] and their complex connectivity [41], the center-surround structure of the ganglion neurons' receptive fields [18], [19], any spatial differences in the density of the receptor distribution [42], as well as any inhibition [22].
To determine the degree of connectivity between our networks, we introduce the notion of convergence, which is defined as the average number of neurons presynaptic to any neuron in a given network. In particular, we use a convergence of 50 for , the sensory-neuron connection matrix, and a convergence of 10 for , the sensory-neuron to receptor connection matrix. The architecture of the network is represented graphically in Fig. 1. For this neuronal network, conceptually, the question is whether its dynamics take advantage of the sparse stimulus structure, and whether its topology can effectively and efficiently transduce the input information to the sensory neurons.
The above question translates to how to design the network parameters so that the information from the original stimulus is best retained by the firing rates embedded in sensory-neuron spike trains when , i.e., how closely we can reconstruct the original stimulus from the sensory neurons' firing rates given the model network's connectivity. A stimulus presented to receptors is considered -sparse, with , when at least one of its transforms into an appropriate frequency or wavenumber space, such as Fourier or wavelet, has at most components whose magnitude exceeds a small threshold [6], [43]. Given such a stimulus, our model sensory-neuron network generates a set of spike trains, which presumably encodes sparse stimulus information.
If we want to use the CS theory as a guiding principle in our model network construction, we immediately encounter a conceptual difficulty because a prerequisite for CS is linear signal measurement, whereas neuronal dynamics are nonlinear. However, it turns out that there is a linear structure corresponding to the input-output relationship embedded within this network. Using coarse-graining methods similar to kinetic theory in nonequilibrium statistical physics, we derive the linearized firing-rate system (2)valid when the neuronal firing rates, , satisfy for all and the membrane potential jump induced by each spike is small, [44], [45]. The firing-rate model (2) was previously derived in the population sense for ensembles of neuronal networks with stochastic inputs of homogeneous strength. However, our work here reveals that through coarse-graining over an ensemble of network realizations differing in initial voltage conditions, in which each network realization is forced by the same set of heterogeneous deterministic inputs, this firing-rate model can be extended to each individual neuron coupled in the network. For weak sensory-neuron coupling-strength and high sensory-neuron firing rates, we therefore obtain a linear network mapping of the stimulus intensities arriving at each receptor onto the firing rates generated by each sensory neuron. In this case, the network is mean-driven, with each sensory neuron receiving a large number of small inputs from its neighbors, which can be approximated by a Poisson spike train of inputs. Under this assumption, we derive a non-linear input-output mapping, which we then linearize in the limit. The linear network mapping (2) is between the -dimensional input vector and the -dimensional output vector of neuronal firing rates ; it is not a map between population-averaged input (the network input) and the population-averaged output (the network firing rate) as in traditional coarse-graining applications [44], [45].
The proximity of the firing rates we have obtained from the I&F model (1) and the linear network mapping (2) is depicted in Fig. 2. The red line displays the dependence of the relative firing rate difference on the overall stimulus intensity . Since the firing rate of each neuron is determined by its unique external input and network connectivity, the error given in Fig. 2 is the sum of the errors for all individual neurons. It is clear that, neuron-by-neuron, the two sets of firing rates agree well with one another over a wide stimulus-intensity range, , i.e., as long as the input to the sensory neurons is not too weak. For much of this regime, especially near , the model sensory neuron firing rates are typically between 20 Hz and 100 Hz, closely resembling experimentally observed firing rates of biologically realistic neurons, such as retinal ganglion cells [34], [46].
With the embedded linear network mapping (2), we arrive at our hypothesis that CS can potentially be a governing principle in transmitting sparse stimuli from the receptors to the sensory neurons, while simultaneously achieving a great reduction in the number of sensory neurons. In signal processing, the well-known Shannon-Nyquist theorem asserts that we must sample a signal with a given bandwidth uniformly at the rate of twice that bandwidth in order to be able to faithfully reproduce it [8]. However, according to CS theory, images that are (approximately) sparse in a wavenumber-space can be reconstructed from random samplings whose number is much smaller than the number of pixels composing the image by finding the reconstruction with the smallest number of nonzero wavenumber-space components. Ref. [43] shows that this difficult optimization problem becomes equivalent to the much simpler question of finding the reconstruction whose wavenumber-space-component magnitudes add up to the smallest sum. Mathematically, one thus replaces a computationally expensive optimization problem in wavenumber space by a much simpler optimization problem, which can be efficiently solved via several optimization algorithms [47], [48].
By applying the CS approach of Candès and Tao to the linear mapping (2), we can reconstruct the stimulus from our model spike trains [7]. Thus, to estimate the sensing accuracy of our model early-sensory-pathway network, we measure the firing rates of each neuron in this network, and use the linear network mapping embedded in this model to carry out the relevant optimization procedure for finding the sparsest stimulus reconstruction. In particular, we reconstruct the stimulus given the rates , which we measure from the full simulation of the I&F network (1), by minimizing the sum , where is the vectorization of the two-dimensional discrete cosine transform of the pixel matrix corresponding to stimulus , subject to the condition that the stimulus components satisfy the linear system (2). (See the Methods Section for details.) It is important to remark that this reconstruction procedure requires only a brief simulation time, generally no more than 100 ms, since any initial transients in the network dynamics are very brief and typically last no more than 25 ms. In the next section, we further analyze the dependence of the CS reconstruction on the simulation time, and demonstrate that successful signal recovery takes place over a biologically realistic time-scale.
We display three sets of results of our CS reconstruction procedure in Fig. 3, for which the stimuli are images of increasing complexity: stripes, dots, and flowers. Visually, the CS algorithm renders recognizable reconstructions of all the objects, and performs best with large shapes, flat surfaces, and gradual transitions, while leaving some graininess, which appears especially pronounced near sharp edges.
In determining the type of networks that can best take advantage of stimulus sparsity and optimally encode information, we study how the relative error, , of the CS stimulus reconstruction depends on the various model network characteristics. We measure this error using the formula where the Euclidean norm, , is defined analogously to the definition in the caption of Fig. 2. To isolate the effect of each characteristic, we vary only one parameter at a time while holding the remaining parameters constant.
First, we address how these CS networks depend on the convergence of the connections between the receptors and the sensory neurons, as represented by the matrix [49], [50]. As shown in Fig. 4a, the error decreases until the optimal convergence of about 10 is reached, and then increases. We remark that the high error for low convergence levels is due to the model sensory network not being able to sample all the receptors, while for high convergence levels all the sensory neurons receive nearly identical input. It should be clear why the convergence 10 is optimal; it is the ratio n∶m of the receptors versus the sensory neurons for our network model. At this ratio, with very high probability, each receptor feeds into precisely one sensory neuron. Due to the random sampling by the sensory neurons, on the other hand, again with high probability, the number of receptors relaying stimuli to a given sensory neuron will be approximately n∶m. Thus, all or most of the stimulus is used by the model sensory network, and there is little or no over- or undersampling.
Likewise, we address the question of how stimulus strength controls the optimality of CS networks. In Fig. 4b, we fix the convergence of and the relative component sizes of the stimulus , and scale linearly the overall stimulus strength . We observe that the reconstruction is best for moderate strength values, with particularly high reconstruction error for low , and slowly degrading reconstruction quality as becomes too large. For the optimal stimulus strength, the sensory neurons are then neither underdriven, such that there are not enough firing events to properly encode network input, nor driven so hard that their interaction becomes too strong, overwhelming the information in the input signal.
In contrast, as displayed in Fig. 4c, the reconstruction error appears to depend little on the convergence of the matrix encoding the connections among the sensory neurons. In particular, for the error size, it makes little difference whether a sensory neuron receives many weak pulses or a few strong pulses from its neighbors, indicating that the amount of fluctuations received from within the sensory neuron network plays a rather small role. The error also appears to be relatively independent of the overall connection strength, , at sufficiently low -values, and then grows linearly with , as shown in Fig. 4d. This reflects the fact that cross-talk among the sensory neurons that is too strong is likely to drown out the signal received from the receptors. Altogether, it thus appears that the connections among sensory neurons neither improve nor degrade the performance of the model network as long as their strengths are moderate. In the case of the retina, we note that while it was previously thought that there is no recurrent connectivity among retinal ganglion cells, recent experimental work shows that there is indeed gap-junction-type coupling among specific types of ganglion cells. [11], [33]–[35], [51], [52]. In either case, as long as the recurrent coupling is not too strong, the model sensory pathway will still accurately encode sparse stimuli. Therefore, the results of this work may more broadly apply to various types of ganglion cells, exhibiting diverse types of connectivity.
Clearly, for a CS network to be dynamically responsive in capturing transient stimuli, the system should be able to rapidly sample the stimulus within a sufficiently short time interval from the stimulus onset for the CS reconstruction. As shown in Fig. 4e, the reconstruction error drops precipitously until the sampling time increases to about 25 ms, and then remains approximately steady. The 25 ms time scale agrees with typical sensory time scales [53], [54]. To address the possibility of minor distortions of information along sensory pathways, we further address how the performance of a CS network is degraded in the presence of noise. As shown in Fig. 4f, we find that the relative reconstruction error grows approximately linearly with the variance of the Gaussian noise added to each stimulus component, demonstrating that a recognizable reconstruction is still achievable even in the presence of relatively high-variance noise.
Since thus far we have used a fixed number of sensory neurons, a natural question to ask is how the performance of a CS network improves as the number of sensory neurons increases. Fig. 4g shows that the performance will in fact improve with additional sensory neurons given a fixed number of receptors and corresponding optimal convergence. Since the reconstruction quality improves significantly less with the addition of sufficiently many sensory neurons, we observe that adding too many sensory neurons may be wasteful from a computational point of view, further justifying the optimality of sensory pathway architecture in processing sparse stimuli.
Hypothesizing that randomness is a key aspect in CS network sampling, we examine a central question of just how randomly sensory neurons need to sample the stimulus in order to achieve optimal sparsity encoding. To answer this question, we first design the connectivity matrix so that all sensory neurons sample receptors from a regular grid. Then, we sequentially remove an original connection in , and replace it by a connection between a randomly chosen receptor and the same sensory neuron. (See Methods Section for details.) From Fig. 4h, we see that the error decreases rapidly until of the initial regular connections have been rewired, and then slowly levels off. Therefore, some degree of randomness is in fact necessary for a viable reconstruction, however the sampling need not be completely random for successful sparsity encoding. In fact, the success of intermediate levels of randomness may help to explain how the localized sampling in receptive fields further improves the performance of the CS network, which we will address later in this section.
Next, we investigate the characteristics of sensory-neuron spike dynamics that are significant in these sparsity-encoding CS networks. We find that the parameter regimes yielding the least error in the stimulus reconstructions are those in which the largest degree of variability or disorder exists among the dynamics of the model sensory neurons. We compute the average sensory network membrane potential, , which roughly models the network local field potential (“LFP”) signal measured experimentally [55], [56], to give an indication of the variability in network dynamics. In Fig. 5a, we plot the “LFP” correlation time as a function of the convergence of the receptor-sensory-neuron connection matrix . It is clear that the “LFP” decorrelates the fastest at the optimal convergence value, indicating relatively aperiodic network dynamics. To quantify the corresponding network information content, we compute the entropy, , of the spike train produced by the network of neurons where denotes the probability distribution of the interspike-interval (ISI) lengths, computed from a binned histogram of ISI's collected from each sensory neuron in the network. In our case, the entropy of the ISIs measures the spike-train information capacity, and therefore gives an indication as to how much possible information can be encoded by the sensory-neuron network over the time-scale of network activity [57], [58]. This entropy reaches its maximum at the optimal convergence, as displayed in Fig. 5b, thereby transmitting the maximum amount of information. It is important to remark that while we specifically use firing rates to reconstruct stimuli, information about the actual sensory-neuron spike trains is embedded in the firing rate statistics. Since the firing rate gives the lowest order of information regarding the ISI distribution, the ISI distribution is of particular interest in quantifying the information encoded by sensory-neuron activity.
In examining the distribution of the ISI's, we observe a rich firing structure among the sensory neurons at the optimal convergence of connectivity matrix . We demonstrate in Fig. 5c how the variance of the ISI distribution depends on the convergence of . (Here, denotes the mean over the distribution.) This variance is clearly maximal at the optimal convergence. Moreover, the ISI structure is further characterized by its near-Gaussian distribution at optimal convergence value, as shown in Fig. 5d, reaching its minimal skewness, , and kurtosis, , which vanish for the Gaussian distribution. From these observations, it is clear that the connectivity between the receptors and sensory neurons plays a large role in determining the information content of the sensory neuron spike dynamics, and by maximizing the information content of these spikes, stimuli may be optimally encoded.
We further corroborate the hypothesis that evolution may have driven early sensory pathways to become CS networks by incorporating a biologically realistic feature, i.e., localized receptive fields, into our model CS network. We discover that this feature indeed improves the performance of the highly idealized CS network we have investigated so far. We model such a receptive field by using a variant of the model in which each sensory neuron samples receptors primarily within a small area, which is closer to biological realism than random sampling [18], [19]. In particular, if the coordinates of the pixel representing a receptor are given by the vector and the coordinates of the receptive-field center corresponding to a chosen sensory neuron are given by the vector , we take the probability that a connection will exist between the two as where represents the probability of a connection if , and is the size of the sensory neuron's receptive field in the units of pixel size. A schematic illustration of this type of sampling is depicted in Fig. 6a, with the parameter values, and , resulting in a convergence of of 25. A reconstruction of the original image from the firing rates produced by this model is shown in Fig. 6b. Note that we found the error of this reconstruction to be 0.19, which is much less than the error of 0.35 we obtained for completely random stimulus-sampling over the entire receptor pool, shown in Fig. 4a. This result, reaching beyond the conventional CS theory, underscores the importance of the local-receptive-field architecture in the evolution of the CS properties of sensory pathways.
We remark that we can also reconstruct moving stimuli using our CS approach. The reconstruction of a ten-snapshot image sequence spaced 25 ms apart is displayed in Fig. 7. In reconstructing each image frame, we only use spikes counted during the time-course over which each respective image is presented. In this way, we consider ten CS recovery problems, with each corresponding to a separate set of observed firing rates. From the highly accurate reconstructions even in the case of moving stimuli, it is clear that the CS architecture is therefore feasible for natural environments in which stimuli are constantly in motion. Moreover, if the image frames instead change every 200 ms, the average reconstruction quality is improved further. As in the case of realistic retinal video processing, correlations between frames and close corresponding dynamical regimes therefore allow for rapid encoding of changing stimuli [59]–[61].
We hypothesize that the CS principle for sparse-stimulus transmission in neuronal networks, as demonstrated in our computational model, should also hold in real neuronal systems in the brain. In more general settings, the underlying linear structure could be recovered using the first-order Wiener kernel from non-linear systems analysis for the entire network [62], [63]. Similarly, in deriving an input-output relationship outside of the mean-driven regime, a linear-nonlinear (LN) model can also be developed through application of a linear spatiotemporal filter and a static nonlinear transformation (e.g. sigmoidal function), which often can be cast in a linear form if the inverse of the nonlinear transformation exists [64]. In either case, once the underlying linear structure is discovered, the presented methodology could in principle be used to attempt to reconstruct sparse stimuli using very few neuronal output measurements.
Mathematically, this work suggests two important extensions to conventional CS theory. First, while compressed sensing is traditionally applied to static linear systems, we demonstrate one possible way of generalizing this theory to dynamical systems that model a large number of interacting agents. Second, the improvement in stimulus encoding yielded by localized random sampling akin to receptive fields suggests that alternative sampling schemes, aside from purely random sampling, may in fact yield better reconstructions so long as there is a sufficient degree of incoherence in the samples such that CS is applicable. From this standpoint, measurement devices engineered with localized random sampling may be able to more successfully encode signals than by applying the completely random sampling conventionally used in compressive sensing data acquisition [65]. Likewise, engineered devices sampling the output of a time-evolving network may also have the capacity to reconstruct network input using compressive sensing combined with an underlying linear input-output network structure similar to the neuronal network studied in this work.
Finally, we point out a new way of looking at the mathematical framework of sparsity. Our findings give rise to a network definition of stimulus sparsity, freeing this concept from any dependence on the particular choice of wavenumber-space or other component-space transform as in conventional definitions of sparsity. In particular, we can define a stimulus as sparse if it can be accurately and efficiently transmitted through a sensory-pathway-type network, such as one that allows for a significant reduction in the numbers of downstream sensory neurons versus the numbers of upstream receptors. This alternative definition of sparsity therefore directly relates the structure of a stimulus to the type of network dynamics it evokes. Rather than indicating sparsity by the number of non-zero signal components, sparsity can alternatively be determined in the network framework by the amount of stimulus information embedded in the evoked network dynamics. Thus, visual images are clearly sparse according to both the networks that sample them completely randomly and those with localized receptive fields.
In the long way towards understanding how the brain forms a specific percept from a given stimulus, one must first understand how the brain samples this stimulus. Our aim here was to examine the hypothesis that the CS principle has evolved to govern the information transduction and retention of sparse stimuli in a sensory pathway while achieving a great reduction in the number of sensory neurons. Our work shows that this hypothesis indeed successfully captures information propagation in our model sensory network. In particular, our results on these network characteristics may provide insight into the CS properties of corresponding networks in the brain.
The correlation time gives the expected amount of time necessary for signal responses to become decorrelated and is defined mathematically as where is the correlation function of the “LFP” , with and time average of the “LFP” is . A short correlation time implies less periodicity and therefore greater variability in the “LFP”.
To reconstruct a stimulus, , from the sensory-neuron firing rates, , we first cast the linearized firing-rate model (2) into a form to which compressed sensing may be applied. To apply compressed sensing in recovering a sparse representation of , we consider the vectorization of the two-dimensional discrete cosine transform of the stimulus pixel matrix, , where denotes the Kronecker product is the , one-dimensional discrete cosine transform matrix with entries , and . In solving the related problem of recovering , the linear model we consider is (3)
Since the cosine transform of the stimulus, , is sparse and the matrix is random, recovering is reduced to minimizing the sum [6], [7] under the constraint (3). Solving this minimization problem is equivalent to solving the well-known optimization problem under the constraint (3). We solve this optimization problem with a greedy algorithm, known as the Orthogonal Matching Pursuit [48]. Once is recovered, we recover the stimulus using the inverted two-dimensional discrete cosine transform of the matrix representation of .
As mentioned in the main body of this paper, CS theory posits that random sampling of sparse images significantly reduces the sampling rate as compared to uniform sampling of finite-bandwidth stimuli, while yielding the same quality of the reproduction. In particular, for uniform sampling, the Shannon-Nyquist theorem requires that finite-bandwidth stimuli must be sampled at the rate of twice their bandwidth in order to achieve a faithful reconstruction [8]. In our case, this would mean sampling by sensory neurons when all the frequencies are represented in the image used as the stimulus. This is because we need to capture each Fourier mode represented in the stimulus in at least two points. On the other hand, the compressed-sensing theory implies that much less frequent sampling should suffice for -sparse stimuli to perfectly reconstruct the stimulus with probability one, in particular, on the order of [6], [7], provided the sampling is sufficiently random. Again, this is because, with probability one, we will thus capture each represented Fourier mode in two points. This is certainly not true if we sample the stimulus on a regular, coarse grid with points in the spirit of the Shannon-Nyquist theorem, unless the stimulus contains nothing but the lowest modes. In fact, such sampling diminishes the resolution. We here elaborate on the illustration of this fact, as depicted in Fig. 4h.
In regularly sampling the stimulus, , sensory neurons sample only receptors that lay on a coarse grid contained within the finer grid of receptors, modeled by the pixel matrix representation of . The coarse regular grid, say of size , is composed of all even-numbered row and column entries of the finer pixel matrix. Fixing convergence at its optimal level, the regular sampling scheme randomly connects sensory neurons to receptors on the coarse grid. To add more randomness and a larger variety of frequency modes to the sampling scheme, we randomly select sensory neurons connected to receptors on the regular grid, and then randomly rewire them with any of the receptors on the pixel matrix.
As displayed in Fig. 4h, even if the sampling is random on a coarse grid, not enough frequency modes may be captured to yield a faithful signal reproduction. Upon random rewirings to the finer grid of receptors, more frequency components may be detected, thereby improving the quality of the reconstruction. However, once the sampling scheme is sufficiently random and enough variety in frequency modes is captured, an accurate reconstruction can be achieved with little improvement following additional rewirings.
|
10.1371/journal.pbio.1002200 | The Orphan Nuclear Receptor TLX Is an Enhancer of STAT1-Mediated Transcription and Immunity to Toxoplasma gondii | The protozoan parasite, Toxoplasma, like many intracellular pathogens, suppresses interferon gamma (IFN-γ)-induced signal transducer and activator of transcription 1 (STAT1) activity. We exploited this well-defined host–pathogen interaction as the basis for a high-throughput screen, identifying nine transcription factors that enhance STAT1 function in the nucleus, including the orphan nuclear hormone receptor TLX. Expression profiling revealed that upon IFN-γ treatment TLX enhances the output of a subset of IFN-γ target genes, which we found is dependent on TLX binding at those loci. Moreover, infection of TLX deficient mice with the intracellular parasite Toxoplasma results in impaired production of the STAT1-dependent cytokine interleukin-12 by dendritic cells and increased parasite burden in the brain during chronic infection. These results demonstrate a previously unrecognized role for this orphan nuclear hormone receptor in regulating STAT1 signaling and host defense and reveal that STAT1 activity can be modulated in a context-specific manner by such “modifiers.”
| Immune responses are orchestrated by a diverse array of secreted ligands, yet the downstream transcriptional responses are coordinated by a relatively small set of key transcription factors, including nuclear factor kappa B (NF-κB) and signal transducers and activators of transcription (STATs). The molecular mechanisms that tailor the output of these immune signaling pathways to generate cell-, tissue-, or context-specific responses are poorly understood. In this study, we exploit a host–pathogen interaction, Toxoplasma gondii infection in mice, using a genetic screen to identify host factors that overcome parasite suppression of STAT1 signaling. We show that the orphan nuclear receptor TLX, a key regulator of brain development, enhances expression of a subset of STAT1-dependent genes in response to IFN-γ stimulation. Through genetic and pharmacological studies, we show that endogenous TLX function is required for triggering appropriate responses to IFN-γ in astrocytes. Moreover, we found that genetic disruption of TLX in mice impairs their ability to mount an effective immune response and control T. gondii infection in the brain. These data suggest that natural or synthetic ligands for TLX might be effective tools for modulating immune responses, particularly in the brain where TLX expression is highest.
| Interferon gamma (IFN-γ) and STAT1 signaling play an essential role in cellular immunity, as indicated by extreme susceptibility to infection in mice and humans carrying mutant alleles for these genes and pathways [1–5]. The binding of IFN-γ to its cell surface receptor leads to phosphorylation, dimerization, and subsequent nuclear translocation of the transcription factor STAT1 [6]. Once inside the nucleus, STAT1 dimers recognize a consensus “gamma-activated sequence” (GAS) element (TTCN3–5GAA) in target genes and initiate a transcriptional program that is essential for resistance to a broad range of pathogens. While the core components of the IFN-γ signaling pathway, including Janus kinase 1 and 2 (JAK1, JAK2), and STAT1, have been known for nearly two decades [7,8], regulatory mechanisms dictating the specificity and strength of STAT1 activity within the nucleus are poorly understood, limiting our understanding of how IFN-γ /STAT1 signaling can be tailored or harnessed to respond to different challenges.
STAT1 activation is known to transcriptionally regulate several hundred genes [9], including pathways involved in host defense, apoptosis, and differentiation [10], but the biological output of STAT1 signaling can vary dramatically depending on the context in which activation occurs. For example, while STAT1 signaling suppresses cell proliferation during hematopoiesis [11,12], IFN-γ-induced STAT1 signaling can also drive hematopoietic stem cells to enter the cell cycle and proliferate to replace leukocytes lost during infection [13]. This apparent duality of STAT1 function also extends to additional immune cells: STAT1 signaling in macrophages activates a potent antimicrobial program and promotes antigen processing and presentation to T cells, yet STAT1 is also required for the ability of tumor-associated macrophages to suppress T cell function [14]. Similarly, while STAT1 drives T helper type 1 (Th1) cell development by activating the transcription factor T-bet [15], it is also required for the suppressive function of regulatory T cells [16,17]. The diversity of outcomes associated with STAT1 activation highlights the need to identify the cellular factors that modify or modulate STAT1 target selection to appropriately tailor the output of this core signaling pathway.
The intracellular protozoan parasite Toxoplasma gondii is a common pathogen of humans and other warm-blooded vertebrates and a valuable model for understanding IFN-γ-mediated immunity. Infection is initiated in the gastrointestinal tract and proceeds through an acute phase characterized by rapid parasite replication within hematopoietic and nonhematopoietic cells. IFN-γ signaling is critical in controlling parasite replication during this phase [18], and mice deficient in IFN-γ or STAT1 rapidly succumb to infection [19–21]. IFN-γ responses do not completely eradicate Toxoplasma, and parasites that evade this response differentiate into a slow-growing cyst form that persists as a latent infection in the central nervous system. During this chronic phase, IFN-γ remains essential in restricting parasites and prevents reactivation to the rapidly dividing form [22–24].
As a testament to the importance of IFN-γ and STAT1 signaling in the control of infections, many pathogens have evolved mechanisms to directly inhibit pathway components [25–28]. Toxoplasma can attenuate IFN-γ receptor-dependent STAT1 signaling [29–33]. We exploited this host–pathogen interaction as the basis for a high-throughput screen in order to identify host factors that when ectopically expressed could overcome the Toxoplasma-dependent block of STAT1 activity. This approach led to the identification and validation of nine transcription factors, including the orphan nuclear hormone receptor TLX (tailless, also known as NR2E1), which acts as a potent enhancer of STAT1 target gene expression. Furthermore, we found that TLX selectively potentiates a subset of STAT1-dependent targets, providing insight into how specific STAT1 programs can be tailored to impact immunity.
In order to identify host cell modifiers of STAT1 signaling in Toxoplasma-infected cells, we utilized a STAT1-responsive luciferase reporter, consisting of two tandemly repeated GAS elements able to bind STAT1 homodimers. Treatment of cells with IFN-γ leads to robust activation of the GAS luciferase reporter, but prior infection with Toxoplasma suppresses this activation by >5-fold (S1A Fig). This did not reflect a general impairment of host cell signaling, as parasites were unable to suppress tumor necrosis factor alpha (TNF-α) induction of an nuclear factor kappa B (NF-κB) reporter (S1B Fig). Overexpression of STAT1 failed to rescue pathway activity, suggesting that Toxoplasma impacts a step downstream of STAT1 stability (S1C Fig). Since STAT1 phosphorylation is required for dimerization, phospho-specific antibodies were used to further characterize pathogen suppression of the STAT pathway. Human osteosarcoma U2OS cells were infected with a low multiplicity of infection (MOI) to leave some cells uninfected, allowing us to evaluate STAT1 phosphorylation in both infected and naïve cells from the same cultures. Infection alone was not sufficient to trigger STAT1 activation (S1D Fig), whereas a 15-min stimulation with IFN-γ triggered phosphorylation of STAT1 in nearly every infected and uninfected cell (S1D Fig). STAT1 dimerization allows for nuclear import and subsequent DNA binding to induce transcription, and immunofluorescence revealed that STAT1 translocation to the nucleus is unaffected by infection (S1E Fig). Taken together, these data are consistent with previous reports [29–33], which indicate that Toxoplasma impairs STAT1 signaling by acting downstream in the pathway, at the level of nuclear STAT1 function (S1F Fig).
Conditions were optimized for high-throughput screening to identify genes that when ectopically expressed restored activity of the STAT1 pathway in infected cells, yielding Zʹ-factor scores > 0.5 (S2 Fig), a measure of assay robustness [34]. Pathway suppression could not be overcome by increasing the IFN-γ concentration (S2C Fig) Moreover, this STAT1 assay is 30 times more sensitive to the STAT1 homodimers triggered by IFN-γ stimulation than the STAT1/STAT2 heterodimers formed upon activation by type I interferon (S2D Fig). These data indicate that Toxoplasma suppression of the STAT1 pathway provides a robust, sensitive, and specific screen to identify enhancers of IFN-γ-induced STAT1-mediated transcription.
The Mammalian Gene Collection (MGC), a library of over 18,000 human and mouse full-length and sequence-validated cDNAs [35], was screened to identify genes able to restore function to the STAT1 pathway upon ectopic expression in Toxoplasma-infected cells (Fig 1A). The primary screen identified 32 cDNAs (17 mouse and 15 human; representing 28 genes) that enhanced STAT1 activity ≥ 2.5-fold in replicate screens of the library and had robust Z-scores ≥ 4 (Fig 1B, inset; S1 Table). Gene ontology (GO) analysis of these 32 STAT1 enhancers indicates that 21 cDNAs are involved in regulation of transcription, a 4.6-fold enrichment relative to the complete MGC library (p-value < 0.001). Of these 21 cDNAs, two represent orthologs of HOX5A, while two others are isoforms of mouse CRTC2. In addition, the screen identified all three ETS2 isoforms present in the library (one human and two mouse). In total, 17 unique transcriptional regulators were identified in the primary screen. Analyzing these 17 genes for Pfam domains identified motifs that defined at least six transcription factor families (Fig 1C). A network analysis (Fig 1D) reveals no previously reported direct protein–protein interactions between STAT1 (red) and the screen hits (black), but eight screen hits have been found to interact with network neighbors (green) known to directly bind STAT1, including the well-known STAT1 regulators PIAS1, CREBBP, and EP300 [36,37].
To assess whether genes identified in this screen are bona fide regulators of STAT1 activity, individual cDNA clones for the 21 putative STAT1 enhancers were sequence verified and tested for their ability to rescue STAT1 activity during parasite infection (S2 Table). Activity was also assayed in uninfected cells transfected with these cDNAs. Seventeen of the 21 cDNAs were found to enhance STAT1-dependent transcription by at least 2.5-fold in infected cells. Fifteen of these 17 cDNAs also activated transcription in uninfected cells, suggesting that the transcription factors identified through this screen represent general enhancers of IFN-γ-dependent STAT1 activation, irrespective of Toxoplasma infection. In order to rule out nonspecific induction of the luciferase reporter, these genes were tested for induction of a control reporter lacking the GAS elements. Six of the 17 STAT1 enhancers, including all Crtc genes and several Hox genes, induced the control reporter and therefore were not pursued further, leaving 11 cDNAs (nine genes, including two orphan nuclear hormone receptors) that specifically enhance STAT1 activity in infected and/or uninfected cells (S2 Table and S3 Fig). Finally, to rule out the possibility that these hits affected parasite fitness, we carried out additional screens of the MGC using transgenic parasites in which luciferase reports either cell invasion [38] or parasite viability [39]. These nine genes did not impact either of those assays (S2 Table), indicating that our screen identified modulators of host cell signaling, rather than direct parasite effectors, consistent with their ability to potentiate STAT1 signaling in the absence of Toxoplasma infection. This expands our knowledge of genes that can regulate this important signaling pathway.
Since the screen employed in this study was based on a reporter that responds to STAT1 homodimers, seven candidate transcription factor genes were tested against six additional pathway reporters to determine if they act specifically on the STAT1 pathway, or whether they might function as enhancers of additional immune-related signaling pathways. U2OS cells were co-transfected with candidate cDNAs along with luciferase reporter constructs under the control of either interferon response factor-1 (IRF1), NF-κB, interferon-stimulated gene factor-3 (ISGF3), serum response factor (SRF), activator protein-1 (AP1), the STAT1 homodimer reporter used in the screen (positive control), or negative control reporters (Fig 2). Cultures were then stimulated with IFN-γ to induce STAT1 and IRF1, IFN-α to induce ISGF3, or TNF-α to induce NF-κB. These particular pathway reporters and stimuli were selected because they represent well-known STAT1-dependent and STAT1-independent inflammatory pathways. Reporters for STAT1, IRF1, ISGF3, and NF-κB all responded to the appropriate stimulus with increased luminescence, while the two control reporters and AP1 and SRF reporters were not activated by any of the stimuli tested (Fig 2). All of the genes tested robustly enhanced IFN-γ-induced STAT1 homodimer activity (Fig 2). Four of the seven genes also activated IRF1 activity, consistent with the known role of IRF1 as a STAT1 target gene [40]. Moreover, these factors were not promiscuous; they displayed little activity on the other reporters (Fig 2). Taken together, these data indicate that this high-throughput screen has identified a set of transcription factors that show clear specificity for the STAT1 pathway.
We identified two orphan nuclear receptors, COUPTF2 and TLX, which exhibited strong enhancement of our STAT1 reporter (S2 Table). Because nuclear hormone receptors are well-known transcriptional regulators, acting as co-activators and/or co-repressors, and are druggable targets, and because TLX is expressed in the brain [41–43], where STAT1 activation is required to control chronic Toxoplasma infection [22–24], we focused our studies on TLX.
First, we set out to identify the spectrum of endogenous genes regulated by TLX. U2OS cells transiently expressing TLX or empty cDNA vector were untreated or stimulated with IFN-γ for 8 h and transcriptionally profiled. Hierarchical clustering of 341 differentially expressed genes (≥1.5-fold, false discovery rate (FDR) ≤ 5%) delineated at least three clusters of co-regulated transcripts (Fig 3A). GO enrichment analysis of these clusters showed that TLX overexpression enhanced transcription of 104 IFN-γ-independent genes involved in neuron differentiation and tissue morphology (Fig 3B, cluster 3), consistent with the fact that TLX is expressed in the brain, where it is an essential regulator of neurogenesis [41]. Amongst these genes are known regulators of brain physiology (Fig 3C), including brain-specific solute carriers (SLC17A7 and SLC30A3), a neuronal calcium sensor (HPCAL4), a neuron specific vesicular protein (CALY), and an essential regulator of dopamine neuron development (CDNK1C) [44], suggesting that they may represent natural targets either directly or indirectly regulated by TLX.
As expected, IFN-γ treatment of U2OS cells enhanced expression of 162 genes involved in immunity and inflammation (Fig 3A and 3B, cluster 1). Importantly, expression of 19 IFN-γ-inducible genes involved in host defense was potentiated by TLX expression (Fig 3A and 3B, cluster 2), including CXCL9, CXCL10, and CXCL11 (Fig 3C, cluster 2), all of which are well-known STAT1-dependent targets.
Next, we sought to identify cell types that expressed high basal levels of endogenous TLX to determine the role of TLX in STAT1-dependent responses. A survey of nuclear receptor expression in the National Cancer Institute (NCI) 60 panel, a collection of cancer cells lines targeted for extensive study, including gene expression profiling [45], suggested that astrogliomas express the highest levels of TLX mRNA, with U251 cells having the highest expression [46]. Transfection of U251 cells with small interfering RNAs (siRNAs) against TLX resulted in a greater than 80% reduction in TLX transcript compared to control (Fig 4A). Next, cells transfected with siRNAs targeting TLX or with a control siRNA were either untreated or stimulated with IFN-γ for 8 h and were subject to expression profiling by microarray. Hierarchical clustering of 1,418 differentially expressed genes (≥1.5-fold, FDR ≤ 5%) delineated three clusters of co-regulated transcripts (Fig 4B). Knockdown of TLX resulted in marked repression of 352 IFN-γ-independent genes associated with cell cycle regulation (Fig 4B and 4C, cluster 3), consistent with the critical role for TLX in maintaining a proliferative state in adult neural progenitor cells [41,47]. Amongst these TLX-dependent transcripts were several genes involved in central nervous system function (Fig 4D), including a synaptic adhesion molecule (CADM1); the neuronal signal transducer, chimerin-1 (CHN1); a serotonin binding glycoprotein (GPM6B); and sorting nexin family member 27 (SNX27), a gene recently shown to regulate developmental and cognitive impairment in Down syndrome [48]. Moreover, four genes previously reported to be TLX dependent in mouse brain [47] were also identified as TLX dependent in this experiment (Fig 4D, asterisks). TLX depletion also resulted in enhanced expression of 411 genes that were enriched for sterol metabolism and endocytosis (Fig 4B and 4C, cluster 2).
U251 astroglioma cells exhibited a robust transcriptional response to IFN-γ treatment, resulting in up-regulation of a similar profile of immune defense genes as seen in U2OS cells (Fig 4B). A subset of these genes (Fig 4A, cluster 1) was TLX dependent and was enriched for GO terms relating to inflammation and antigen presentation (Fig 4C, cluster 1). These IFN-γ- and TLX-dependent genes included CXCL9 and UBD, as well as guanylate-binding proteins (GBPs) (Fig 4E). Taken together with our ectopic expression studies, these data identify TLX as a transcription factor that regulates the steady-state expression of STAT1-independent genes involved in brain function, brain development, and cell cycle while enhancing the output of a subset of IFN-γ-dependent target genes.
Nuclear receptors can regulate gene expression either directly through DNA binding or indirectly by physically interacting with other transcription factors [49]. To determine whether DNA binding was required for TLX to regulate STAT1-dependent transcripts, U2OS cells were transfected with TLX constructs lacking either the DNA binding domain or the ligand binding domain. As expected, wild-type TLX markedly enhanced IFN-γ induction of our STAT1 reporter, and this induction was completely abrogated if only the DNA binding domain (TLX ∆DBD) or ligand binding domain (TLX ∆LBD) was used (Fig 5A). Similarly, when CXCL9 and CXCL10 expression were used as readouts of STAT1 function, both domains of TLX were also required for efficient induction of these genes following IFN-γ stimulation (Fig 5B). Taken together, these data show that TLX requires both DNA binding activity and ligand binding activity to enhance STAT1-mediated transcription.
Although endogenous ligands for TLX have not been described, a recent small-molecule screen identified famprofazone, a nonsteroidal anti-inflammatory drug, as a synthetic ligand for TLX that induces transrepression [50]. We reasoned that if endogenous levels of TLX potentiate the expression of specific IFN-γ-inducible STAT1 target genes, then famprofazone should inhibit this response. To test this, U251 astroglioma cells were pretreated with famprofazone and subsequently stimulated with IFN-γ, IFN-α, or TNF-α, and the expression of CXCL10 and OAS2 was measured by RT-qPCR (Fig 5C). CXCL10 and OAS2 were most potently induced by IFN-γ and IFN-α, respectively—consistent with their known role as canonical targets of these cytokines—whereas TNF-α had the weakest effect on both of these targets. Famprofazone treatment dramatically impaired the ability of IFN-γ and IFN-α to induce CXCL10 (Fig 5C), while a more modest impairment was observed on TNF-α induction of CXCL10. Interestingly, famprofazone also impaired induction of OAS2, but only when IFN-γ was used as the stimulus. In contrast, TNF-α and IFN-α induction of OAS2 was unaffected by famprofazone. These data suggest that TLX regulates STAT1 function in a stimulus- and target-specific manner.
The observation that TLX required both a DNA binding domain and a ligand binding domain for enhancement of CXCL9 and CXCL10 (Fig 5B) suggested that TLX might interact directly with the promoter of select STAT1 target genes. We hypothesized that this could increase the amount of phosphorylated STAT1 at the promoters of these genes, thereby leading to enhanced transcription. To test this hypothesis, we carried out chromatin immunoprecipitation (ChIP) in U2OS cells using antibody specific for phosphorylated STAT1, followed by quantitative PCR (qPCR) for the region of the CXCL9 and CXCL10 promoters where STAT1 is known to bind (Fig 6) [51]. Cells ectopically expressing a control vector showed increased ChIP signal at both promoters after a 2-h IFN-γ stimulation, relative to control antibody. Compared to control vector, cells expressing TLX showed a 3-fold and 2-fold increase in phosphorylated STAT1 (pSTAT1) binding at the CXCL9 and CXCL10, respectively (Fig 6). These data demonstrate that TLX can enhance transcription of IFN-γ-dependent genes by enhancing promoter occupancy of phosphorylated STAT1.
Although studies have explored the cell types expressing TLX in the developing and adult brain, whether TLX expression is regulated, in particular during a proinflammatory insult to the brain, is unknown. Given our finding that TLX potentiates expression of IFN-γ-inducible chemokines, we reasoned that in vivo infection with Toxoplasma—a potent inducer of IFN-γ production—might alter TLX expression in the brain. To test this, mice were infected and allowed to progress to chronic infection, at which point whole brains were removed, sectioned, and stained with a polyclonal antibody to TLX. Brain sections from naïve animals showed only modest, nuclear-localized staining in the granular cell layer of the dentate gyrus, a region previously described to contain TLX-positive neural progenitor stem cells (S4 Fig) [41]. In contrast, in the infected brain, numerous cell types stained positive for TLX (Fig 7), including cells with an apparent leukocyte morphology that were found near parasite cysts (Fig 7A and 7B, arrow). In addition, neurons within the cerebral cortex stained intensely for TLX (Fig 7C and 7D). These data demonstrate, for the first time, that although TLX is expressed selectively in granular layer of the dentate gyrus in the normal adult brain, CNS infection can induce TLX expression in a variety of cell types, many of which are proximal to the microbial insult.
The critical role for IFN-γ and STAT1 in restricting Toxoplasma, taken together with our finding that TLX is induced in the brain during infection and can regulate IFN-γ-dependent expression of molecules such as CXCL9, CXCL10, GBP4, and GBP5, prompted us to examine whether TLX impacted infection or pathogenesis. Mice carrying a floxed allele of TLX were crossed to mice expressing Cre under the control of Mx1, allowing inducible deletion of TLX [47]. Following treatment with Poly(I:C) to induce deletion, mice were rested for 2 wk and then infected with Toxoplasma. Animals were allowed to progress to chronic infection when parasites had established long-lived infection in the brain, at which point two key parameters of the immune response to this parasite were examined: induction of a potent Th1 response marked by IL-12 production and control of parasite replication.
Dendritic cells are a critical source of IL-12 during Toxoplasma infection, and their ability to produce this cytokine is required for parasite control [52]. In addition, IL-12 is STAT1-dependent during Toxoplasma infection [20], as well as in other protozoan infections [53]. Therefore, we compared dendritic cell IL-12 production in our wild-type (WT) and TLX-deficient mice. As expected, splenic dendritic cells recovered from control TLXf/f infected mice produced IL-12 after in vitro treatment with brefeldin A and monensin (Fig 7E and 7F). This was associated with control of chronic infection as brain tissue stained with antisera to parasite antigens revealed intact cysts with well-formed cysts walls (Fig 7G). In contrast, Mx1Cre-TLXf/f mice showed a near complete loss of IL-12 production by splenic dendritic cells (DCs) (Fig 7E and 7F) and had a marked increase in parasite burden in the brain, as demonstrated by the presence of parasites outside of cysts (Fig 7H, arrows). Furthermore, these animals presented with a higher parasite burden as assessed by qPCR for parasite DNA (Fig 7I). Taken together with our in vitro data, these results suggest that TLX plays a critical and previously unrecognized role as a STAT1 enhancer that is required for protection from Toxoplasma infection
This report presents the development and application of a host–pathogen interaction screen to interrogate a poorly understood aspect of STAT1 signaling: the regulation of STAT1 transcriptional activity in the nucleus. This strategy identified a novel set of genes that enhance STAT1 function and are highly enriched in transcription factors. The orphan nuclear receptor TLX was amongst the strongest STAT1 enhancers identified, and we have shown that TLX not only enhances the expression of endogenous STAT1 target genes following IFN-γ stimulation but is also required for control of Toxoplasma infection. Taken together, our data raise the possibility that TLX, as well as natural or synthetic ligands for this orphan nuclear receptor, represents an important new drug target to modulate cellular immunity and inflammation.
The use of transcription factors to elicit distinct transcriptional programs driving biological function is a common theme. This is particularly evident in the mammalian immune system, in which a small number of core transcription factors (e.g., STATs, IRFs, and NF-κBs) orchestrate diverse innate and adaptive responses. STAT1 exemplifies this plasticity in that it acts in different settings to promote both activated and suppressive macrophage and T cell function. The context-specific activity of transcription factors and signaling pathways that converge on them can be explained by a variety of mechanisms including epigenetic regulation [54], signal intensity or duration [55], and altered transcription factor complexes, since transcription factors form elaborate and dynamic multimeric complexes with co-activators/co-repressors and basal transcriptional machinery at the promoters of target genes [56]. Consistent with this notion, a recent study found that cell-type-specific responses mediated by transforming growth factor beta (TGF-β)-induced SMAD3 signaling is dictated by a small set of cell-type-specific transcription factors interacting with SMAD3 and directing its promoter binding activity [57]. Our data suggest that transcriptional modifiers also contribute to the context-specific activity of STAT1 and that TLX, in addition to its roles outside of STAT1 signaling, enhances the transcription of specific IFN-γ-regulated immune genes.
Little is known about the natural targets of TLX outside of neuronal progenitors or the involvement of TLX in immunity or as a regulator of STAT1 signaling. Our transcriptional profiling experiments reveal the spectrum of genes and functional categories regulated by TLX in two different cellular contexts. Interestingly, another nuclear receptor, the liver X receptor (LXR), in addition to its direct role in lipid metabolism, also binds to STAT1 during IFN-γ stimulation, resulting in suppression of STAT1 targets in macrophages, astrocytes, and microglial cells [58–60]. These data provide an important complement to the LXR/STAT1 interactions, by highlighting another druggable target that could drive enhanced STAT1 function in the brain, as well as in cells of the immune system. While TLX has been described as being exclusively expressed in only a select subset of cells in the brain, these data were largely derived from transgenic reporter mice [41]. Recent studies using immunohistochemistry [61] or qPCR [62] suggest that TLX may be expressed more widely than previously thought in the brain, as well as in a range of immune cell types, including CD4 and CD8 T cells, and monocytes [62].
In addition to defining TLX as a novel STAT1 enhancer, our screen revealed additional enhancers, a subset of which was previously reported to impact IFN-γ /STAT1 target genes. Upstream stimulatory factor-1 (USF1) and zinc finger X-linked duplicated family member C (ZXDC) have previously been shown to enhance IFN-γ-dependent transcription of major histocompatibility (MHC) class II [63–65]. USF1 forms a complex with STAT1 on the class II transactivator (CIITA) promoter, a well-known STAT1 target [63], and is proteolytically degraded by Chlamydia in order to block IFN-γ-induced transcription of STAT1 target genes [66,67]. In addition, ZXDC directly binds CIITA to enhance its function [64,65]. We also identified all three isoforms of ETS2 present in the library as STAT1 enhancers, which demonstrates the robustness of the screening assay. ETS2 physically interacts with CP300 and CBP [68]—both nuclear proteins that act as STAT1 enhancers by linking transcription factors to the basal transcriptional machinery [37]. Finally, another orphan nuclear receptor, COUPTF2, was identified in our screen. Interestingly, both TLX and COUPTF2 have recently been shown to share high homology, particularly in the region of the ligand binding domain, suggesting a functional relatedness.
While the genetic screen described in this report capitalizes on pathogen suppression of IFN-γ responses, a characteristic of a wide range of host–microbe interactions, these results highlight the potential of functional genomic screens to identify regulators of immune signaling pathways more broadly, representing a novel approach for the systematic identification of genes that modulate transcription factor activity. In addition, the data presented here specifically highlight the previously unrecognized role of the orphan nuclear receptor TLX in regulating STAT1 activity. Just as the glucocorticoid receptor and estrogen receptor are drug targets in inflammation and cancer, our data suggest that ligands for TLX may constitute new therapeutic targets to modulate inflammation and host defense in the brain.
RH strain Toxoplasma were maintained by serial passage in human foreskin fibroblast monolayers as described previously [69]. Me49 cysts were obtained from either Swiss Webster or CBA donor mice, enumerated by light microscopy, and used to infect mice by intraperitoneal injection of 20 cysts. TLXf/f mice [47] were a kind gift from Dr. Ron Evans (The Salk Institute) and were crossed to Mx1-Cre mice to create TLXf/f Mx1-Cre mice, allowing inducible deletion of TLX. TLX was deleted using 200 μg/mouse of Poly(I:C) (Imgenex) administered intraperitoneally every 3 d (five administrations total). Animals were rested for 10 d after the final dose before being infected with 20 Me49 cysts administered intraperitoneally. The human osteosarcoma (U2OS) and astroglioma (U251) cell lines were obtained from the American Type Culture Collection (ATCC) and maintained as recommended. This study was carried out in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. Protocols were approved by the Institutional Animal Care and Use (IACUC) committee of the University of Pennsylvania (animal welfare assurance number A3079-01). The University of Pennsylvania Animal Care and Use Programs are fully accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC).
The MGC is a publically available cDNA library of complete open reading frames from the mouse and human genome [35]. cDNAs are packaged in a cytomegalovirus (CMV) promoter-based overexpression vector (sport6 vector, Invitrogen). Screening of MGC v2 was carried out in 384-well luminescence CulturPlates (Perkin Elmer) prespotted with 40 ng/well of each cDNA clone [70]. U2OS cells (7500/well) were reverse transfected for 24 h with MGC cDNA (or empty sport6 vector for control wells) and the STAT1 luciferase reporter (40 ng/well; Panomics), using Fugene 6 transfection reagent (0.24 ul/well; Roche). Cells were infected with a 10:1 ratio of RH strain Toxoplasma for 2 h and subsequently stimulated with recombinant mouse IFN-γ (Peprotech) for 7 h. Plates were assayed by adding BriteLite luciferase reagent (PerkinElmer) and measuring luminescence on an Analyst HT (Molecular Devices) set for 0.1-s integration time. The entire MGC library was screened in duplicate. Cells, transfection reagent, and cytokines were dispensed to each plate using a Matrix WellMate (Thermo Scientific). To minimize evaporation and edge effects during incubation, plates were covered with metal lids with a rubber gasket.
For whole-genome expression microarray, RNA was isolated using the RNeasy Plus kit (Qiagen). Biotin labeled complementary RNA (cRNA) was generated using the Illumina TotalPrep RNA amplification kit. Total RNA and cRNA quality were assessed by Bioanalyzer (Agilent). Illumina HumanHT-12 version-4 expression beadchips were hybridized with cRNA from two biological replicates per condition and scanned on an Illumina BeadStation 500GX. Scanned images were converted to raw expression values using GenomeStudio v1.8 software (Illumina). Data analysis was carried out using the statistical computing environment, R (v3.0.2), the Bioconductor suite of packages for R, and RStudio (v0.97). Raw data were background subtracted, variance stabilized, and normalized by robust spline normalization using the Lumi package [71]. Differentially expressed genes were identified by linear modeling and Bayesian statistics using the Limma package [72,73]. Probes sets that were differentially regulated (≥1.5 fold, FDR ≦ 5%; after controlling for multiple testing using the Benjamini-Hochberg method [74,75]) were used for hierarchical clustering and heatmap generation in R. Clusters of co-regulated genes were identified by Pearson correlation using the hclust function of the stats package in R. Data have been deposited on the Gene Expression Omnibus (GEO) database for public access (GSE55751).
GO enrichment analysis was carried out using the Database for Visualization and Integrative Discovery (DAVID) [76,77]. Enrichment of GO terms was defined relative to the complete MGC library or the whole human transcriptome for screen data and microarray data, respectively. Protein domains were identified using Pfam [78]. Network analysis of protein–protein interactions among human orthologs of the 17 nonredundant, putative transcription factors and STAT1 was carried out using Cytoscape v2.6.3 [79,80] and the Michigan Molecular Interaction (MiMI) plug-in [81]. MiMI was queried to find all “nearest neighbor” genes shared by at least two of the 17 query genes. Zʹ-factor analysis of STAT1 reporter assay was used to optimize conditions for screen [34]. A modified robust Z-score was calculated for each cDNA as described previously [82]. All experiments were repeated two-to-four times. Means and standard deviations were calculated from biological replicates. Significance was determined using a Student’s t test. Statistical analysis and data visualization were performed with GraphPad Prism 4 and DataGraph 2.3 (Visual Data Tools).
At 4–8 wk postinfection, infected mice were perfused with 40 ml of ice-cold PBS to remove peripheral blood. For splenocyte preparation, spleens were dissected, dissociated, and subjected to hypotonic red blood cell lysis to generate a single cell suspension that was used for ex vivo cytokine analysis, splenocytes were stimulated for 4 h with brefeldin A and monensin, and then cells were rinsed, stained for surface markers with CD3-Pacific Blue, CD19-Pacific Blue, B220-Pacific Blue, CD11b-APC, and CD11c-PE-Cy7 at 4°C, and fixed with 4% PFA in PBS for 10 min at RT. Intracellular IL-12 staining with IL-12p40-PE was detected by staining in FACS buffer containing 0.5% saponin (Sigma, St. Louis, Missouri). DCs were identified as Dump (CD3, CD19, NK1.1)-, CD11c+. Data were collected on a BD LSRFortessa cell analyzer (BD Bioscience) and analyzed using FlowJo software (TreeStar, Ashland, Oregon). Antibodies were purchased from BD Biosciences (San Jose, California) and eBioscience (San Diego, California). For detection of phospho-STAT1, U2OS cells were infected for 2 h with Toxoplasma parasites engineered to express tdTomato and subsequently stimulated with 10 ng/ml of rIFN-γ (Peprotech). At various times poststimulation, cells were trypsinized (for flow cytometry only), fixed with 2% formaldehyde, permeabilized 10 min with cold methanol at 4°C, and stained with Alexa-488-conjugated monoclonal antibody specific for phosphorylated tyrosine residue 701 of STAT1 (clone 4a; BD Biosciences).
For validation of primary hits from the high-throughput screen, cDNA clones were expanded from bacterial stocks of the library, and DNA was isolated using a HiSpeed Maxi Kit (Qiagen). Each clone was sequence verified and retested in six replicate wells in 384-well format using the conditions described above for the full screen. Each clone was also tested for its ability to regulate STAT1 in uninfected cells, as well as its ability to trigger a control luciferase reporter lacking GAS elements. Additional reporters were also used for monitoring ISGF3 (Stratagene), IRF1 (ActiveMotif), NF-κB, SRF, and AP1 (Clontech), as described in the text.
U251 cells were plated at 200,000 cells per well in 6-well plates and allowed to adhere overnight. Cells were transfected with either 10 nM siRNA to human TLX (Qiagen; target 5ʹ-CCGGTTGATGCTAACACTCTA-3ʹ, sense 5ʹ-GGUUGAUGCUAACACUCUATT-3ʹ, and antisense 5ʹ-UAGAGUGUUAGCAUCAACCGG-3ʹ) or 10 nM siRNA to luciferase as a negative control in HiPerfect transfection reagent (Qiagen) and incubated for 72 h before stimulating with rIFN-γ (Peprotech) for 8 h. Total RNA was isolated and used for either qPCR or expression profiling by microarray as described below. For inhibitor experiments, U251 cells were treated with 20 μM famprofazone (Santa Cruz Biotechnology), for 4 h before stimulation with 10 ng/ml rIFN-γ. RNA was isolated 8 h poststimulation and used for qPCR.
Full-length human TLX (NR2E1) cDNA (GenBank accession BC028031) from the MGC was used as a template for PCR with primers specific for the DNA binding domain (forward: 5ʹ- ccatctcgagATGAGCAAGCCAGCCGGA-3ʹ; reverse: 5ʹ- ccattctagaTTAGCGGATGGTGGACGTCCG-3ʹ) and the ligand binding domain (forward: 5ʹ-ccatctcgagATGGAATCAGCTGCCAGACTTCTCTTCATGAG-3ʹ; reverse: 5ʹ- ccattctagaTTAGATATCACTGGATTTGTACATATCTGAAAGCAGTC-3ʹ). Primer nucleotides in bold indicate start or stop codons; underlined are restriction sites for directional cloning, and italics indicate a 4 nt pad region permitting efficient restriction of PCR amplicon ends. PCR products were gel purified, digested overnight with XhoI and XbaI, and cloned into the pCMV-Sport6 plasmid with DNA ligase (New England Biolabs). Inserts were sequence verified, clones were cultured overnight, and DNA was isolated by maxiprep. Truncation mutant constructs were then used in STAT1 luciferase reporter and qPCR assays as described above.
U2OS cells were transfected in 10 cm dishes with 6 μg/dish of either Sport6-empty control plasmid or Sport6-hNR2E1 using Fugene 6 reagent. Twenty-four hours post-transfection, cells were left untreated or were stimulated with 20 ng/ml of IFN-γ and then cross-linked with 1% formaldehyde for 10 min before quenching with 125 mM Glycine for 5 min. Cells were recovered by scraping, pooled, and 20–40 million cells were pelleted and treated with cell lysis buffer (10 mM Tris pH 8.0, 10 mM NaCl, and 0.2% NP-40) for 10 min on ice. Nuclei were lysed with 50 mM Tris pH 8.0, 10 mM EDTA, and 1% SDS for 10 min at room temperature. Lysate was diluted in immunoprecipitation buffer (20 mM Tris pH 8.0, 2 mM EDTA, 150 mM NaCl, 1% Triton X-100, and 0.01% SDS) and sonicated seven times for 30 s each time, with 1 min on ice between each sonication. Lysis and dilution buffers were supplemented with protease and phosphatase inhibitors. Sonicated samples were further diluted to 3.6 ml each, and 1 ml was used for immunoprecipitation overnight at 4°C with either rabbit monoclonal antibody to pSTAT1 (Cell Signaling Technology, clone 58D6) or control rabbit IgG (Cell Signaling Technology). Bound chromatin was pulled down with protein G Dynabeads (Life Technologies) for 3 hat 4°C. Beads were washed and incubated with elution buffer (50 mM Tris pH 8.0 and 10 mM EDTA) at 65°C for 30 min. Chromatin was recovered, and cross-links were reversed by overnight incubation at 65°C. Samples were treated with RNAse A for 2 h at 37°C, followed by treatment with proteinase K for 30 min at 55°C. DNA was purified over PCR purification columns (Qiagen) and used for qPCR with primers designed based on known binding sites of STAT1 in the promoters of CXCL9 and CXCL10 [51]. CXCL9 primers were forward: 5ʹ-CAGATCCAAGGGAATTTCTGC-3ʹ and reverse: 5ʹ-TGTGCCAAAGGCTATCAGTG-3ʹ. CXCL10 primers were forward: 5ʹ-TGCCCTGACAAACTAATGAGC-3ʹ and reverse: 5ʹ-CAAGGCATATTCTGCACCAG-3ʹ.
Whole brain recovered from Swiss Webster mice at 4–6 wk postinfection were either snap frozen in OCT embedding media or fixed in 10% neutral buffered formalin before embedding in paraffin. For frozen tissue, 5 μm sections were cut on a Leica CM 3050 S cryostat and stained with 5 μg/ml rabbit polyclonal antibody to the N-terminus of human TLX (LifeSpan Biosciences), and bound antibody was detected using 5 ng/μl of alexa-488-conjugated goat anti-rabbit secondary. For paraffin sections, antigen retrieval was carried by microwaving slides in citrate buffer (pH 6). Images were captured on Nikon E600 microscope outfitted with a Nikon Digital Sight DS-FI1 camera (bright-field) and a Roper Scientific Photometrics CoolSnap EZ camera (fluorescence).
To detect Toxoplasma by immunohistochemistry, paraffin sections were deparaffinized, rehydrated, and endogenous peroxidase blocked using 0.3% H2O2 in PBS for 10 min at room temperature. Sections were blocked with 2% normal goat serum before being incubated overnight at 4°C with rabbit polyclonal antibody specific for the SAG1 protein of Toxoplasma. Bound antibody was detected with 1 μg/ml biotinylated goat anti-rabbit IgG using the VectaStain kit (Vector Labs), followed by DAB substrate kit (Vector Labs) according to the manufacturer’s instructions. Sections were counterstained with hematoxylin and images collected on a Nikon E600 fluorescent microscope (Nikon, Tokyo, Japan) and analyzed using NIS-Element (Nikon). Quantification of parasite DNA by qPCR was performed as previously described [83]. Briefly, 50 mg of tissue was disrupted by repeated passage through an 18 gauge needle, and DNA was isolated using the High Pure PCR template preparation kit (Roche). Real-time PCR specific for the Toxoplasma B1 repeat region was used to quantify the amount of parasite DNA from 500 ng of DNA purified from tissue. Samples were amplified using Toxoplasma B1 primers (forward: 5ʹ-TCCCCTCTGCTGGCGAAAAGT-3ʹ and reverse: 5ʹ-AGCGTTCGTGGTCAACTATCGATTG-3ʹ) and Power SYBR Green PCR Master Mix and a 7500 Fast Real-Time PCR System. A standard curve prepared from known amounts of purified Toxoplasma DNA was used for quantification. Analysis was performed with system software v1.3.1 (Applied Biosystems, Warrington, United Kingdom).
|
10.1371/journal.ppat.1001318 | Dendritic Cells and Hepatocytes Use Distinct Pathways to Process Protective Antigen from Plasmodium in vivo | Malaria-protective CD8+ T cells specific for the circumsporozoite (CS) protein are primed by dendritic cells (DCs) after sporozoite injection by infected mosquitoes. The primed cells then eliminate parasite liver stages after recognizing the CS epitopes presented by hepatocytes. To define the in vivo processing of CS by DCs and hepatocytes, we generated parasites carrying a mutant CS protein containing the H-2Kb epitope SIINFEKL, and evaluated the T cell response using transgenic and mutant mice. We determined that in both DCs and hepatocytes CS epitopes must reach the cytosol and use the TAP transporters to access the ER. Furthermore, we used endosomal mutant (3d) and cytochrome c treated mice to address the role of cross-presentation in the priming and effector phases of the T cell response. We determined that in DCs, CS is cross-presented via endosomes while, conversely, in hepatocytes protein must be secreted directly into the cytosol. This suggests that the main targets of protective CD8+ T cells are parasite proteins exported to the hepatocyte cytosol. Surprisingly, however, secretion of the CS protein into hepatocytes was not dependent upon parasite-export (Pexel/VTS) motifs in this protein. Together, these results indicate that the presentation of epitopes to CD8+ T cells follows distinct pathways in DCs when the immune response is induced and in hepatocytes during the effector phase.
| Malaria causes the deaths of 0.5–2 million people each year, mainly in Africa. A safe and effective vaccine is likely needed for the control or eradication of this disease. Immunization by irradiated malaria-infected mosquitoes has been shown to protect people against malaria. Irradiated parasites do not divide and cause infection but are capable of activating specialized killer cells called CD8+ T cells, which can protect against live parasites. Because vaccinating people with irradiated mosquitoes is not practical, we wanted to understand which parasite molecules are targeted by CD8+ T cells. These molecules may then be formulated into a safe and effective vaccine. CD8+ T cells do not automatically recognize every parasite molecule, but instead fragments of parasite proteins must be displayed on the surface of infected cells to be seen by CD8+ T cells. Our data show that CD8+ T cells recognize parasite proteins secreted by the parasite into the infected cell. This suggests that such proteins could be important components of malaria vaccines.
| Immunization with irradiated Plasmodium sporozoites to induce sterile protection against live parasite challenge is a powerful model for malaria vaccination [1]. Protective immunity is mediated in part by CD8+ T cells specific for the circumsporozoite (CS) protein of Plasmodium [2], [3]. Plasmodium specific CD8+ T cells have been shown to be primed by dendritic cells (DCs) [4], [5], [6], [7]. In particular, we have found that after sporozoite inoculation into the dermis by infected mosquitoes, antigen is presented by DCs in the skin-draining lymph node to initiate the CD8+ T cell response [4]. Primed CD8+ T cells then exit the priming site and migrate to the liver where they can eliminate infection after recognizing antigen presented by hepatocytes [4]. Thus CD8+ T cell mediated immunity requires antigen presentation by two different cell types – DCs and hepatocytes. Determining how DCs and hepatocytes process and present Plasmodium antigens is essential for the rational identification of vaccine candidates. Since immunization with irradiated sporozoites represents the gold standard for malaria vaccination it is important to know which sporozoite antigens are presented by DCs. Perhaps more vital still, is to understand which molecules are presented by hepatocytes, as only those molecules presented to effector cells can be the targets of protective immunity.
Microbial and tumor epitopes presented by MHC class I usually derive from proteins in the cytosol that are proteolytically cleaved into small peptides by the proteasome. These peptides are translocated from the cytosol into the ER by the TAP transporter for loading onto class I MHC molecules, which then traffic towards the cell surface (reviewed in [8]). Many parasites, however, reside within a parasitophorous vacuole (PV) and their proteins are not necessarily secreted into the host cytosol. The processing and presentation of intracellular parasite antigens is therefore complex and still poorly understood. Toxoplasma gondii antigens have been reported to reach the cytosol for class I processing via fusion of the PV and the host ER; from the host ER antigens may be retrotranslocated into the host cytosol for processing [9]. Leishmania major antigens may bypass the host cytosol altogether as antigen presentation appears to be TAP independent. Instead it is believed that L. major-derived peptides are directly loaded onto MHC Class I in the phagolysosome [10].
The in vivo processing of Plasmodium sporozoite or liver stage antigens has not been studied. Unlike Toxoplasma or Leishmania, Plasmodium does not infect professional APCs and it is not known how DCs acquire sporozoite antigen. Likewise, the presentation of antigens by hepatocytes to effector cells is also poorly understood. In-vitro evidence suggests that hepatocytes are capable of presenting Plasmodium antigen and that this may be proteasome dependent [11], requiring the export of parasite antigen to the hepatocyte cytosol by unknown mechanisms. It has been proposed that Pexel/VTS motifs, known to be important for the export of proteins out of the PV in Plasmodium blood stages [12], [13], could also be involved in the transport of liver stage antigens to the hepatocyte cytosol for processing and presentation by class I MHC [14].
In this study we aimed to identify key cellular and molecular features of the antigen processing pathways employed by DCs and hepatocytes. We aimed to determine if Plasmodium CS processing requires the use of the cytoplasmic TAP dependent pathway to transport the processed epitope from the cytosol to the ER and allow binding of the peptide to class I MHC. In addition, we wanted to investigate whether the CS antigen is phagocytosed by presenting cells or if it is directly deposited or secreted into the cytosol of DCs or hepatocytes. To address these questions we generated P. berghei parasites that express a mutant CS protein containing the model SIINFEKL H-2Kb restricted epitope. Using this parasite in conjunction with knockout and mutant mice we have been able to generate the clearest picture to date of the processing of the CS protein from both sporozoite and liver stages.
A major obstacle to determining how Plasmodium antigens are presented to T cells is the lack of defined H-2b restricted epitopes which severely limits in vivo studies, as many transgenic mice, which are critical to study basic aspects of immunology, are generated on a C57Bl/6 (H-2b) background. To overcome this, we generated P. berghei CS5M parasites in which the endogenous CS gene was replaced with a modified CS gene carrying 5 mutations that changed the natural H-2Kd restricted epitope SYIPSAEKI to SIINFEKL, an H-2Kb restricted epitope (Figure 1A and B). P. berghei CS5M parasites were apparently normal as they infected mosquitoes and mice similarly to parental P. berghei ANKA (Figure S1). Most importantly P. berghei CS5M parasites stimulated a robust SIINFEKL specific response in C57Bl/6 mice upon immunization (Figure 1C), and activated SIINFEKL-specific CD8+ T cells from previously generated TCR transgenic mice [15] were able to eliminate the liver stages of P. berghei CS5M (Figure 1D).
It is important to emphasize that our approach differs significantly from the more common strategy of inserting an entire foreign gene into a parasite and then tracking the immune responses to the foreign molecule. In the P. berghei CS5M parasite SIINFEKL is inserted in place of a well-defined natural epitope, leaving intact the neighboring residues to ensure correct proteasomal processing, thus the model epitope is presented exactly as the natural CS epitope. This makes the P. berghei CS5M parasite an excellent system in which to study antigen processing and presentation. Moreover, we anticipate that P. berghei CS5M will be a powerful tool for use in future studies of antigen specific immune responses to malaria sporozoites.
We initiated our studies on the presentation of Plasmodium antigen by investigating whether DCs present irradiated sporozoite antigen via the canonical TAP dependent pathway. Wild type and TAP-1 deficient mice [16] were immunized intra-dermally in the ear with sporozoites and 2 days later CD11c+ DCs were isolated from the draining lymph nodes. To assess antigen presentation the DCs were co-cultured with CFSE-labeled SIINFEKL specific transgenic cells. Antigen presentation was quantified by measuring the expansion of the transgenic cell population 3 days after immunization. While DCs isolated from wild type animals induced extensive proliferation of the SIINFEKL specific cells, DCs from immunized TAP-1 deficient animals were unable to induce proliferation (Figure 2A). The failure of TAP-1 deficient DCs to induce proliferation could only be due to a processing defect as TAP-1 deficient DCs pulsed with exogenous SIINFEKL peptide were fully capable of inducing antigen specific T cell proliferation (Figure S2).
To determine if TAP-1 is required in vivo after immunization via the natural route of infection, wild-type and TAP-1 deficient animals that had received SIINFEKL specific TCR transgenic CD8+ T cells were immunized by the bites of irradiated mosquitoes infected with P. berghei CS5M parasites. We observed a robust antigen specific CD8+ T cell response after immunization of wild type mice; however, immunized TAP-1 deficient animals failed to mount a significant CD8+ T cell response in either the draining LN, spleen or liver (Figure 2B). Together these data indicate that the presentation of the CS protein by DCs is strictly TAP dependent.
Given that the priming of sporozoite specific T cells is TAP dependent, the CS protein must reach the cytosol of the DC for antigen processing. Since Plasmodium parasites have not been observed to productively infect DCs [17], [18] it is not obvious how sporozoite antigen accesses the DC cytosol. One possibility is that CS antigen from sporozoites is cross-presented via an endosome-to-cytosol pathway in which sporozoite antigen is phagocytosed and then retrotranslocated into the cytosol [19]. Alternatively, CS may be deposited in DCs during the process of cell traversal - a process in which sporozoites pass through the cytosol of cells, without forming a vacuole around themselves [20], [21], [22].
To distinguish between these possibilities we evaluated the induction of CD8+ T cell responses in animals which have a single-point mutation in the molecule Unc93B1 (3d mice). This mutation causes several impairments to endosome function including defects in signaling via the endosomal TLRs and in cross presentation [23]. We reasoned that if there were defects in T cell priming in these animals it would strongly indicate a role for endosomes in antigen processing by DCs. We found that DCs isolated from immunized 3d mice were less capable of priming SIINFEKL specific T cells in vitro compared to wild type controls (Figure 3A). This defect appears to be in the processing of antigen, as exogenous peptide is efficiently presented by DCs from 3d mice (Figure S2). Nonetheless, ex vivo antigen presentation assays provide only a snapshot of sporozoite antigen presentation at a single time point whereas we have recently shown that prolonged antigen presentation is required for full T cell priming [24]. Thus we assessed T cell priming in vivo after immunization by mosquito bites. We found that the difference observed in ex vivo experiments was amplified in vivo as 3d mice had severely decreased SIINFEKL specific responses in the spleen and liver compared to wild type mice (Figure 3B).
The role of endosomes in the presentation of sporozoite antigen by DCs was further confirmed in experiments in which cross-presenting DCs subsets were depleted in vivo by treatment with cytochrome c (cyt c; Figure S3) [25], [26], [27]. Upon taking up cyt c cross-presenting DCs retrotranslocate it into the cytosol where it can induce apoptosis. In contrast non cross-presenting cell subsets are unaffected as they break down any cyt c that has been taken up in lysosomes. In agreement with the data from 3d mice we found significant reductions in the priming of SIINFEKL specific T cells in cyt c treated animals after immunization via mosquito bites (Figure 3C). Together these data demonstrate that the majority of sporozoite antigen is probably processed via the endosome-to-cytosol pathway.
Given that the presentation of sporozoite antigen by DCs occurs via the endosome, we hypothesized that opsonization of parasites might enhance the priming of CD8+ T cells [28], [29]. Accordingly we incubated parasites with the anti-CS mAb 3D11 [30] prior to immunization. Unexpectedly, we found that opsonized parasites induced much reduced proliferation of CD8+ T cells compared to sporozoites treated with irrelevant antibody [31] (Figure 4). This intriguing result indicates that opsonization inhibits rather than potentiates the delivery of sporozoite derived CS protein to the DC class I processing pathway. This surprising result is not completely unprecedented – opsonized T. gondii parasites appear to be taken up by DCs via complement and Fc receptors and directed away from the cross presenting pathway and towards break down by lysosomes [9]. To determine if this occurs after opsonization of Plasmodium sporozoites we also treated sporozoites with F(ab′)2 fragments of the 3D11 mAb which cannot be recognized by Fc receptors and do not efficiently fix complement. However 3D11 F(ab′)2 fragments were as efficient as intact antibody at inhibiting T cell priming. Thus it may be that opsonization (and F(ab′)2 treatment) affect T cell priming by immobilizing parasites [32] and thus interfering with a number of processes which may be important for T cell priming. These include parasite migration to the skin draining lymph nodes, invasion of cells in the skin and the shedding of antigen from the sporozoite surface [4], [17], [33].
Because effector cells must kill infected hepatocytes, it is also required that hepatocytes present processed antigen to CD8+ T cells. Therefore, in addition to DCs, we were also interested in determining how hepatocytes process antigen for presentation to effector cells. To determine if antigen is processed by hepatocytes via the same endosome-to-cytosol pathway employed by DCs, activated SIINFEKL specific CD8+ T cells were transferred to TAP-1 deficient, 3d and cyt c treated mice that were subsequently infected with P. berghei CS5M parasites. The read-out for epitope presentation is T-cell mediated inhibition of liver stage development i.e. if the epitope is presented, activated CD8+ T cells will recognize it and will eliminate liver stage parasites. We also tried to visualize antigen presentation by immuno-fluorescence with the mAb 25-D1.16 which recognizes Kb-SIINFEKL complexes [34]; however, in common with other researchers we found that this technique was not sensitive enough to detect epitopes on the surface of parasite infected cells [35].
Using our in vivo functional assay we found that effector CD8+ T cells had no inhibitory effect on parasite development in the livers of TAP-1 deficient animals while they were fully capable of eliminating parasites in wild type mice (Figure 5A), clearly indicating that in hepatocytes, as in DCs, CS must reach the cytosol for antigen processing. However, in sharp contrast to DCs, we found that hepatocytes do not process antigen via endosomes since effector CD8+ T cells were capable of efficiently eliminating parasites from the livers of 3d or cyt c treated mice (Figure 5B and C). Thus hepatocytes unlike DCs do not appear to process antigen by an endosome to cytosol pathway, rather, hepatocytes present antigen that has been deposited or secreted by the parasite directly into the cytosol.
Our findings that antigen presentation in hepatocytes requires CS to enter the host cytosol but is independent of the endosomal pathway, raise the question as to how CS traffics to the hepatocyte cytosol. A previous report in which the 2 Pexel/VTS motifs in the N terminal domain of CS were mutated, suggested that CS export to the cytosol was eliminated in the absence of functional Pexel/VTS motifs [14]. To determine whether Pexel/VTS motifs are critical for the entry of CS into the class I processing pathway of infected hepatocytes we generated P. berghei CS5M parasites that carried mutations in key residues of both Pexel/VTS motifs as well as the SIINFEKL epitope (P. berghei CS5MΔP1–2; Figure S4). We mutated the Pexel/VTS sequences to the sequence that was previously suggested to abolish CS export into the cytoplasm of infected hepatocytes [14]. In fact we were able to observe punctate staining of CS in the cytosol of both P. berghei CS5M and P. berghei CS5MΔP1–2 infected Hepa1-6 cells (Figure 6A and B), and more importantly, we found that the P. berghei CS5MΔP1–2 parasites were killed as efficiently as P. berghei CS5M by effector CD8+ T cells (Figure 6C). This indicates that Pexel/VTS motifs are not required for the entry of CS into the cytosol of hepatocytes for antigen presentation to effector CD8+ T cells. However, in agreement with the previous study we did observe that parasites with mutated Pexel/VTS motifs in the CS protein have a ∼10-fold decrease in infectivity (Figure 6C). Finally we found that DCs efficiently present the epitope from the CS protein of parasites lacking the Pexel/VTS motifs (Figure 6D). This was not entirely unexpected as our previous findings suggested that DCs likely acquire the CS antigen by phagocytosis which is unlikely to be affected by host cell targeting sequences.
In this study we demonstrate that the process of antigen presentation required for the priming of sporozoite specific T cells and for the elimination of liver stage parasites are distinct. The difference in antigen presentation between DCs and hepatocytes has important consequences for malaria vaccine development based on irradiated sporozoites. If other Plasmodium antigens are processed similarly to CS, it is likely that DCs, which acquire antigens by phagocytosis, could stimulate T cell responses to a broad range of secreted and non-secreted antigens. In contrast hepatocytes can only present antigens that are secreted into the cytosol of infected or traversed cells; these antigens are, however, the potential targets of protective immunity as they induce effector cells to eliminate liver stage parasites. Thus, irradiated sporozoites may induce a range of irrelevant as well as protective immune responses. Moreover it is possible that irradiated sporozoites will fail to induce protective responses to various liver stage antigens presented by hepatocytes, that are not expressed by sporozoites. This appears to be the case for the liver stage antigen Hep17: irradiated sporozoites do not induce detectable Hep17 specific CD8+ T cells; however, vaccine-induced T cells specific for this antigen are protective against Plasmodium liver stages [36].
We observed that both T cell priming and parasite elimination by T cells were strictly TAP dependent. Thus in both DCs and hepatocytes antigen must reach the cytosol for presentation. In DCs this appears to occur via an endosome-to-cytosol pathway as determined by two independent in vivo methodologies: the use of 3d mice and treatment of mice with cyt c. However, unlike the defect in TAP1 deficient mice, the reduction in T cell priming in both 3d and cyt c treated mice was not complete. This may indicate that a small amount of antigen is directly deposited in the cytosol of DCs by traversing sporozoites. Alternatively cross-presentation may not be fully ablated in these models. 3d mice carry a single point mutation in one molecule (Unc93B1) which may retain some residual functionality [23], while the depletion of cross-presenting DCs by cyt c may not be absolute, particularly in the lymph nodes. The function of Unc93B1 in antigen presentation is not clear, though it may be involved in translocating elements of the cross-presentation machinery to the endosome similar to the way it mediates the movement of TLRs to endosomes [37]. An intriguing recent study showed that 3d mice were highly susceptible to T. gondii infection [38]. The authors suggest that this was not due to an impairment of CD8+ T cell control of parasites as the activation of CD8+ T cells appeared normal in 3d mice – however they were only able to look at bulk T cell populations and not antigen specific cells.
Further research will be required to determine what receptors DCs use to take up sporozoites and which pattern recognition molecules interact with sporozoites to facilitate cross presentation. One unexpected finding was that opsonization of sporozoites did not enhance the presentation of the CS antigen by DCs. One hypothesis is that opsonization may immobilize parasites [32] and thus interfere with a variety of processes that may be important for T cell priming including antigen shedding, and migration to the draining lymph nodes for presentation [4], [17], [33]. Alternatively opsonization may prevent parasites from infecting cells in the skin where they could continue to provide antigen to the immune system [18], [39]. The inability of DCs to present antigen from immobilized parasites may explain why irradiated parasites are capable of inducing a protective CD8+ T cell response, but heat killed parasites are not [3], [40]. These data also have important implications for vaccine design since they imply that there would be difficulties in priming or boosting sporozoite specific CD8+ T cell responses in individuals with high anti-CS antibody titers. Thus it may be hard to induce effective CD8+ T cell responses in individuals who have already been naturally exposed to parasites or immunized with vaccines such as RTS,S that are designed to induce strong anti-sporozoite antibody responses [41].
Using the 3d and cyt c treated mice we showed that in contrast to T cell priming, parasite elimination was unaffected in mice with reduced capacity to cross-present antigen. This is in agreement with the findings of a previous in vitro study [11] which found no evidence for endosomes having a role in antigen presentation by infected cells. The previous study also showed that proteasome and Golgi inhibitors blocked antigen presentation, which is compatible with our finding that antigen presentation occurs via the classical TAP-dependent pathway [11]. Together these data suggest that cell killing occurs only after direct antigen presentation by the infected hepatocyte itself.
A key direction for future research will be to identify how antigens enter the host cell for presentation. We were unable to find a role for Pexel/VTS motifs in targeting the CS protein to the host cell cytosol as suggested by a previous study [14]. Our data are based on fluorescence microscopy 6 hours post-infection when the highest amounts of CS can be observed in the cytosol [42], [43] and, more importantly, our functional assay to measure the elimination of parasites by T cells. The fact that Pexel/VTS motifs are not required for the entry of CS to the class I processing pathway suggests that liver stage proteins may be exported to the hepatocyte by other mechanisms. In particular, it suggests that the CS protein may contain another motif that facilitates its export out of the PV into the infected host cell. Alternatively, liver-stage antigens might also be exported to the class I processing pathway if the Plasmodium PV can fuse with the hepatocyte ER as appears to occur in Toxoplasma infected DCs [9].
Together our data provide the most complete description to date of the processing of sporozoite and liver stage antigen. Using the P. berghei CS5M parasite we have demonstrated that DCs cross-present sporozoite antigen via an endosome-to-cytosol pathway. Of most importance, we show that CS must be delivered to the hepatocyte cytosol for presentation to effector cells. If this is true for other antigens, it is likely that antigens secreted into the hepatocytes of either infected or traversed cells constitute the major targets of anti-liver stage CD8+ T cell mediated immunity. Secretion to the hepatocyte is likely a complex process given our finding that Pexel/VTS motifs are not required for the entry of CS to the class I processing pathway; however, unraveling this process will be key to the identification of vaccine candidates.
All animal procedures were approved by the Institutional Animal Care and Use Committee of the Johns Hopkins University (Protocol Number MO09H41) following the National Institutes of Health guidelines for animal housing and care.
5–8 week old female C57Bl/6 were purchased from NCI (Frederick, MD). TAP-1 deficient animals were purchased from Jackson (Bar Harbor, ME). Unc93B13d mice were obtained from the Mutant Mouse Resource Center (University of California, Davis, CA). OT-1 mice (carrying a transgene specific for the SIINFEKL epitope) were kindly provided by David Sacks (Laboratory of Parasitic Disease, National Institute of Allergy and Infectious Disease, Bethesda, MD).
P. berghei CS5M parasites were generated by transfection of P. berghei ANKA with the linearized pR-CS5M plasmid as previously described [44]. pR-CS5M was derived from the plasmid pR-CSwt [45] as follows. A Kpn1-Xho1 fragment including the entire CS gene was excised from pR-CSwt into a pBluescript SK- (Stratagene) backbone to generate the plasmid pIC-CSwt. A SexA1 site was introduced by mutation of G to A at position 714 in the CS gene (silent in Gln238) and a BsmF1 site was introduced by a mutation of T to C at position 810 (silent in Asp270) using the QuikChange XL site directed mutagenesis kit (Stratagene). The SexA1-BsmF1 fragment was excised and replaced with a ∼100 bp insert including the SIINFEKL epitope in place of the SYIPSAEKI sequence (formed from the oligos S8ins F and S8insR; see Table S1) to generate the plasmid pIC-CS5M. The Kpn1-Xho1 fragment from pIC-CS5M was excised and ligated into the backbone of pR-CSwt to generate the pR-CS5M plasmid. pR-CS5M was linearized with the enzymes Kpn1 and Sac1.
P. berghei CS5MΔP1–2 parasites were generated similarly to P. berghei CS5M (Figure S1). The plasmid pR-CS5MΔP1–2 was generated as follows. Arg32 and Leu34 in the CS gene on the pIC-CS5M plasmid were mutated to Alanines by using the QuikChange site directed mutagenesis kit with the primers PEXEL1 F and PEXEL1 R (see Table S1), which include a Bsm1 site. Arg66 and Leu68 were mutated similarly with the primers PEXEL2 F and PEXEL2 R that include an ApaB1 site. The resulting plasmid was designated pIC-CS5MΔP1–2. The Kpn1-Xho1 fragment of the pIC-CS5MΔP1–2 plasmid was ligated into the pR-CSwt backbone to generate the pR-CS5MΔP1–2 plasmid used for transfection.
Lymph node and spleen myeloid DCs were prepared essentially as described [46]. Briefly, spleens or lymph nodes from immunized mice or naive mice were taken, chopped finely and digested with 1 mg/ml collagenase. The single cell suspension of spleen cells was then separated over a Nycodenz gradient (density, 1.075 g/ml) and the DC-rich low-density fraction was taken. To further enrich the DC population, negative selection was performed on the collected fraction using magnetic bead separation with anti-CD3, anti-GR1, anti TER119, anti-B220 and anti-Thy1.2 antibodies. Final purity of CD11c+ DC was about 70%. To assess Ag presentation ex vivo, splenic myeloid DCs (1×105) were mixed with 5×104 purified naive CFSE-labeled CD8+-transgenic cells in a single V-bottom well of a 96-well plate. 60–65 h later, the cells were harvested, and CFSE dilution in the transgenic cell population was used as a measure of Ag presentation.
Where possible SIINFEKL-specific T cell priming was measured after immunization by the bites of 10–20 irradiated mosquitoes. Prior to biting, a low number (2×103) of CD45.1+ OT-1 cells were transferred to mice and the expansion of the CD45.1+ CD8+ (SIINFEKL-specific) population were measured by flow cytometry 10 days later to allow time for the responses to reach detectable levels. In some experiments it was necessary to perform immunizations with needle injected sporozoites (e.g. where the sporozoites were treated with antibodies prior to immunization). In these experiments 5×105 congenic CD45.1+ OT-1 cells were adoptively into mice, which were immunized the following day. The cells would be labeled with 0.6 µM CFSE using the Vybrant Cell Tracker kit according to the manufacturer's instructions (Invitrogen Life Technologies), and antigen presentation was inferred from proliferation of CD45.1+ CD8+ cells in the draining lymph nodes after 3 days. Use of a high number of transgenic cells is acceptable in these experiments as we are using the cells as a readout of antigen presentation not measuring particular T cell phenotypes. ELISPOTs to measure peptide-specific IFN-γ secreting cells were performed as described [47] and used to detect endogenous SIINFEKL responses.
F(ab′)2 fragments from the 3D11 mAb (class: mouse IgG1) were prepared by incubation with immobilized Ficin in the presence of 4 mM cysteine according to the manufacturer's instructions (Pierce). F(ab′)2 fragments were isolated from intact antibody and Fc fragments by passing twice over a Protein A column. Purity of F(ab′)2 fragments was verified by SDS-PAGE under non-reducing conditions.
SIINFEKL-specific effector cells were purified from mice that had received 5×105 naïve CD45.1+ OT-1 cells and then been immunized with 2×106 pfu VV-OVA [48]. 8–10 days later spleens were taken from the immunized mice and the lymphocytes were purified by spinning over lympholyte M (Cedarlane Laboratories). A total of 2×106 effector/SIINFEKL specific CD8+ T cells were transferred to each recipient mouse.
Quantification of liver stage parasites was performed as previously described [49]. Briefly, 40 hours after challenge, livers were excised and parasite load was determined by quantitative PCR for P. berghei 18S rRNA using SYBR Green (Applied Biosystems).
Single cell suspensions of lymphocytes were obtained by grinding spleen cells or lymph node cells between the ground ends of two microscope slides and filtering twice through 100 µm nylon mesh. Liver lymphocytes were isolated from perfused livers by grinding, filtration through a 70 µm mesh and separation over a 35% percol gradient as described [50].
Hepa1-6 cells were grown on coverslips in a 48 well plate and allowed to reach ∼80% confluence prior to infection with ∼3×104 parasites. 6 hours later the slides were washed and fixed for 15 minutes with 4% formaldehyde prior to permeablilization with 100% methanol for 10 minutes. The cells were then blocked with 3% BSA for 45 minutes. The parasite cytosol was labeled with anti-Plasmodium HSP70 mAbs [51] followed by secondary staining with Alexa594 anti-mouse IgG (Molecular Probes). The cells were then stained with anti-P. berghei CS mAb (3D11) directly conjugated to FITC. Slides were mounted with ProLong antifade with DAPI (Molecular Probes). Images were acquired on a Nikon Eclipse 90i microscope with a Hamamatsu Orca-ER camera attachment using Volocity software (Perkin Elmer). Images were analyzed and assembled using ImageJ software (open source from NIH).
Statistical analysis was performed using Prism 4 software (GraphPad Software), unless otherwise stated, means were compared by two-tailed Student's t tests. Analysis of all flow cytometry data was performed using FlowJo software (TreeStar).
|
10.1371/journal.pgen.1000846 | Nuclear Pore Proteins Nup153 and Megator Define Transcriptionally Active Regions in the Drosophila Genome | Transcriptional regulation is one of the most important processes for modulating gene expression. Though much of this control is attributed to transcription factors, histones, and associated enzymes, it is increasingly apparent that the spatial organization of chromosomes within the nucleus has a profound effect on transcriptional activity. Studies in yeast indicate that the nuclear pore complex might promote transcription by recruiting chromatin to the nuclear periphery. In higher eukaryotes, however, it is not known whether such regulation has global significance. Here we establish nucleoporins as a major class of global regulators for gene expression in Drosophila melanogaster. Using chromatin-immunoprecipitation combined with microarray hybridisation, we show that Nup153 and Megator (Mtor) bind to 25% of the genome in continuous domains extending 10 kb to 500 kb. These Nucleoporin-Associated Regions (NARs) are dominated by markers for active transcription, including high RNA polymerase II occupancy and histone H4K16 acetylation. RNAi–mediated knock-down of Nup153 alters the expression of ∼5,700 genes, with a pronounced down-regulatory effect within NARs. We find that nucleoporins play a central role in coordinating dosage compensation—an organism-wide process involving the doubling of expression of the male X chromosome. NARs are enriched on the male X chromosome and occupy 75% of this chromosome. Furthermore, Nup153-depletion abolishes the normal function of the male-specific dosage compensation complex. Finally, by extensive 3D imaging, we demonstrate that NARs contribute to gene expression control irrespective of their sub-nuclear localization. Therefore, we suggest that NAR–binding is used for chromosomal organization that enables gene expression control.
| The eukaryotic genome is spatially distributed in a highly organized manner, with chromosomal regions localizing to well-defined sub-nuclear positions. This organization could have a profound effect on chromatin accessibility and transcriptional activity on a genome-wide level. Using high-resolution, genome-wide, chromatin-binding profiles we show that the nuclear pore components Nup153 and Megator bind to quarter of the Drosophila genome in form of chromosomal domains. These domains represent active regions of the genome. Interestingly, comparison of male and female cells revealed enrichment of these domains on the male X chromosome, which represents an exceptionally active chromosome that is under dosage compensation control to equalize gene expression due to differences in X chromosome number between males and females. Based on extensive 3D image analysis, we show that these chromosomal domains are contributed by both peripheral as well as intranuclear pool of these proteins. We suggest that chromosomal organization by nucleoporins could contribute to global gene expression control.
| The spatial organisation of DNA, both at the nucleotide and chromosomal levels, allows efficient storage of genetic information inside the nucleus. However, DNA-dependent processes such as transcription, require the chromosomal structure to be modified in order to allow access to this information.
The regulation of chromatin accessibility is an intensely studied subject [1],[2]. Molecular and genomic investigations have examined how nucleotide sequences and ATP-dependent chromatin-remodelling enzymes specify the locations for nucleosomal-binding, and how histone-modifying enzymes modulate the stability of histone-nucleic acid interactions. These enzymes are recruited to precise genomic loci with the aid of sequence-specific DNA-binding transcription factors. In turn, particular histone modifications influence transcription factor-binding to target sites on the genome, so controlling transcriptional initiation. Despite the importance of these cis- and trans-acting factors on the local chromosomal environment and the transcription of nearby genes, it has become increasingly clear that they explain just one level at which chromatin is regulated [3],[4].
The eukaryotic genome is spatially distributed in a highly organised manner, with entire chromosomal regions localising to well-defined sub-nuclear positions [5]. This organisation has a profound effect on chromatin accessibility and transcriptional activity on a genome-wide level [6]–[8]. For instance, chromosomal regions at the nuclear envelope tend to form closed heterochromatin, a structure that is generally indicative of transcriptional repression [9]. Genomic studies in Drosophila melanogaster and humans established that lamins—proteins lining the nuclear membrane [10]—are major contributors to sub-nuclear localisation and gene regulation [11],[12]. Comparisons of binding profiles with gene expression data and histone marker information showed that chromosomal regions containing dense lamin-binding were transcriptionally repressed.
Although the nuclear periphery has been primarily associated with repression, recent evidence has also suggested a role for membrane components in transcriptional activation [9], [13]–[16]. The nuclear pore complex is a large structure comprising about 30 protein subunits, and it is the primary channel through which macromolecules traverse the nuclear envelope [17]. Interestingly, investigations in Saccharomyces cerevisiae identified subunits of the nuclear pore complex that preferentially bound transcriptionally active genes [18]. Moreover, several target loci such as GAL2 and INO1 were found to relocate from the interior to the periphery upon activation [13], although there were exceptions to this behaviour [19]–[22]. Thus, it is becoming increasingly clear that nuclear periphery components can have both positive and negative influence on gene regulation.
Since there are differences in the composition of the nuclear envelope—such as the lack of lamins—it is important to also study the contribution of nuclear envelope components in gene regulation in higher organisms [9], [17], [23]–[25]. So far just one study has explored the global interactions of nucleoporin subunit Nup93 with human chromosomes 5, 7 and 16 [26]; the publication reported only a low density of binding sites, and their influence on gene regulation was inconclusive.
Recently, we revealed a biochemical association between nucleoporins and the dosage compensation apparatus in higher eukaryotes including humans [27]. In Drosophila, the Male Specific Lethal (MSL) complex offsets the imbalance in the number of sex chromosomes in males and females by doubling the expression of genes on the male X chromosome [28],[29]. By purifying enzymatically active MOF complexes, we identified interactions with the nucleoporins Nup153 and Megator (Mtor). Strikingly, depletion of either subunit resulted in the loss of dosage compensation in male cells. Therefore, our work suggested a vital role for nucleoporins in promoting transcriptional activation on a large-scale.
Here, we present the first genome-wide study of nucleoporin-binding in a higher eukaryote. Using chromatin immunoprecipitation followed by hybridisation to high resolution tiling microarrays, we show that Nup153 and Mtor interact with 25% of the Drosophila genome in large domains spanning 10–500 kb in size. These regions—which we term nucleoporin associated regions (NARs)—contain large numbers of highly expressed genes, and are enriched for markers of active transcription including RNA polymerase-binding and histone H4 lysine 16 acetylation. Additionally, we reveal a remarkably high density of NARs on the male X chromosome, which correlate extremely well with the binding pattern of the dosage compensation complex. Finally, we demonstrate that chromosomal regions bound by these nucleoporins are composed of peripheral as well as non-peripheral pools of these proteins but interestingly the X chromosomal target regions are preferentially localised closer to the nuclear periphery. In summary, we firmly establish nucleoporins as a major class of chromatin-binding proteins in higher eukaryotes, with a general role in transcriptional regulation and three-dimensional chromosomal organisation. Finally we show for the first time, the importance of nucleoporin-binding not only as a mechanism for transcriptional control, but also in maintaining a complex organism-level biological system namely dosage compensation.
We produced DNA-binding profiles for nuclear pore components Mtor and Nup153 in Drosophila male SL-2 and female KC cell lines using chromatin immunoprecipitation followed by hybridisation to Affymetrix tiling arrays [30],[31] (Figure 1). Raw data were processed as in Kind et al (2008) to minimise false-positive signals from aberrant array probes (Figure S1) [32].
The ChIP-chip profiles for the two proteins strongly correlate, indicating they bind to similar locations throughout the genome (r = 0.77 and 0.88 for SL-2 and KC cells respectively; Figure 1D, Figure S4). We confirmed the reproducibility of results by performing three biological replicates for each condition (r = 0.73), and we validated binding at 18 control genes by real-time PCR in triplicate (Figure S2).
Both Mtor and Nup153 exhibit extensive binding across the whole genome, and together they bind to 42% of the Drosophila genome (calculated as a fraction of base-pairs covered with two-fold cut-off). Thus nucleoporins represent a new class of global chromatin-binding proteins for higher eukaryotes.
Visual inspection of the ChIP-chip profiles reveals that Nup153 and Mtor interact with the genome in a manner not observed for traditional transcription factors (Figure 1B and 1C) [33]. Instead of associating with discrete loci, nucleoporins bind extended chromosomal regions that alternate between domains of high-density binding with those of low occupancy.
In order to analyse the visual observations in a statistically rigorous fashion, we quantified binding that takes place within a 10 kb sliding window that was scanned along the genome (see Materials and Methods). Windows containing more than 70% binding (as a proportion of array probes with positive binding signal) were classified as Nucleoporin Associated Regions (NARs), and neighbouring windows reaching this threshold were grouped together as continuous NARs. The detection method is robust: the 70% threshold ensures that no NARs are found when binding sites are randomly distributed across the genome and we identify very similar sets of NARs for windows ranging 5 kb to 500 kb in size. Moreover, application of the domain-finding approach described by Guelen et al [11] returns over 80% agreement with our method (in terms of base-pairs classified as NARs).
There is considerable NAR-occurrence (Figure 1A–1C); in male SL-2 cells, a total of 1,384 NARs cover a quarter of the entire Drosophila genome (25Mb and 29Mb for Nup153 and Mtor respectively) and in female Kc cells 1,865 NARs occupy a similar proportion of the genome (33Mb and 35Mb for Nup153 and Mtor respectively; Figure S3). Most domains range in size from 10 kb to 100 kb, although some even extend to over 500 kb (Figure 1F, Figure S4). Most nucleotide positions within NARs are occupied by both Nup153 and Mtor. Moreover, even where the overlap is not perfect, NARs tend to occur in similar genomic loci (Figure 1E; Figure 1B chromosomal positions 560,000–600,000). Most importantly, NARs occur in gene-rich areas that encompass over 4,700 protein-coding genes whose activities might be affected by nucleoporin-binding.
A direct relationship between nucleoporin-binding and gene expression has not been established so far in higher eukaryotes. Therefore, we explored the impact of NARs on transcriptional regulation by examining the activity of genes encoded within these regions (Figure 2; Tables S1, S2).
We measured gene expression levels using Affymetrix GeneChips (see Materials and Methods). Using the present-absence calls defined by the MAS5.0 algorithm [34], we detected the expression of 6,478 and 6,219 genes in SL-2 and Kc cells respectively. These genes are preferentially located within NARs: 63% of genes inside NARs are expressed compared with just 40% outside, indicating a significantly elevated transcriptional activity in the former (p-value <2.2e−16). This observation is supported by data quantifying RNA polymerase II-occupancy (Figure 2; Tables S1, S2); by mapping publicly available ChIP-chip data [35], we find the Pol II-binding is highly enriched at the promoters of genes inside NARs compared with those outside (p-value <2.2e−16).
Recent publications demonstrated that histone modifications, MOF acetyltransferase- and lamin-binding are robust genome-wide indicators of transcriptional activity. In both SL-2 and KC cells, acetylated histone H4 lysine (H4K16Ac) and MOF-binding [32]—strong markers for active transcription—are extremely prominent within NARs (Figure 2; Tables S1, S2; p-value <2.2×10−16). In contrast, histone H3 lysine 27 tri-methylation [36] and lamin-binding [12]—markers of transcriptional repression—are enriched outside NARs (Figure 2, Figure S5; Tables S1, S2; p-value <2.2e−16).
Finally, we confirmed a causal link between nucleoporin-binding and transcriptional regulation by measuring gene expression levels following RNAi-mediated knock-down of Nup153 (Figure 2, Figure S7; Tables S1, S2). The depletion results in large and wide-spread transcriptional changes in cells collected after seven days: 5,684 genes −40% of Drosophila genes represented on the array—are differentially expressed in SL-2 cells (p-value <0.05). Moreover, there is a large enrichment of down-regulated genes within NARs (29% of all genes; 40% of ‘present’ genes) compared with non-NARs (19% of all genes; p-value <2.2e−16). We obtain similar enrichments for cells collected five days after RNAi-treatment, and also upon Mtor-depletion (data not shown). These observations strongly indicate that nucleoporin-binding promotes a high-level of transcriptional activity, which may be due to the formation of an open chromatin environment.
One of the most important manifestations of gene expression control in higher eukaryotes is dosage compensation for different number of sex chromosomes between the two sexes. In Drosophila—in which females have two X chromosomes but males possess only a single X—the dosage compensation complex offsets the imbalance in gene content by doubling the expression of the male X chromosome. Thus, the chromosome represents an outstanding example of an exceptionally highly transcribed genomic region.
In order to explore the association of Nup153 and Mtor with the dosage compensation complex further, we compared the patterns of nucleoporin-binding in male SL-2 and female Kc cells (Figure 1A, Figure 3A–3D, Figure S3). There is a dramatic difference between the two sexes: in females, NARs are evenly distributed throughout the entire genome with only a 1.2-fold difference in % NAR occupancy between chromosome X (7.4Mb and 33% for Nup153; 8.0Mb and 36% for Mtor) and autosomes (26.0Mb and 27% for Nup153; 27.1Mb and 28% for Mtor); but in males, NARs are overwhelmingly biased towards the X chromosome (14.9Mb and 67% for Nup153; 16.6Mb and 75% for Mtor) compared with the autosomes (9.7Mb and 10% for Nup153; 12.0Mb and 12% for Mtor) with a 6-fold difference in occupancy. Further, domains on the male X chromosome (median length = 62Kb, 94Kb for Nup153 and Mtor respectively) are much longer than those found on any other chromosomes (median length = 22Kb for Nup153 and Mtor in male autosomes, ∼35Kb for female autosomes and X chromosome).
Having established that the nucleoporins are enriched on the male X chromosome, we explored the association with the dosage compensation system further. Recently, we demonstrated that the members of the dosage compensation complex—MSL1, MSL3 and MOF—preferentially bind to the male X chromosome [32]. A comparison of this previously published dataset with our current analysis shows that NARs on the male X chromosome coincide very well with the binding sites of the dosage compensation complex (Figure 3E).
We also tested the effects of Nup153-depletion on MSL1 and MOF-binding to 10 known target loci using chromatin-immunoprecipitation followed by qPCR. X-chromosomal binding is severely reduced for both proteins (Figure 3F), and the additional binding to autosomal targets is lost for MOF (Figure S8). The effects are clearly specific to Nup153, as depleting another nucleoporin, Nup50 does not influence MSL1 and MOF-localisation and binding (Figure S9; data not shown). Moreover, the observations are not due to an effect on MSL protein concentrations or defects in the RNA export pathway [27]: we previously showed that MSL levels remain unaffected in Nup153 and Mtor-depleted cells; and impairment of the major export pathways through NFX1-depletion does not disrupt the localisation of the MSL complex to the X chromosomes.
Although nucleoporins are primarily located at the nuclear periphery, some display dynamic association with the nuclear pore complex [37], and it remains unclear whether nucleoporin-chromatin interactions would affect transcription at the periphery or within the nucleoplasm. Therefore, we assessed the spatial localisation of different chromosomal regions within the nucleus using three-dimensional imaging of Fluorescence In Situ Hybridisation (3D-FISH) in male and female cells (Figure 4). We selected 26 chromosomal regions of average length 15–20 kb for analysis (Table S3), comprising 18 NAR (targets T1-18) and 8 non-NAR loci (targets N1-8). An independent lamin-bound locus (target L105) was used as a positive control representing a region previously shown to localise at the nuclear periphery [12].
First we checked the localisation of Nup153 and Mtor themselves (Figure S6). Immunostaining of SL-2 cells and salivary glands from male larvae confirm that both proteins predominantly reside in the nuclear periphery, although we also detected some staining within the nucleus. This is consistent with earlier reports that these proteins are dynamic components of the nuclear pore complex, with the capacity to shuttle between different sub-nuclear locations [25],[37].
Next, we used DAPI and lamin protein-immunostaining to assess the nuclear localisation of our target loci. We display a selection of images in Figure 4A: the lamin protein in green defines the nuclear boundary, the DAPI in blue the distribution of genomic DNA, and the FISH signal in red specifies the position of the target locus. In order to account for cell-to-cell variation in localisation that results from the dynamic behaviour of chromatin, we measured the distance between the FISH signal and nuclear boundary for a large number of samples (44<n<91). Size differences between nuclei were normalised by representing distances as a percentage of the nuclear radius. In Figure 4B, we show the expected distribution of distances for a simulated locus situated at the periphery; for a FISH signal with 30% radius, we find that most measurements lie between 0% and 30% of the distance to the centre of the nucleus. In contrast simulations for a signal positioned halfway between the periphery and the centre results in a distinct, more symmetrically shaped distribution, with most measurements falling between 20% and 60% of the distance to the centre (Figure 4C; Figures S10, S11; Videos S1, S2, S3, S4).
The lamin-bound L105 locus displays a distribution that is heavily skewed towards the periphery (Figure 4D); however the profile is broader than the simulation, signifying that the locus is present at the interior of the nucleus at least part of the time. On the other hand, target N2 resembles that of the non-peripheral simulation (Figure 4E), albeit with a broader distribution, which indicates that the locus predominantly resides in the interior. Since both loci are NAR-independent, they were assigned as in vivo controls representing peripheral and non-peripheral localisation.
Many NAR-target distributions show almost perfect overlap with L105, demonstrating that they are preferentially situated at the periphery (Figure 4F–4G; see Materials and Methods); interestingly however a subset of NAR loci displays distributions that are indicative of non-peripheral localisation (Figure 4I). For non-NARs, targets such as N1 display good overlap with the negative control N2 (Figure 4H), but some are found at the periphery. It is clear, therefore, that many targets regions tested here do not conform to the behaviour expected from NPC-binding.
In fact, we find that NARs from chromosome X tend to reside at the periphery (6 out of 10 targets; Table S4), whereas only a small number of autosomal NARs do so (1 out of 8; Table S4). This is reflected in the aggregate distributions, in which X-chromosomal loci display the characteristic skewed profiles compared with autosomal regions (Figure 4J–4L). Among non-NARs (Figure 4K–4M), autosomal loci are invariably non-peripheral, whereas the X chromosomal targets display a tendency for peripheral localisation; the positioning of the latter is probably influenced by neighbouring NARs as there is such a large amount of binding on the X chromosome. For comparison, peripheral localisation of the X chromosome is absent in female Kc cells (data not shown). Thus in striking contrast to prior expectations, we reveal that interior as well as peripheral populations of nucleoporins bind chromatin and mediate transcriptional activity at NARs. Furthermore, interactions with the X chromosome promotes peripheral localisation of the chromosome—most likely as a result of the overwhelming amount of binding in males—but this is generally not the case for autosomes.
Finally to confirm the influence of nucleoporins on localisation, we tested the effects of RNAi-mediated Nup153-knockdown for six loci: three peripheral X chromosomal NARs (T4, T5, T7), a non-peripheral X chromosomal NAR (T11), a non-peripheral autosomal NAR (T9) and the non-peripheral control (N2). For each we compared the distribution of Nup153-depleted samples against a mock EGFP RNAi-treatment (Figure 5, Figure S7). All three peripheral targets on the X chromosome displace to a more intra-nuclear position upon loss of Nup153 (Figure 5A–5C; p-value <0.05), but in contrast there was no significant change for any of the non-peripheral loci (Figure 5D–5F; p-value >0.05). These data suggest that the sub-nuclear positioning of peripheral NARs—specifically those on the male X—depends on the presence of Nup153, whereas the localisation of intra-nuclear loci is independent regardless of whether they are bound by nucleoporins.
The classical view of transcriptional regulation describes the interplay of transcription factors, histones and associated enzymes with DNA in order to recruit the transcriptional machinery to the appropriate genomic loci. However, it has become increasingly clear that these interactions explain only one level at which gene expression is controlled. At a genome-wide level, the spatial organisation of chromosomes within the nucleus is increasingly considered to have a profound effect on chromatin structure and transcriptional activity [5]. In particular, studies in yeast indicate that members of the nuclear pore complex might promote transcription by recruiting chromatin to the nuclear periphery [14],[18]. However, the importance of such regulation in higher eukaryotes has remained unresolved [26].
In this study, we established conclusively that nucleoporins play a central role in mediating transcriptional regulation in a complex, multicellular organism. For the first time in any higher eukaryote, we generated a genome-wide profile of nucleoporin-binding; contrary to preliminary observations, binding is widespread, occurring across 40% of the genome. Thus, we reveal that nucleoporins—Nup153 and Mtor in particular—represent a major new class of global chromatin-binding proteins.
Intriguingly, these proteins interact with the genome differently to traditional transcription factors. Rather than associate with individual loci, nucleoporins bind continuous sections of chromosomes at very high density. Termed NARs, these regions extend up to 500kb in length and occupy 25% of the entire Drosophila genome. Moreover, NARs are functionally important as they demarcate regions of open chromatin and transcriptional activity, which is lost on depletion of Nup153. It is significant that the male X chromosome—a prime example for hyper-transcription—is almost entirely occupied by NARs. Therefore, we suggest that Nup153 and Mtor may stimulate transcription by promoting the formation of an open chromatin environment.
In dramatic contrast to expectations, nucleoporin-binding does not automatically lead to localisation at the nuclear periphery, though the male X chromosome is an exception in this regard. Since Nup153 and Mtor are known to be dynamic components of the nuclear pore complex, it appears likely that both peripheral and intra-nuclear pools of nucleoporins contribute to chromatin-binding. Given the dynamic nature of chromatin-localisation, it is also possible that NARs are located at the periphery in a very transient manner, and further developments in imaging techniques will help clarify this. Where NAR-formation and peripheral localisation do coincide however, Nup153 is necessary for sustained positioning.
Chromosomal domains have been implicated in the formation of three-dimensional sub-nuclear structures to coordinate the expression of otherwise distant loci [38] such as the human beta-globin genes [39],[40]. We speculate that NARs may indicate the genomic regions required for the assembly of these transcription factories on a very large scale. Within this context, the dynamic nature of Nup153 and Mtor is significant, as re-localisation of these proteins might allow a basis for global transcriptional control in response to cellular cues. Additionally, given the primary function of the nuclear pore complex in transporting macromolecules to and from the nucleus, Nup153 and Mtor may provide a means to couple transcriptional control with post-transcriptional events. We stress however that the mechanisms behind such processes are the subject of intense research activity and many controversies remain.
Finally, the special link with dosage compensation confirms the importance of nucleoporin-binding not only as a molecular mechanism for transcriptional control, but also in maintaining a complex, organism-level biological system.
Chromatin immunoprecipitation combined with microarray hybridisation (ChIP-chip), and qPCR experiments were performed as described previously in Kind et al [32]. Primer sequences are provided in Text S1.
Numerical data from Affymetrix Drosophila Tiling 2.0R Arrays (Dm35b_MR_v02) were processed as in Kind et al [32]. Briefly, array data were background corrected using GCRMA and quantile normalised [41]. Log2 (ChIP/input) ratios were calculated using the average from three replicate experiments. Log2 ratios were then smoothed by averaging the signal within a 500 bp window centred on each probe (Figure S1).
Chromosomal regions with high densities of Nup153- and Mtor-binding were identified by sliding a 10 kb window along each chromosome, centred on the start position of each probe. NARs were defined as continuous chromosomal regions containing positive binding signal (ie, log2 ratio >0) for more than 70% of probes. We also implemented the two-stage domain-finding method described by Guelen et al [11]. Our method recovered at least 80% of all probes defined as domains by the Guelen approach.
Nup153 and Nup50 were depleted as previously described in Mendjan et al [27]. Briefly, cells were incubated with dsRNA for five or seven days with a boost on day two. Cells were subsequently harvested for Western blot analysis, ChIP, gene expression profiling, or immunofluorescence experiments. Control experiments were performed using mock treatment (EGFP RNAi).
Gene expression was measured using Affymetrix Drosophila2 GeneChips in triplicate for each condition. Data analysis was performed using publicly available packages in the BioConductor Software Suite [43]. Raw .CEL files were processed using RMA [44] and probe-sets were mapped to genes using the annotation from the Ensembl database (v41) [45].
In control (EGFP-treated) cells, expressed genes were identified as those outputting MAS5.0 ‘present’ cells in all three replicates [34]. For comparisons of Nup153-depleted and mock-treated cells, differentially expressed genes were determined using the Limma package [46]; p-values were corrected for multiple-testing using FDR [47] and a significance threshold of p-value<0.05 was selected.
We compared the overlap between NARs and genomic features. For ease of comparison, all data were mapped onto the Drosophila genome provided by the Ensembl database (v. 41) [45]. Accompanying each entry is the statistical significance of the difference in the amount of genomic feature found within NARs and non-NARs.
(i) Histone H4 lysine K16 acetylation (H4K16Ac; p<2.2e−16; t-test): processed ChIP-chip profiles obtained from Kind et al [32]. (ii) MOF-binding (p<2.2e−16; Fisher test): processed ChIP-chip profiles obtained from Kind et al [32]. (iii) RNA PolII-occupancy (p<2.2e−16; Fisher test): PolII-bound genes obtained from Muse et al [35]. For visualisation purposes in Figure 2, bound genes were represented as 1kb windows centred on the transcription start site. (iv) Gene density (p-value <2.2e−16; Wilcoxon test): number of genes as annotated by the Ensembl database within a 20kb sliding window with a 1 kb offset. (v) Expressed genes (p-value <2.2e−16; Fisher test): gene expression measured using Affymetrix Drosophila2 GeneChips as described above. (vi) Down-regulated genes upon Nup153-depletion (p-value <2.2e−16; Fisher test): differentially expressed genes in RNAi-treated cells compared with untreated cells as described above. (vii) Lamin-binding (p-value <2.2e−16; Fisher test): processed ChIP-chip data were obtained from Pickersgill et al [12]. Note that the study used low-resolution cDNA arrays, and therefore unlike the human study, the authors were unable to detect high-density lamin-associated domains. (viii) Histone H3 lysine 27 tri-methylation (H3K27me3; p-value <2.2e−16; Fisher test): processed ChIP-chip profiles obtained from Schwartz et al [36].
DNA FISH on SL-2 cells was performed as previously described by Lanzuolo et al [48]. Briefly for DNA FISH 1×106 cells were centrifuged, re-suspended in 0.4 ml of medium and placed for 30 min at room temperature on a poly-lysine-coated slide (10 mm diameter). After rinsing with PBS, the cells were fixed with 4% paraformaldehyde in PBT (PBS, 0.1% Tween 20) for 10 min at room temperature. Cells were then washed three times with PBT, incubated for 1 h at room temperature with RNAse A (100 µg/ml in PBT). After rinsing with PBS, cells were incubated with 0.5% Triton in PBS for 10 min at room temperature. Cells were rinsed again with PBS and incubated with 20% glycerol in PBS for 30 min at room temperature. Cells were then frozen in liquid nitrogen, thawed at room temperature and soaked in 20% glycerol in PBS, repeatedly four times. After washing the cells again with PBS three times, they were incubated for 5 min in 0.1N HCl, briefly rinsed in 2XSSC twice, and stored in 50% formamide, 2XSSC, 10% dextransulphate, pH 7.0. Fluorescent probes were prepared with the FISH Tag DNA Kit (Invitrogen, Carlsbad, CA), dissolved in the hybridization mixture (50% deionized formamide, 2XSSC, 10% dextransulphate, salmon sperm DNA at 0.5 mg/ml), applied to cells and sealed under coverslips with rubber cement. Probe and cellular DNA were denatured simultaneously on a hot block at 78°C for 3 min. Hybridization was carried out in a humid atmosphere at 37°C for 1 d. After hybridization, slides were washed in 2XSSC three times for 5 min at 37°C, and in 0.1XSSC three times for 5 min at 45°C, rinsed in PBS twice and counter-stained with DAPI.
For immuno-FISH, the following procedure is added after washing with 0.1XSSC at 45°C. Wash twice with 2XSSC 5 min each at RT. Blocking with (TNT buffer; 0.1M Tris-HCl pH 7.5, 0.15M NaCl, 5% BSA) for 1 h at RT. Anti-lamin antibody is incubated for overnight at 4°C in TNT buffer, wash with wash buffer three times for 5 min. Second antibody is applied in TNT buffer for 2–3 h at RT, wash with wash buffer (0.1M Tris-HCl pH 7.5, 0.15M NaCl), including DAPI staining as described above. Cells were mounted on the glass slide with FluoromountG (Southern Biotech. Birmingham, AL). Three-dimensional image stacks were taken with Leica SP5 confocal microscope (Leica Microsystems, Exton, PA) using an x63 oil immersion objective with a numerical aperture of 1.4, and zoom 3.2±0.2.
To perform DNA FISH on target and non-target probes, approximately 15 kb region were chosen, except for the repeated sequence, in the genome and amplified by PCR from genomic DNA with 5–10 primers pairs, each covering around 0.5–3 kb. Primer sequences are available on request.
To determine quantitatively the three-dimensional position of the FISH signal within the nucleus, we used the ImageJ software [49]. The nuclear envelope was initially defined by segmentation of the DAPI image using the automated Otsu thresholding algorithm. The boundary definition was then refined against the lamin-staining, flagging significant deviations between the two signals if necessary. Figure S10 shows a schematic diagram of the procedure. We also display a distribution of radii calculated for 62 nuclei, demonstrating that the DAPI and lamin signals provide very consistent definitions of the nuclear boundary. Segmented images were then stacked in order to recreate the three-dimensional nucleus.
Next we calculated the distances between the FISH signal and the nuclear boundary (Figure S10, S11, Videos S1, S2, S3, S4). The segmented three-dimensional images of the nucleus were converted into a three-dimensional distance map using the Local Thickness plug-in (http://www.optinav.com/Local_Thickness.htm). We thresholded the FISH images to identify voxels within the nucleus that corresponded to the FISH signal and we measured the distances between all such voxels and the closest point on the nuclear boundary. For each nucleus we calculated the mean distance, and then for each test locus, we use the set of mean distances for all nuclei to plot the distance distribution. Similar results were obtained when we used the centre of mass of the FISH signal as the reference point instead of the mean distances for individual voxels (data not shown).
In total, we examined 1,712 nuclei (35–91 samples for each target locus; total 1,172 nuclei for NAR; total 540 nuclei for non-NARs). For a given target, we compiled all distance measurements from all relevant nuclei to produce a distribution of distances as shown in Figure 4 and Table S4.
The lamin L105 and N2 non-NAR targets were selected as in vivo controls with representative distributions for peripheral and non-peripheral sub-nuclear localisation. We compared the localisation of each target locus by comparing its distance measurements against the L105 and N2 controls separately. Statistical significance was calculated using the Wilcoxon test, with a FDR-corrected threshold of p <0.05. Briefly, a non-significant p-value (ie p-value >0.05) compared with the L105 distribution is indicative of peripheral localisation, whereas a non-significant p-value (i.e. p-value >0.05) compared with the N2 distribution is indicative of non-peripheral localisation.
Microarray data are available in the ArrayExpress databaset [42] under accession numbers E-MEXP-2523 (gene expression data) and E-MEXP-2525 (ChIP-chip data).
|
10.1371/journal.pcbi.1005599 | Systematic, network-based characterization of therapeutic target inhibitors | A large fraction of the proteins that are being identified as key tumor dependencies represent poor pharmacological targets or lack clinically-relevant small-molecule inhibitors. Availability of fully generalizable approaches for the systematic and efficient prioritization of tumor-context specific protein activity inhibitors would thus have significant translational value. Unfortunately, inhibitor effects on protein activity cannot be directly measured in systematic and proteome-wide fashion by conventional biochemical assays. We introduce OncoLead, a novel network based approach for the systematic prioritization of candidate inhibitors for arbitrary targets of therapeutic interest. In vitro and in vivo validation confirmed that OncoLead analysis can recapitulate known inhibitors as well as prioritize novel, context-specific inhibitors of difficult targets, such as MYC and STAT3. We used OncoLead to generate the first unbiased drug/regulator interaction map, representing compounds modulating the activity of cancer-relevant transcription factors, with potential in precision medicine.
| Most transcription factors are considered “undruggable” in conventional drug discovery. However, a large number of them are discovered to be key tumor dependencies. Thus, targeting these difficult targets has been a challenge for cancer drug discovery. Here, we introduce a novel method, OncoLead, that applies biological networks to identify candidate inhibitors that either directly or in-directly block the activities of these targets. This approach is confirmed by known target-inhibitor interactions in public databases. Furthermore, we predicted new inhibitors for MYC and STAT3, which are validated by in vitro assays.
| While the number of high-value, candidate therapeutic target proteins has increased dramatically over the past five years, most of them lack a corresponding FDA-approved or late-stage investigational (i.e., clinically relevant) small-molecule inhibitor. Furthermore, a large number of these are considered undruggable and may thus benefit from small molecules inducing potent, albeit indirect inhibition, within a specific tumor context. For instance, ibrutinib, a Bruton’s Tyrosine Kinase (BTK) inhibitor, can effectively abrogate aberrant NF-kB activity in human B cells, with clinically relevant application to treatment of the ABC subtype of diffuse large B cell lymphoma [1].
A key problem in addressing this challenge is the lack of generalizable methodologies for the efficient and systematic prioritization of small molecule compounds as direct or indirect inhibitors of an arbitrary protein of interest. Throughout this manuscript, we will use the word ‘compound’ for short to refer to small molecule compounds. Consistently, by compound targets and compound activity we refer to the proteins targeted by the small molecule compound and its pharmacological activity, respectively. Indeed, high-throughput screens (HTS) mostly rely on ad hoc, experimental gene reporter assays, whose design, testing, optimization, and miniaturization is laborious and inefficient. In addition, most of these assays are limited to reporting on the activity of a single target protein or of a specific protein class (e.g., protein kinases [2]). Computational HTS approaches, such as quantitative structure activity relation (QSAR) analysis [3] and virtual screening [4], rely on availability of structural models for both the ligands and the target protein and thus on prior knowledge from related compound’s binding assays or from X-ray/NMR target structure elucidation [3]. For instance, the similarity ensemble approach (SEA), which predicts new target-ligand relationships based on their similarity to established target-ligand sets, is widely adopted [5]. However, results completely depend on the availability of ligand analogs, whose structure has been previously elucidated.
Critically, these methods lack cell-context specificity and are limited to assessing only direct, high-affinity binding compounds, thus missing small-molecule compounds that may indirectly modulate the activity of a target protein, as is the case for ibrutinib. These compounds cannot be assessed by QSAR, because they do not represent high-affinity ligands of the target protein of interest but rather of one of its major context-specific up-stream regulators. In addition, these methods are not effective for protein families that lack specific binding pockets, such as transcription factors (TFs) [6], even though these comprise many of the best established tumor dependencies. Indeed, TFs such as ESR1, NOTCH1, MYC, GATA3, and ERG, among many others, are frequently aberrantly activated in cancer [7]. In addition, many TFs have been recently elucidated as Master Regulators of tumor cell state, which are organized in highly interconnected modules or tumor checkpoints [8], including key synthetic lethal combinations, such as STAT3, CEBPB, and CEBPD in mesenchymal glioblastoma [9] or CENPF and FOXM1 in malignant prostate carcinoma [10].
Recently, several perturbational strategies have been proposed to measure differential gene expression following systematic chemical perturbations of specific cell lines, such as the connectivity map (CMAP) [11] and the Library of Integrated Network-based Cellular Signatures (LINCS) [12]. However, since most small molecule compounds affect the activity rather than the expression of target proteins, these data cannot elucidate targets but rather their ability to modulate the entire gene expression signature of a cell. We recently introduced DeMAND, a method for the interrogation of cell context specific networks, to infer drug mechanism of action (MoA) [13]. While being very efficient to capture direct as well as indirect context-specific targets [13], DeMAND requires at least six gene expression profiles per compound. As a result, while it is very effective for elucidating the MoA of individual compounds of interest, it is not optimally suited to the reverse problem, i.e., prioritizing candidate protein inhibitors from large-scale perturbational profiles, especially when fewer than six perturbational profiles per compound are available.
We thus developed OncoLead, a novel and highly generalizable methodology for the efficient and systematic identification of small molecules that directly or indirectly inhibit a target protein of interest. OncoLead leverages the Virtual Inference of Protein activity by Enriched Regulon analysis (VIPER) algorithm [10, 14]—a network-based algorithm for the assessment of protein activity from gene expression data—to assess the effect of a panel of drugs on protein activity from individual expression profiles. We limit our analysis to ~7,000 regulatory proteins (RPs), including ~2,000 transcription factors (TFs) and ~5,000 signaling proteins (SIGs), whose regulatory ‘activity’ may be modulated by a small-molecule compound. While these represent only ~30% of the human genome, they capture an important component of relevant tumor dependencies that may benefit from targeted inhibitor availability.
Briefly, given two cellular states (e.g., baseline and compound-perturbed), OncoLead uses the differential expression of a protein’s transcriptional targets (i.e., its regulon) as an accurate and highly reproducible multiplexed endogenous reporter assay for its activity [15, 16]. For a given RP, the regulon comprises its context-specific direct or indirect transcriptional targets [17]. This approach is especially well suited to the screening of large libraries of compounds for two reasons: first, it can accurately infer compound-mediated protein activity modulation from a single perturbational profile (e.g., RNASeq following perturbation); second, its performance is essentially unaffected when RNASeq depth is reduced from 30M to 0.5M reads [18], thus allowing highly-multiplexed characterization of the activity of compounds at low cost.
We first show that OncoLead can effectively assess differential activity for established targets of the compound, even when these are not differentially expressed following compound’s perturbation. To accomplish this goal, we leveraged two public databases including the Connectivity MAP (CMAP) [11] and the Library of Integrated Network-based Cellular Signatures (LINCS) (http://lincs.hms.harvard.edu/), as well as one in vivo dataset, containing gene expression profiles (GEPs) obtained post-treatment from patients’ tumor tissue. For each cell line or tissue represented in the datasets, the analysis was performed using networks representing the transcriptional targets of the candidate compound-targeted proteins in tissue lineage-matched contexts. We used the algorithm to assemble the first comprehensive, cell-context-specific map of inhibitors targeting RPs. The associated resource, which includes a comprehensive map of RP-compound’s interactions, is available as a supplementary file linked to this publication. We then show that the algorithm is effective in elucidating novel tumor-specific inhibitors of undruggable targets. Specifically, OncoLead was highly effective in inferring novel breast-cancer-specific inhibitors of MYC and STAT3, which were experimentally validated.
OncoLead assesses whether a compound is an effective inhibitor/activator of a given regulatory protein, based on its effect on the transcriptional level of the protein’s regulon—i.e. its set of direct and indirect transcriptional targets—to infer the regulatory protein’s differential activity; see Methods and [18]. For simplicity, we call compound’s mode-of-action (CMoA) to the full repertoire of proteins, whose activity is significantly affected following perturbation with the compound. These include both direct targets as well as context-specific downstream effectors of compound’s activity, and thus effectively representing the context-specific compound’s MoA.
Clearly, the accuracy of our inferences of protein activity depends on the quality of the protein regulons. Due to lineage specific chromatin remodeling and co-factor availability, protein regulons are highly cell context specific [19, 20]. In this work, we used the ARACNE algorithm [21] for context-specific inference of the regulatory network. We have previously shown that regulons inferred by ARACNE are particularly suited for VIPER analysis [18]. As shown in Table 1, ARACNE-based regulon inference was performed using tumor-context matched gene expression profiles (GEP) from The Cancer Genome Atlas consortium (TCGA) [22], and relevant tumor context matched GEO datasets [23], when available.
We further complemented the ARACNE networks by incorporating evidences from other resources, including direct TF to target genes interaction evidences from chip-seq or chip-chip data (ChEA database) [25], direct or indirect protein-protein interactions from the STRING database [24], and indirect functional associations inferred upon RNAi-mediated gene silencing experiments collected from the GEO database [26] (see Methods). Integration of these different evidences was performed at the inferred protein activity level (see Methods and Fig 1A).
To quantitatively assess interactome quality, we computed the Interactome Reliability Scores (IRS) as the area under the curve (AUC) representing the number of statistically significant OncoLead-inferred CMoA proteins as a function of the p-value threshold (see Methods). The rationale, as previously discussed [10], is that less accurate interactome models yield fewer statistically significant proteins, and thus lower IRS than the more accurate ones. Indeed, IRS scores decreased monotonically when protein interactions were increasingly randomized (0%–100%) using a degree-preserving randomization algorithm [27] (Fig 1B). Furthermore, confirming our hypothesis, tissue-matched interactomes systematically achieved the best IRS performance against the corresponding cell line specific signatures (Fig 1C).
Gene expression signatures (GES) representing each cell line following compound’s perturbations were then analyzed using OncoLead on multiple networks to generate integrated results. This produced a sparse 3-dimensional matrix of protein activity signatures [ΔAP,L,C], representing the relative differential activity (treatment with compound vs. DMSO control) of each target protein, P, expressed as Normalized Enrichment Score (NES), in cell line L, with compound C. This matrix thus provides a quantitative representation of the CMoA of all tested compounds, across all profiled cell lines (S1 and S2 Tables).
We then expanded this analysis by leveraging an extensive collection of gene expression profiles, representing treatment of multiple cell lines with various compounds and shRNAs targeting different genes, available from the LINCS repository (http://lincs.hms.harvard.edu/). These datasets provide a limited representation, restricted to only 978 reporter genes (L1000) measured by a multiplexed Luminex assay. Within the constraints of such reduced representation, we used this dataset to build an experimental gold standard dataset (GSD) of compounds affecting the activity of specific target genes, by matching the signature of compound’s perturbations to those of shRNA mediated silencing. We limited our analysis to 1,365 compounds yielding statistically reproducible transcriptional responses (see Methods), and 92 shRNA-mediated silencing assays for which (a) target gene silencing could be confirmed by L1000 measurements at >3 standard deviations from the controls mean and (b) the gene was represented in the interactome as regulator (i.e. RP). This resulted in distinct gene silencing assays for each cell line, from a minimum of 16 in VCAP prostate cancer cells to a maximum of 151 in A375 melanoma cells.
To assemble a suitable experimental GSD, compound’s perturbations were matched to gene-silencing assays by Pearson correlation analysis of the corresponding, cell-matched L1000 signatures. Thus, each of the 1,365 most reproducible compound’s perturbations were associated to a list of shRNA-mediated gene silencing assays, ranked from the one with the most correlated to the one with the most anti-correlated L1000 signature. The rationale is that the gene-silencing assays with signatures most correlated to a compound’s perturbation signature represent proteins whose activity is inhibited by the compound.
We then computed the DTPA of each target protein for each of the perturbed cell lines, using the full gene expression profile, from regression analysis of the L1000 signature, see Methods. Thus, for each target protein, we rank-sorted all compounds by DTPA score, from its strongest predicted candidate inhibitor (i.e., that with the largest negative DTPA) to its strongest activator (i.e., that with the largest positive DTPA). Finally, we assessed these predictions by reciprocal gene set enrichment analysis (GSEA)[30] of the OncoLead-predictions against the experimentally-prioritized target modulators in the GSD. Specifically, for each target protein, we computed the NES representing enrichment of DTPA ranked inhibitors in statistically significant GSD inhibitors (p = 0.05).
Enrichment was statistically significant for most proteins targeted by small molecule compounds (NES > 1.96; p < 0.05, shown in green for RPs, Fig 2C and S2 Fig). This includes 112/151 proteins in A375 cells (74%), 68/106 in A549 cells (64%), 33/45 (73%) in HA1E cells, 19/25 (72%) in HCC515 cells, 75/145 (52%) in HEPG2 cells, 105/137 (52%) in HT29 cells, 97/120 (81%) in MCF7 cells, 85/97 (88%) in PC3 cells, and 16/16 (100%) in VCAP cells. Overall 609/842 testable proteins (72%) yielded OncoLead-inferred candidate inhibitors that were strongly enriched in experimentally assessed ones, based on the GSD. This is especially remarkable considering that LINCS L1000 assays directly measure expression of only 978 genes. As a result, on average, only 1/20th of regulon targets is directly measured by these assays while other targets are imputed. In addition, shRNA-mediated silencing may have significant off-target effects. Taken together, these data suggest that the method represents an effective strategy to prioritize candidate inhibitors for arbitrary proteins of interest.
To test whether OncoLead may be effective in elucidating the targets of specific compounds in vivo, we used gene expression data obtained from patient-derived tumor biopsies before and after therapeutic intervention. Specifically, we leveraged a dataset generated by Miller et.al (GSE20181)[31], consisting of primary breast tumor samples profiled after Letrozole treatment, including at 30-days (short term: ST) and 90-days (long-term LT), compared to pre-treatment profiles (PT). Letrozole blocks estrogen synthesis in postmenopausal patients by inhibiting the aromatase enzyme. This abrogates estrogen receptor activation in breast cancer cells. Individuals profiled in this dataset include 36 estrogen deprivation responsive and 14 non-responsive patients. Response was assessed based on whether significant tumor size reduction was observed at 90-days post-treatment.
Four differential expression signatures were analyzed, including ST:PT (30-days VS. pre-treatment) and LT:PT (90-days VS. pre-treatment), across both responsive and non-responsive patients. DTPA vectors were obtained by OncoLead analysis of these signatures, using the TCGA patient-derived Breast Carcinoma interactome (Table 1). As expected, ESR1 activity was significantly reduced following Letrozole treatment in the responsive group (Fig 2D), with longer treatment inducing stronger ESR1 activity reduction (pLT:PT = 0.01; pST:PT = 0.046). Strikingly, however, OncoLead-inferred ESR1 activity was not significantly affected by Letrozole in non-responsive patients (pLT:PT = 0.51; pST:PT = 0.095). Furthermore, differential ESR1 expression was not statistically significant following Letrozole treatment (p-value > 0.05, by Student’s t-test) at either time point and for either responsive or non-responsive patients, suggesting that CMoA analysis correctly captured ESR1 inhibition even though its expression levels were not affected (Fig 2D).
Since the IRS of each perturbation summarizes the effect of such perturbation on the inferred activity of regulatory proteins, we decided to use the IRS as a metric for the bioactivity of the small molecule compound. Specifically, we evaluated the IRS score across all cell lines and compound’s perturbations in the Connectivity Map (CMAP) dataset. As expected, progressive degradation of the gene expression signatures, by randomly permuting increasingly larger subsets of gene expression values, was associated with a proportional decrease in the IRS (Fig 1D). We selected Irinotecan for this test because it showed one of the highest IRS values in CMAP.
Among the 1,294 CMAP compounds, HDAC, topoisomerase, CDK, and estrogen receptor antagonists presented the largest overall IRS in the MCF7 luminal breast cancer cell line. These compounds represent well-known cancer drug classes, currently under investigation in breast cancer clinical trials, and thus likely to be highly bioactive in these cells. The same analysis performed on the other two CMAP cell lines (i.e., PC3 and HL60), consistently identified HDAC, HSP90, NF-KB, topoisomerase, proteasome and protein synthesis inhibitors among the compounds with highest IRS. Interestingly, we observed a low IRS for a large proportion (~55%) of the compounds profiled in MCF7 cells in CMAP, suggesting poor bioactivity of those compounds at the profiled concentrations.
The specific highest IRS in MCF7 (breast cancer, BRCA), HL60 (acute promyelocytic leukemia, APL), and PC3 (prostate cancer, PRAD) cells was achieved by fulvestrant (ESR1 antagonist), tretinoin (all-trans retinoic acid), and pioglitazone, respectively. Estrogen antagonists represent the standard of care in BRCA adjuvant therapy[32]. Indeed, based on NCI60 data[33], fulvestrant achieves 50% growth inhibition (GI50) at a substantially lower concentration in MCF7 compared to HL60 (-LogGI50 (M) = 8 vs. 5.2). Similarly, all-trans retinoic acid represents the standard of care in APL[34]. Indeed, based on drug sensitivity profile in COSMIC data, HL60 cells were more sensitive to tretinoin than MCF7 and PC3 (LogIC50 (uM) = 0.83, 4.0 and 5.8 for HL60, MCF7, and PC3, respectively). Finally, pioglitazone, a PPAR-γ agonist approved by the FDA as anti-diabetes drug, showed higher IRS in PC3 compared to the other two cell lines. Interestingly, PPAR-γ agonists are currently in phase 2 clinical trials for AR-independent prostate cancer[35].
Our analysis provides a comprehensive map of transcription regulator (TR), whose activity is modulated by a large repertoire of compounds across the CMAP and LINCS perturbational databases. This represents a large set of cellular contexts, including: MCF7 (breast), PC3 (prostate), HL60 (leukemia) in CMAP, and A375 (skin), A549 (lung), HA1E (kidney), HCC515 (lung), HEPG2 (liver), HT29 (colon), MCF7 (breast), PC3 (prostate), and VCAP (prostate) in LINCS (S1 and S2 Tables). Specifically, a compound-TR activity matrix was generated for each cell line, with rows representing TR proteins and columns representing compound’s perturbations. This matrix can be used to perform a variety of analyses, including identifying optimal context-specific compounds to inhibit or activate an arbitrary TR protein, as well as to compute the similarity between compounds.
Traditional drug discovery has been focused on one-drug / one-target strategy. More recently, the approach has been expanded to poly-pharmacological effects of the drugs, meaning that one drug can interact with multiple on-target and off-targets resulting in combination of wanted and unwanted effects [51]. However, MoA discovery is still limited to compound’s direct binders [4, 5]. In this work, we expand the concept of drug MoA by including direct and indirect mediators of compound’s effect. We limit our approach however, to proteins having a transcriptional regulatory role in the cell (TR). Given their central role as regulators of cell state [10, 17], limiting CMoA definition to TRs dramatically reduce the dimensionality of the problem with no reduction of sensitivity to detect compound’s biological activity. Due to signaling and transcriptional network rewiring, inclusion of non-direct targets in CMoA makes the approach exquisitely cell context specific.
While small molecule compounds affect mostly the activity of the directly or indirectly targeted proteins, we currently have no methods in our biochemistry arsenal to directly measure protein activity. While important advances in the field of proteomics have made possible the quantification of proteins and protein isoforms in a close-to proteome-wide fashion, protein activity is not solely determined by protein or protein isoform abundance, but also depends on proper cellular localization and interaction with co-factors. Gene expression, on the other hand, can be actually profiled with high accuracy and at relative low cost, making it one of the phenotypic read-outs of choice for current drug-screen efforts. mRNA abundance however, is not directly associated with coded protein activity, especially after compound-mediated short-term perturbations, but can be interpreted, using context-specific models of transcriptional regulation, to infer changes in the regulatory proteins activity [10].
At the core of our approach is the VIPER algorithm [10, 18], which makes possible several fundamental characteristics: (1) the analysis can be performed at the single-sample level, allowing its application to FDA-approved drug repositioning in cancer precision medicine protocols (see below), (2) results are very robust to regulatory model accuracy and expression profile quality, and (3) results are almost insensitive to partial transcriptome coverage, making it particularly suited for the analysis of low-depth RNAseq expression profiles following drug perturbation [18]. In fact, we have shown that VIPER-activity signatures obtained from 1 million reads per sample are virtually identical to the ones obtained from 30 million reads, while GES were dramatically different. This remarkable quality enables us to infer drug MoA from ultra-low cost high-multiplex expression profile analysis following drug perturbation. VIPER can accurately identify regulators whose activity is modulated by the compound. However, in ~10% of the cases, it may switch the effect directionality (e.g., infer a protein as activated when is in fact inhibited) [20]. This occurs because, at steady state, auto regulatory loops may exist that induce inverse correlation between protein activity and mRNA expression. As a result, targets may be correctly inferred but their relationship (activated vs. repressed targets) may be inverted.
We recently introduced DeMAND, a method for the elucidation of compound MoAs, based on compound’s perturbational gene expression profiles[13]. Although OncoLead and DeMAND are both network-based, they are fundamentally different and complementary, both in their formulation and, more importantly, in their practical applications. More specifically, while DeMAND focuses on compound-mediated dysregulation of protein interactions, thus requiring at least six distinct perturbations for reliable predictions, OncoLead directly infers changes in protein activity based on their regulon differential expression, thus requiring a single sample. Using the same benchmarks that were used to evaluate performance of the DeMAND algorithm, the perturbational profiles of fourteen compounds in LY3, we tested the algorithm complementarity of the OncoLead and the DeMAND. OncoLead outperformed DeMAND almost by the same margin by which DeMAND outperformed a naïve t-test analysis, as reported in[13] (S4A Fig). When integrating the predictions from the two methods, each protein was assigned the best of the OncoLead or DeMAND scores. The integrated results outperformed both individual method predictions, confirming the significant complementarity of these methods (S4A Fig). For example, while DeMAND missed the direct doxorubicin target (TOP2A), this was effectively captured by OncoLead (ranked 25 out of 6,819 targets, p = 4.7x10-22). Conversely, while OncoLead missed geldanamycin’s target HSP90AA1, it was effectively discovered by DeMAND (ranked 6 out of 7499 targets, p = 1x10-25). In general, OncoLead is better suited to analyzing proteins that have a direct regulatory role, while DeMAND is more effective on proteins involved in non-regulatory interactions, such as those involved in heat stress responses. As a result, the integration of both methods is optimally suited for the elucidation of compound’s MoA.
A clear limitation of our approach however, is that the definition of drug MoA is limited to the proteins represented as regulators in the network models. This leaves out non-protein mediators of drug effect as well as proteins not considered as regulators (TRs in this manuscript). However, because any drug-induced phenotypic change reflected in transcriptome changes will be partially mediated by transcriptional regulators, genome-wide coverage of TRs by the regulatory models should be enough to capture, at least partially, the MoA of any bioactive compound.
This study presents a new framework for defining compound’s MoA. We have shown that OncoLead-MoA captures the direct targets for known drugs while GES does not. Moreover, by interpreting drug-induced GES with models of transcriptional regulation, OncoLead-MoA is insensitive to signature noise while providing information about the bioactivity of the compounds, even when no replicated profiles are available. This is because non-informative signatures, generated from non-bioactive compounds exposure, will be poorly explained by the regulatory models and hence, their IRS scores will tend to be small, compared to informative, bioactive compound-derived expression signatures (S4B Fig).
We envision a strong impact of our approach for drug repositioning in precision medicine. Specifically, our method can be applied to infer drugs targeting patient-specific and currently ‘undruggable’ targets. For instance, we have shown that tumor subtype-specific MRs constitute tumor dependencies[8–10], which are usually non-tractable from the current repertoire of FDA-approved drugs. However, inhibitors for such MRs can be inferred from gene expression data of appropriate models after drug perturbation, as we have shown here for MYC and STAT3. We can envision a framework for cancer precision medicine in which we first infer the MRs for a single-patient in an unbias genome-wide fashion. Then by matching the patient’s MR profile with the full FDA-approved drug OncoLead-MoA, in the spirit of the Connectivity Map [11], we select drugs or drugs combinations that not only target the top patient MR, but that comprehensively target a significant proportion of them. The elevated noise in single-tumor gene expression profiles makes this approach to be unfeasible if based only on gene expression. Conversely, interpretation of the signatures based on regulatory models, which are based on tenth to hundreds of genes per protein regulon, makes OncoLead results highly reproducible [20], and the single-patient MR signatures, as well as the single-sample based drug MoA obtained from drug-screen experiments, extremely robust (S4C Fig). If MRs are enriched in tumor dependencies [9, 10], then this approach should prioritize drugs being specifically toxic for the patient’s tumor.
GSEA uses the Kolmogorov-Smirnov statistic and tests the enrichment of a gene set on a gene signature generated by t-test, fold-change or other methods[30, 53].
The distance between drugs based on gene expression or CMoA profiles was computed using signature distance algorithm. It takes the 5% most up-regulated and down-regulated genes (gene expression value or OncoLead activities) of one sample 1 (2) and computes its enrichment on the other sample 2 (1) to get value DIS1-2 (DIS2-1). Then take the average of DIS1-2 and DIS2-1 to get the distance between sample 1 and 2.
The distance between drugs based on GI50 sensitivity profiles from NCI60 was computed using spearman correlation of the sensitivity scores across different cell lines.
MCF7 and SNB19 cells were cultured in RPMI (Invitrogen) and supplemented with 10% FBS (Invitrogen) and 1% penicillin-streptomycin (Cellgro). Cell viability was measured using CellTiter-Fluor Cell Viability Assay (Promega, G6080). Cignal STAT3 reporter (luc) kit was purchased from Quiagen (CCS-9028L). MYC reporter was bought from Qiagen/SABiosciences (CCS-012L). Reporter activity was determined by Dual-Glo Luciferase Assay System (Promega, E2920). Compounds were purchased from Sigma, Prestwick and Spectrum.
MCF7 cells in RPMI 10% FBS with antibiotics were plated at the density of 2x104 cell/well onto 96-well flat-bottom plate one day before transfection. Cells were transfected with MYC reporter using the jetPrime (Polyplus) delivery system. Reporter mixture contains an inducible MYC responsive firefly luciferase construct and constitutively express Renilla construct (ratio 40:1). A mixture of non-inducible firefly luciferase reporter constitutively expressing Renilla construct was used (ratio 40:1) as a negative control. Constructs constitutively expressing GFP, firefly luciferase and Renilla luciferase constructs (40:1:1) were used as positive control.
24 hours after cell seeding, culture medium was replaced with fresh medium (100 ul) and drugs were added to the cells in duplicates. 24 hours after drug treatment, cell viability and MYC reporter activity was measured. Serial dilutions of drugs were prepared in DMSO to keep the same final concentration of DMSO at 0.8% (S4 Table). DMSO only was used as a negative control for drugs. Gene reporter activity following drug treatment was normalized to cell viability (Firefly/CellTiter-Fluor) and compared to the negative controls (DMSO).
MCF7 cells were first transfected with STAT3 and stimulated using 40ng/ml IL6. 24 hrs later cells were treated with drugs at 10uM. For the control, corresponding amount of DMSO was added to the cells. 24hrs after the drug treatment, cell viability was determined and reporter activity was measured. Each experiment was performed in triplicate. The experimental procedure for SNB19 was the same for MCF7 except stimulation using IL6. Galliela Lactone and Static were included in the experiment as positive controls and DMSO was used as negative control.
|
10.1371/journal.pgen.1005042 | Maternal Co-ordinate Gene Regulation and Axis Polarity in the Scuttle Fly Megaselia abdita | Axis specification and segment determination in dipteran insects are an excellent model system for comparative analyses of gene network evolution. Antero-posterior polarity of the embryo is established through systems of maternal morphogen gradients. In Drosophila melanogaster, the anterior system acts through opposing gradients of Bicoid (Bcd) and Caudal (Cad), while the posterior system involves Nanos (Nos) and Hunchback (Hb) protein. These systems act redundantly. Both Bcd and Hb need to be eliminated to cause a complete loss of polarity resulting in mirror-duplicated abdomens, so-called bicaudal phenotypes. In contrast, knock-down of bcd alone is sufficient to induce double abdomens in non-drosophilid cyclorrhaphan dipterans such as the hoverfly Episyrphus balteatus or the scuttle fly Megaselia abdita. We investigate conserved and divergent aspects of axis specification in the cyclorrhaphan lineage through a detailed study of the establishment and regulatory effect of maternal gradients in M. abdita. Our results show that the function of the anterior maternal system is highly conserved in this species, despite the loss of maternal cad expression. In contrast, hb does not activate gap genes in this species. The absence of this activatory role provides a precise genetic explanation for the loss of polarity upon bcd knock-down in M. abdita, and suggests a general scenario in which the posterior maternal system is increasingly replaced by the anterior one during the evolution of the cyclorrhaphan dipteran lineage.
| The basic head-to-tail polarity of an animal is established very early in development. In dipteran insects (flies, midges, and mosquitoes), polarity is established with the help of so-called morphogen gradients. Morphogens are regulatory proteins that are distributed as a concentration gradient, often involving diffusion from a localised source. This graded distribution then leads to the concentration-dependent activation of different target genes along the embryo’s axis. We examine this process, which differs to a surprising extent between dipteran species, in the scuttle fly Megaselia abdita, and compare our results to the model organism Drosophila melanogaster. In this way, we not only gain insights into how the mechanisms that establish polarity function differently in different species, but also how the system has evolved since these two flies shared a common ancestor. Specifically, we pin down the main difference between Drosophila and Megaselia in the altered function of the maternal Hunchback morphogen gradient, which activates target genes in the former, but not the latter species, where it has been completely replaced by the Bicoid morphogen during evolution.
| Axis formation and segment determination in the vinegar fly Drosophila melanogaster are among the most thoroughly studied developmental processes today [1–5]. They offer an ideal starting point for the comparative study of development and the evolution of pattern-forming gene regulatory networks. Axis formation in flies is based on the graded distribution of morphogens established through a number of different maternal regulatory systems. In this study, we will be focusing on two of those in particular: the anterior and posterior systems [4].
In D. melanogaster, maternal protein gradients are either formed by localisation of mRNA at the anterior or posterior pole of the embryo, or by regionally specific translational repression of ubiquitous maternal transcripts [4,5]. The anterior system centres around the anterior determinant Bicoid (Bcd). bcd mRNA is localised to the anterior pole of the embryo and an antero-posterior (A–P) protein gradient forms through diffusion from that source [6–8]. Bcd regulates the translation of uniformly distributed maternal mRNA of caudal (cad) [9,10], which leads to a graded distribution of Cad protein with high concentration levels in the posterior [6,9,11–14]. In addition, Bcd acts as a concentration-dependent transcriptional regulator of zygotically expressed segmentation genes—such as gap or pair-rule genes [6,15–20]. In the case of the posterior maternal system, nanos (nos) mRNA is localised in the posterior pole region forming the source of the Nos protein gradient [21–23]. Unlike Bcd, Nos is not a transcriptional regulator: its only role is to translationally regulate ubiquitous maternal hunchback (hb) mRNA, leading to an anterior gradient of maternal Hb protein [24–26].
Evidence for the presence of localised determinants in dipterans goes back to early studies that utilised UV irradiation or RNAse treatment on embryos of chironomid midges (Fig. 1, Nematocera: Culicomorpha). These experiments produced mirror-duplicated abdomens, so-called bicaudal phenotypes, in which anterior structures are missing and replaced by duplicated organs usually found in the posterior [27–29]. The observed effects were attributed to the destruction of an anteriorly localised mRNA. However, the identity of the anterior determinant is still unknown in the majority of dipteran infraorders. The bcd gene arose through a duplication of the hox3 factor zerknüllt (zen) at the base of the cyclorrhapha (Fig. 1) [30–33]. While its spatial distribution and role as transcriptional regulator are highly conserved among cyclorrhaphans [30–32,34–39], it is not present in other flies.
Interestingly, anterior UV irradiation of D. melanogaster embryos—or mutations to the bcd gene—never produce bicaudal phenotypes [40,41]. This hints at the presence of an additional non-localised factor. This factor is hb, which contributes to axis specification and A–P polarity in D. melanogaster. The ubiquitous distribution of its maternal mRNA may explain why it is resistant to localised UV irradiation. This interpretation is consistent with the fact that only embryos lacking both bcd and hb show bicaudal phenotypes in this species [24,42,43].
While the roles of bcd and hb in axis specification appear to be somewhat redundant in D. melanogaster, the situation is different in other cyclorrhaphan flies. The hoverfly Episyrphus balteatus, for example, has secondarily lost maternal hb expression (Fig. 1) [38,44]. Consequently, knock-down of bcd by RNA interference (RNAi) leads to bicaudal phenotypes in this species [38].
In this paper, we study axis specification and maternal regulation of segmentation genes in another non-drosophilid cyclorrhaphan species, the scuttle fly Megaselia abdita (Fig. 1). M. abdita belongs to the most basally branching cyclorrhaphan lineage, the Phoridae [45,46]. While maternal cad expression has been lost in this species (Fig. 1) [47], hb retains its maternal contribution [31]. In light of this, it is surprising that knock-down of bcd does lead to bicaudal phenotypes. We investigate the regulatory causes of this phenomenon through a detailed study of the establishment and regulatory role of maternal gradients in M. abdtia. Our results reveal that the anterior and posterior systems are much less redundant compared to D. melanogaster. In particular, the difference between the two species can be explained by the loss of gap gene activation through maternal hb in M. abdita. Our results indicate that the role of the posterior system in axis specification has been lost in E. balteatus and M. abdita, while it still retains some of its ancestral functionality in D. melanogaster. In this general scenario, the anterior system is gradually replacing the posterior one during the evolution of the cyclorrhaphan flies.
The posterior maternal system is based on maternal gradients of Nos and Hb protein. In M. abdita, nos mRNA is localised posteriorly during early cleavage stages (Fig. 2A) becoming restricted to the pole cells by C10 (Fig. 2B) as in D. melanogaster. Previous reports have documented ubiquitous maternal hb mRNA [31] as well as conserved zygotic hb expression in an anterior and a posterior domain [48]. Antibody stainings reveal a distribution of Hb protein very similar to the zygotic mRNA pattern during the late blastoderm (Fig. 2C,D). Furthermore, an anterior Hb protein gradient is present at cleavage and early blastoderm stages (Fig. 2E). In order to investigate the role of the posterior system in the formation of this gradient, we treated M. abdita embryos with nos RNAi. These embryos show no effect on hb mRNA, while ectopic Hb protein is present in the posterior of the embryo (effect detectable in 15 out of 16 RNAi-treated embryos; Fig. 2F). We conclude that the maternal Hb gradient is set up through translational repression by Nos in M. abdita as in D. melanogaster.
The anterior maternal system of M. abdita is less conserved than the posterior one. Unlike D. melanogaster [9,10] and E. balteatus [44], M. abdita lacks maternal cad transcripts [47] and consequently maternal Cad protein. Zygotic expression of cad, on the other hand, is qualitatively similar in D. melanogaster, E. balteatus, and M. abdita [9–12,14,37,42,47–49]. The only notable difference is that abdominal cad expression reaches further anterior in the latter two species compared to Drosophila [44,48].
In order to test how zygotic cad expression is regulated in M. abdita, we knocked down bcd, hb, and the head gap gene orthodenticle (otd). In bcd RNAi-treated embryos, we observe a derepression of cad transcripts in the anterior (38/48; Fig. 3A–F). At cleavage cycle 13 (C13), cad expression appears uniform throughout the embryo (Fig. 1D). During early C14A (time class 2, T2), cad becomes expressed at higher levels in the anterior than in the posterior (Fig. 3E). This effect is specifically confined to the region that is free of cad expression in wild-type embryos (compare to Fig. 3B). At later stages, an ectopic domain resembling the posterior cad stripe forms in the anterior (Fig. 3F). Similar ectopic cad stripes have been observed in the anterior of D. melanogaster bcd mutants [14], cad reporter assays in D. melanogaster [47], and E. balteatus embryos treated with bcd RNAi [38].
In hb knock-down embryos, we observe a small anterior expansion of cad expression in a minority of specimens (4/13; Fig. 3G–I; S1 File). Anterior derepression is much more subtle in this case than in bcd knock-downs (Fig. 3D–F). This effect is similar to hb mutants of D. melanogaster [42].
Given the difference between bcd and hb knock-downs, we investigated potential additional contributions by otd, a factor known to act as a transcriptional repressor of cad in the jewel wasp Nasonia vitripennis [50]. otd expression is lost in bcd RNAi-treated embryos (12/16; S1 Fig). However, expression of cad appears normal in embryos treated with otd RNAi (25/25; Fig. 3J–L; S1 File). This indicates that otd is not involved in cad regulation, consistent with results from D. melanogaster [50] and E. balteatus [44].
In summary, anterior repression of cad in D. melanogaster is due mainly to a combination of translational repression by Bcd—acting on ubiquitous maternal cad mRNA—and transcriptional repression by hb—acting on the zygotic abdominal cad domain [14,42]. Transcriptional regulation of cad by Bcd plays a minor role, if any [47]. In contrast, repression of cad by Bcd occurs predominantly at the transcriptional rather than the translational level in M. abdita, similar to E. balteatus [38]. Our evidence does not conclusively establish whether this interaction is direct. However, we have shown that potential intermediate factors such as Otd and Hb are not involved in cad regulation, or show regulatory effects that are far too subtle to account for anterior repression in M. abdita.
Previous work has shown that bcd mRNA is localised anteriorly in M. abdita [30–32,48], and that it regulates hb transcription through the P2 promoter [31,37]. To assess the effect of bcd on gap gene regulation and embryo polarity in general, we characterised expression patterns of the trunk gap genes hb, giant (gt), knirps (kni), Krüppel (Kr), and the pair-rule gene even-skipped (eve) in M. abdita embryos treated with bcd RNAi. We used single- and double-stained embryos to assess severity of the knock-down and spatial registration of expression patterns—between gap domains (Fig. 4) as well as between Kr and the pair-rule gene eve (Fig. 5). We take advantage of the variable knock-down efficiency in RNAi experiments, which acts similar to an allelic series in classical genetics, to measure the sensitivity of specific gap domain boundaries towards decreasing levels of Bcd. In general, we find that all of these boundaries are highly sensitive to changes in Bcd concentration (Figs. 4 and 5; see also S1 File).
Wild-type embryos of M. abdita show a broad, bcd-dependent, anterior domain of zygotic hb expression, which gradually retracts from the pole (Fig. 4B) [31,37]. The posterior boundary of this domain shifts in anterior direction over time [48], unlike its equivalent in D. melanogaster. In embryos treated with bcd RNAi, we observe an anterior cap of hb expression which never retracts from the pole (35/42; Fig. 4C–F; S1 File). It reduces in size with the severity of the bcd knock-down (Fig. 4C–E) indicating dependence on Bcd concentration. Similar anterior domains have been observed in embryos derived from bcd mutant mothers in D. melanogaster [51] and in bcd RNAi-treated embryos of E. balteatus [38]. In both of these cases, the anterior cap of hb expression has been interpreted as an anterior mirror duplication of the posterior hb domain [38,51]. The posterior hb domain is also conserved in M. abdita (Fig. 4B) [48]. It exhibits a slight anterior expansion in some embryos treated with bcd RNAi (Fig. 4C–F; S1 File). In contrast, the posterior hb domain remains unaffected in D. melanogaster embryos lacking bcd [51].
Wild-type embryos of M. abdita have a broad anterior domain of gt, with a stationary posterior boundary, plus a posterior domain that shifts anteriorly over time (Fig. 4H) [48]. In embryos treated with bcd RNAi, we observe either loss (10/18) or strong reduction (8/18) of the anterior gt domain at early stages (before T3), while most embryos exhibit expression in a small anterior cap at later time points (14/15; Fig. 4I–L; S1 File). This anterior cap retracts from the pole around T8 (1/1). As for hb, the extent of anterior gt expression decreases with increasing strength of the knock-down effect (Fig. 4I–K). We interpret these observations as follows: delay and reduction of anterior gt expression are due to a lack of activation by Bcd, while the late anterior cap domain may be induced by ectopically expressed Cad (see Fig. 3E, F). The effect of bcd knock-down on the posterior gt domain is more modest. This domain is always present in bcd RNAi embryos but exhibits some anterior displacement of both its boundaries (Fig. 4I–L; S1 File). D. melanogaster embryos from bcd mutant mothers show a similar anterior displacement of the posterior gt domain, but no expression of gt in the anterior [52,53]. In contrast, E. balteatus embryos treated with bcd RNAi show broad derepression of gt, whose expression is only excluded from the anterior and posterior tip of the embryo [38].
In wild-type embryos of M. abdita, kni is expressed in an L-shaped anterior head domain, plus an abdominal domain that shifts to the anterior over time (Fig. 4N) [48]. In embryos treated with bcd RNAi, the head domain disappears, while the abdominal domain of kni expands and becomes displaced towards the anterior (38/38; Fig. 4O–R; S1 File). As in the case of hb and gt, the amount of expansion depends on the severity of the knock-down. This is qualitatively similar to embryos derived from bcd mutant mothers in D. melanogaster, but the effect is more severe in M. abdita and resembles kni expression in bcd mutants which are also heterozygous for maternal hb [24]. The effect of Bcd on kni is even more pronounced in E. balteatus where kni becomes drastically derepressed—showing ubiquitous expression in extreme cases—in embryos treated with bcd RNAi [38].
Wild-type M. abdita embryos have a central Kr domain, which is wider than its equivalent in D. melanogaster (Fig. 4A, G, M) [48]. As is the case for other gap domains, it shifts anteriorly and contracts over time. In embryos treated with bcd RNAi, the central domain of Kr expands towards the anterior (94/116; Fig. 4C–E, I–K, O–Q; Fig. 5B–F; S1 File). Yet again, the extent of the expansion is correlated with the strength of the knock-down. In the strongest cases, Kr expression is entirely missing (22/116; Fig. 4E, K, Q). A similar expansion of the central Kr domain has been observed in embryos from bcd mutant mothers in D. melanogaster [24]. However, these embryos never show a complete lack of Kr expression; it is only abolished by the additional removal of maternal hb [24,54]. Knock-down of bcd in E. balteatus, which lacks maternal hb expression altogether, leads to a complete absence of Kr expression in all RNAi-treated embryos [38].
In summary, our results suggest that Bcd is a concentration-dependent transcriptional regulator of gap genes in M. abdita. The observed effects of Bcd on gap gene expression are more severe than in D. melanogaster (resembling gap gene patterns in mutants affecting both bcd and hb), but milder than in E. balteatus.
M. abdita embryos treated with bcd RNAi can exhibit a bicaudal phenotype with complete axis polarity reversal and mirror-duplicated posterior structures in the anterior [31]. These severe knock-down phenotypes have their plane of symmetry at abdominal segment 5 (A5), and express four eve stripes—the two anterior ones probably being mirror-duplicated stripes 6 and 7 [31]. Such polarity reversal is never observed in embryos derived from bcd mutant mothers in D. melanogaster [41], only in embryos that lack both bcd and maternal hb [24,43,54]. While the former still have a residual Kr domain, the latter lack Kr expression completely. Polarity reversal is also observed in E. balteatus embryos treated with bcd RNAi, which show no Kr expression at all [38].
We tested the relationship between the bicaudal phenotype and the presence or absence of Kr by co-staining bcd knock-down embryos for both eve and Kr (Fig. 5). The pair-rule gene eve is expressed in seven stripes in wild-type M. abdita embryos (Fig. 5A) [44,55,56]. Weak bcd knock-down phenotypes show a full complement of seven eve stripes that are displaced towards the anterior, with a correspondingly mild anterior displacement of Kr (Fig. 5B; compare to Fig. 4C, I, O). Increasing severity of the knock-down results in the progressive loss of anterior eve stripes and more pronounced anterior displacement of the central Kr domain (Fig. 5C–E; compare to Fig. 4D, J, P). In the strongest cases, we detect four eve stripes only (as in [31]), and no or very little Kr expression (Fig. 5F; compare to Fig. 4E, K, Q). This suggests that the absence of Kr expression is correlated with polarity reversal in bcd knock-down embryos.
Why does lack of Bcd induce a bicaudal phenotype in M. abdita if it has a maternal Hb gradient very similar to D. melanogaster? To answer this question, we compared the role of maternal Hb in gap gene regulation in both species.
We have previously characterised the effect of Hb on Kr, kni, and gt in M. abdita [48]. Expression of kni and gt in embryos treated with hb RNAi is very similar to the corresponding patterns in hb mutants of D. melanogaster. In contrast, the effect of Hb on Kr differs between the two species: both show an anterior expansion of the central Kr domain (24 out of 53 RNAi-treated embryos in M. abdita), but only D. melanogaster embryos lacking maternal Hb exhibit a decrease in Kr expression levels [24,54]. We never observe such down-regulation in M. abdita embryos treated with hb RNAi (S2 Fig) [48]. Together with the absence of Kr expression in strong bcd knock-down phenotypes (Fig. 4E, K, Q, Fig. 5E), this indicates that Hb is unable to activate Kr in M. abdita.
In contrast, several authors have interpreted the reduced levels of Kr expression in hb mutants as evidence for activation of Kr by Hb in D. melanogaster [24,42]. However, it has never been shown whether this activating effect is direct or indirect—via repression of the repressor Kni by Hb (see [5], for a detailed discussion). To distinguish between these two possibilities, it is necessary to suppress kni in a background lacking maternal and zygotic hb. Direct activation is supported if levels of Kr expression remain low in embryos lacking both hb and kni, while an indirect effect via kni is supported if Kr levels are restored in these embryos compared to hb mutants alone. Unfortunately, it is not straightforward to create such double mutants, since both hb and kni are located on the same chromosome in the D. melanogaster genome, and germ line clones must be induced to eliminate both maternal and zygotic activities of hb. This may be the reason why this experiment has never been carried out. To overcome this challenge, we used RNAi-mediated double knock-down of hb and kni, and knock-down of hb in a kni mutant background.
In D. melanogaster hb knock-down embryos, we observe anterior expansion and strong down-regulation of Kr (5/9; Fig. 6A, B; S2 File), as well as considerable anterior displacement of kni (3/5; S3 Fig; S2 File). These patterns correspond precisely to Kr and kni expression in embryos mutant for both maternal and zygotic hb [24]. Similarly, kni knock-down embryos show a Kr pattern which is identical to that observed in kni null mutants: we observe no posterior expansion of Kr (Fig. 6E; S2 File), in accordance with a recent quantitative study [57], but in disagreement with earlier qualitative reports [58–60]. These results indicate that our early embryonic RNAi knock-downs mimic strong null mutant phenotypes.
In D. melanogaster hb/kni double knock-down embryos, we observe an anterior expansion of Kr, but no restoration of expression levels (12/18; Fig. 6C; S2 File). We confirm this result in kni mutant embryos treated with hb RNAi, which exhibit an identical anterior expansion of Kr and no restoration of expression levels (12/14; Fig. 6F; S2 File). Taken together, these results demonstrate that kni is not responsible for Kr down-regulation in D. melanogaster embryos lacking maternal and zygotic Hb. Therefore, activation of Kr by Hb is direct in this species. In contrast, this activatory role is absent in M. abdita where Hb acts as a repressor only, which leads to a lack of Kr expression and mirror symmetrical expression of the remaining gap genes in bcd knock-down embryos (see also Conclusions).
In D. melanogaster, maternal and zygotic Cad contribute to the activation of posterior gap domains [13,42] and—at least partially independently of gap gene regulation—activate posterior stripes of pair rule gene expression [11,50,61–64]. To investigate the exclusively zygotic contribution of Cad to gap and pair-rule gene expression in M. abdita, we characterised the expression patterns of hb, gt, Kr, kni, and eve (Fig. 7A–L; S1 File) as well as the cuticle phenotype (Fig. 7O) of embryos treated with cad RNAi. The cad knock-down phenotype of M. abita exhibits deletions of all segments posterior of T3, and T3 itself is also disrupted in some embryos (Fig. 7O). This phenotype is more similar to D. melanogaster than to E. balteatus. Embryos of the latter treated with cad RNAi exhibit a strongly reduced cephalopharyngeal skeleton, in addition to an almost complete loss of abdomen and thorax [44]. In contrast, D. melanogaster embryos mutant for both maternal and zygotic cad have an intact head and thorax and, although there is extensive loss of abdominal segments, often even retain some abdominal structures [11]. The fact that the M. abdita phenotype is stronger than that of D. melanogaster suggests that cad still plays an essential role in posterior segmentation in this species despite the loss of its maternal contribution.
In light of this, it is surprising that knock-down of cad in M. abdita does not have a strong effect on gap gene expression. The only clearly detectable defect is a slightly reduced posterior hb domain (7/16; Fig. 7A–C). All other domains of hb, gt, Kr, and kni seem unaffected (Fig. 7A–L; see also S1 File). Expression levels of Kr, kni, and gt appear similar to wild-type, although we cannot completely rule out a marginal decrease due to lack of sensitivity of our enzymatic detection method. This stands in contrast to D. melanogaster, where expression levels in the abdominal domain of kni and the posterior domain of gt are reduced in mutants lacking both zygotic and maternal cad (while hb and Kr are expressed as in wild-type) [13,42,50]. In E. balteatus cad knock-down embryos, anterior hb and Kr are normal, while the posterior kni, gt, and hb domains are absent or severely reduced [38,44].
To test if activation of gap genes by Cad is present but redundant with the complementary contribution by Bcd, we characterised the expression of kni and gt in embryos treated with RNAi against both bcd and cad (Fig. 8; see also S1 File). We observe a large anterior displacement in the position of the abdominal kni domain (36/36), as is seen in bcd RNAi-treated embryos. This was associated with a strong reduction in expression levels, particularly before T3 (Fig. 8B; 14/15), though after this stage levels of expression begin to resemble those in the wild-type. Embryos treated with bcd or cad RNAi alone, never show such reduction (Fig. 4N–Q, Fig. 7J, K). Expression of gt is absent before T2 (6/10), and only becomes detectable as a weak posterior domain at later stages (Fig. 8D, F). In contrast to bcd RNAi-treated embryos (Fig. 4H–L), we do not observe any anterior displacement of this domain (see S1 File). In the anterior, we observe a cap of gt expression at the late blastoderm stage (Fig. 8H), which closely resembles the anterior cap in embryos treated with RNAi against bcd alone (Fig. 4I,J; S1 File). Our observations stand in contrast to those from D. melanogaster mutants lacking both maternal and zygotic cad and bcd. Such mutants show complete absence of both kni and gt expression [13].
Taken together, our results suggest that Cad contributes to early activation of both abdominal kni and posterior gt in M. abdita, in a way which is largely redundant with activation by Bcd. Surprisingly, late expression of both kni and gt in the posterior of the embryo seems to be at least partially independent of both Bcd and Cad activation. This suggests that a third, yet unknown, factor must contribute to gap gene activation in this species.
Finally, we investigated the contribution of M. abdita cad to pair-rule expression. In embryos treated with cad RNAi, we observe a reduction in the number of eve stripes: 2 out of 12 embryos showed three, 5/12 four, and 5/12 five eve stripes (Fig. 7M,N). Similarly, D. melanogaster embryos mutant for both maternal and zygotic cad have four eve stripes [50]. The most drastic effect of cad on pair-rule gene expression is observed in E. balteatus, where embryos treated with cad RNAi exhibit the loss of all but the first stripe of eve [44].
Taken together, our evidence demonstrates that zygotic cad still plays an important role in the determination of posterior segments of M. abdita. In contrast to D. melanogaster and E. balteatus, where eliminating cad has a clearly detectable effect on gap gene expression [13,42,44], it is largely redundant for gap gene activation in M. abdita. This implies that cad performs its pattern-forming role mainly at the level of the pair-rule genes in this species.
In this study, we have investigated the establishment of maternal gradients and their role in gap gene regulation in the scuttle fly M. abdita. We compare our results with the evidence from the vinegar fly D. melanogaster as well as the marmalade hoverfly E. balteatus (Fig. 9). On the one hand, we find that important aspects of maternal regulation are highly conserved among cyclorrhaphan flies. Bcd acts as a concentration-dependent transcriptional regulator, and Cad is involved in posterior patterning in all three species. On the other hand, we find a number of interesting differences between M. abdita, E. balteatus, and D. melanogaster.
The first difference concerns the regulation of cad. Even though maternal cad expression can be detected in nematocerans, and basally branching non-cyclorrhaphan brachycerans (Fig. 1), maternal expression of cad has been lost in M. abdita [47]. Zygotic expression of cad is qualitatively similar between species, but reaches further anterior in M. abdita and E. balteatus than in D. melanogaster, creating a large overlap of cad and hb in these flies. Consistent with the absence of strong repression between these two genes, hb only weakly affects cad expression in M. abdita. In contrast, cad is completely de-repressed anteriorly in bcd knock-down embryos (see Fig. 3). There is some evidence from reporter assays that Bcd may regulate cad transcriptionally in D. melanogaster as well [47]. The situation is much less ambiguous in the case of E. balteatus, where cad is strongly up-regulated in the anterior upon bcd RNAi knock-down [38]. This similarity between M. abdita and E. balteatus suggests that transcriptional repression of cad by Bcd is much more prominent in these flies compared to D. melanogaster. Whether this interaction is direct in any of the three species remains to be shown.
The second difference concerns the roles of bcd and hb in axis specification and gap gene patterning. Knock-down of bcd in M. abdita and E. balteatus leads to bicaudal phenotypes, as observed in bcd/hb double mutants but not in bcd mutants in D. melanogaster [24,41–43]. It is important to note that the situation in M. abdita is distinct from both D. melanogaster and E. balteatus (Fig. 9). More positional information is retained in bicaudal embryos, resulting in a more anterior (A5) plane of symmetry, compared to A6 in the latter two species [24,31,38]. This difference is also reflected at the level of gap gene expression. Severe M. abdita knock-down phenotypes for bcd, which lack Kr expression, show a sequence of hb-gt-kni-gt-hb domains along the antero-posterior axis (Fig. 9) (this paper and [31]). D. melanogaster hb/bcd double mutants only have overlapping central gt and kni domains (Fig. 9) [24,51,52,65]. E. balteatus knock-down embryos show an almost complete de-repression of gt and kni throughout the embryo (Fig. 9) [38].
The anterior gradient of Bcd is an evolutionary innovation of the cyclorrhaphan lineage (Fig. 1) [30–33]. The evidence suggests that it is completely sufficient for axis specification and embryo polarity in M. abdita and E. balteatus. In contrast, both maternal Bcd and Hb contribute synergistically to axis specification and gap gene patterning in D. melanogaster. While differences in the effect of Bcd between D. melanogaster and E. balteatus are easily explained by the absence of maternal hb in the latter [38], it is less straightforward to pinpoint the cause for polarity reversal in bcd knock-down embryos of M. abdita. Our evidence suggests that this difference lies in the ability of maternal Hb to activate Kr in D. melanogaster, but not M. abdita (see Fig. 4, and [48]). Kr expression in the anterior of the embryo is correlated with the maintenance of polarity in D. melanogaster bcd mutants, and weak bcd knock-down phenotypes in M. abdita (Figs. 4 and 5). In D. melanogaster, maternal Hb is required for Kr expression in the absence of Bcd [24,42], and we have shown here that this activating interaction is indeed direct and not caused by the indirect repression of the Kni repressor (Fig. 6).
It remains unclear whether activation of Kr by maternal Hb has been gained in D. melanogaster or lost in M. abdita. However, there is some evidence that favours the latter scenario. Maternal hb expression is strongly conserved across arthropods far beyond the cyclorrhaphan lineage [66–74], and hb is involved in axis patterning in many of the species where it has been studied [67,68,70–72,75,76]. Most interestingly in our context, Hb activates Kr in the flour beetle Tribolium castaneum [75], the honeybee Apis mellifera [72], the hemipteran milkweed bug Oncopeltus fasciatus [76], and the cricket Gryllus bimaculatus [69]. The fact that this activating role of hb is conserved, and is only present in the one cyclorrhaphan species that retains some activity of maternal Hb in axis formation, seems to suggest that it may represent the ancestral state, and that activation of Kr by Hb has been lost in M. abdita and E. balteatus.
We have previously demonstrated that the gap gene system of M. abdita compensates for the significant differences in the distribution of maternal factors compared to D. melanogaster, such that gap gene expression converges to equivalent patterns in both species by the onset of gastrulation [48]. Such compensatory evolution is called developmental system drift or phenogenetic drift [77–81]. At the level of the gap genes, this is achieved through quantitative changes in the strength of otherwise wholly conserved gap-gap interactions [48]. In contrast, our study shows that system drift at the level of maternal-to-gap interactions is mediated by both quantitative and qualitative differences in gene regulation. While inter-species differences in the effect of Bcd and Cad mainly consist in changes in activation strength, the activating role of Hb on Kr has changed in a qualitative way: while Hb activates Kr in D. melanogaster, this activating role is absent in both M. abdita and E. balteatus (Fig. 9).
In summary, we observe a trend towards replacing the role of maternal Hb with activity of the anterior maternal system—Bcd and Cad—in non-drosophilid cyclorrhaphan lineages through the process of developmental system drift. This is reflected by the stronger phenotypes of bcd and cad knock-downs in both E. balteatus and M. abdita compared to D. melanogaster. In this view, axis formation and gap gene patterning in D. melanogaster retains more ancestral characteristics than these early-branching non-drosophilid cyclorrhaphans. Further corroboration of these insights will have to come from functional studies of axis specification and gap gene patterning in an appropriate outgroup (Fig. 1): non-cyclorrhaphan brachycerans or emerging nematoceran model systems such as the chironomid midge Chironomus riparius or the moth midge Clogmia albipunctata.
M. abdita fly culture, embryo collection and fixation were carried out as described in [82,83]. Enzymatic mRNA in situ hybridisation, image acquisition, and data processing were carried out as described in [84,85]. We use an embryonic staging scheme—homologous to the one already established for D. melanogaster [86]—which is described in detail in [56]. Embryo morphology and developmental timing are remarkably similar in both species. Embryos are classified into cleavages cycles C1–C14A according to nuclei number; C14A is further subdivided into eight time classes T1–8 based on nuclear and membrane morphology.
Polyclonal antiserum was raised against M. abdita Hb protein expressed by means of a pET-DEST42 vector (Invitrogen) containing a full length cDNA insert. Purified Hb protein dissolved in 6M urea was used to raise rat antibodies by Primm Biotech (primmbiotech.com) using standard protocols. For antibody stains, wild-type blastoderm-stage embryos were collected after 4 hrs of egg laying and stained with a colorimetric protocol adapted from the in situ protocol published in [85]. In brief, fixed and dehydrated embryos were re-hydrated by washing 1x5min in PBT/methanol (embryos were allowed to sink before the solution was removed), 2x in PBT, and 1x20 min in PBT. Embryos were washed 1x, then blocked with 2x30 min in western blocking reagent (Roche) in PBT followed by incubation with primary antibodies in blocking solution overnight. Unbound antibody was removed washing 3x in PBT followed by 4x15 min washes in PBT. Embryos were then re-blocked and incubated with secondary antibodies conjugated with alkaline phosphatase (Roche) at 1:3000 in blocking solution for 1 hr. Unbound antibody was removed as before. To prepare for staining, embryos were washed 2x5 min in AP buffer (100 mM NaCl, 50 mM MgCl, 100 mM Tris pH 9.5, 0.1% tween-20). Staining was carried out in the dark by the addition of AP buffer containing 0.1 mg/ml NBT and 0.05 mg/ml BCIP. Staining was stopped with 3x1 min followed by 3x10 min washes in PBT. Nuclei were counter-stained by a 10-min incubation in PBT containing 0.3 μM DAPI, followed by 3x washes and 3x10 min washes in PBT. Embryos were cleared through a series into 70% glycerol:PBS, of which 30 μl were mounted per slide. All washes were done on a nutator.
We used RNAi knock-down protocols adapted from [31,37,87]. See [48] for further details.
All expression boundaries plotted as graphs were extracted from NBT/BCIP stained embryos, except for Kr expression in M. abdita bcd RNAi-treated embryos, where boundaries were extracted from FastRed stains. Differences in expression levels in Fig. 6 and S2 Fig were assessed through simultaneous staining of wild-type and RNAi-treated embryos using NBT/BCIP to ensure a robust signal.
Quantified expression data for M. abdita wild-type and RNAi knock-down embryos are available online through figshare (http://dx.doi.org/10.6084/m9.figshare. 1252195; [88], and the SuperFly database (http://superfly.crg.eu; [89]). Plots of gene expression boundaries from RNAi-treated or mutant embryos can be found in S1 File (M. abdita) and S2 File (D. melanogaster).
nos (KP232978) was cloned from cDNA using data from our published early embryonic transcriptome (http://diptex.crg.es; MAB_comp4961) [46]. All other genes were cloned as described in [48].
Embryo collection, fixation, RNAi treatment, and in situ hybridisation in D. melanogaster was carried out as for M. abdita [85,87]. D. melanogaster kni mutants correspond to deletion strain 3127 (Bloomington Drosophila Stock Center) with genotype Df(3L)ri-79c/TM3, Sb1. Homozygous mutants were detected by an absence of FastRed kni staining during in situ hybridisation.
|
10.1371/journal.pgen.1007268 | Activation of DAF-16/FOXO by reactive oxygen species contributes to longevity in long-lived mitochondrial mutants in Caenorhabditis elegans | Mild deficits in mitochondrial function have been shown to increase lifespan in multiple species including worms, flies and mice. Here, we study three C. elegans mitochondrial mutants (clk-1, isp-1 and nuo-6) to identify overlapping genetic pathways that contribute to their longevity. We find that genes regulated by the FOXO transcription factor DAF-16 are upregulated in all three strains, and that the transcriptional changes present in these worms overlap significantly with the long-lived insulin-IGF1 signaling pathway mutant daf-2. We show that DAF-16 and multiple DAF-16 interacting proteins (MATH-33, IMB-2, CST-1/2, BAR-1) are required for the full longevity of all three mitochondrial mutants. Our results suggest that the activation of DAF-16 in these mutants results from elevated levels of reactive oxygen species. Overall, this work reveals an overlapping genetic pathway required for longevity in three mitochondrial mutants, and, combined with previous work, demonstrates that DAF-16 is a downstream mediator of lifespan extension in multiple pathways of longevity.
| The use of genetic model organisms has permitted the identification of a large number of genes that influence longevity. These genes have been grouped into different pathways of lifespan extension, which have been proposed to modulate longevity by distinct mechanisms. In this work, we explore the mechanisms underlying longevity in three long-lived mitochondrial mutants in C. elegans. We find that all three mutants show upregulation of DAF-16/FOXO target genes and that DAF-16 as well as multiple proteins that function with DAF-16 are required for their longevity. Since DAF-16 has previously been shown to be responsible for the increase in lifespan resulting from decreasing insulin-IGF1 signaling, this indicates that different pathways of lifespan extension have overlapping mechanisms, and that DAF-16/FOXO is a common downstream mediator of longevity.
| Mitochondria serve a variety of critical functions within the cell including energy production, intracellular signaling and metabolism. Accordingly, major disruptions of mitochondrial function lead to detrimental effects including lethality. In contrast, mild disruption of mitochondrial function has been shown to increase lifespan.
The first demonstration that impairment of mitochondrial function could extend longevity came from the worm C. elegans. The gene clk-1 was identified in a screen for maternal effect mutations that slow development and defecation cycle length. Mutations in clk-1 were found to increase lifespan [1,2], and this gene was later found to encode a hydroxylase involved in the biosynthesis of ubiquinone [3]. This same screen identified nine other clk genes, all of which decrease mitochondrial function and increase lifespan [4]. Since then multiple other genes affecting mitochondrial function have been shown to increase lifespan in C. elegans, including isp-1, which encodes the Rieske iron-sulfur subunit of complex III of the mitochondrial electron transport chain [5], nuo-6, which encodes a subunit of complex I of the mitochondrial electron transport chain [6], lrs-2, which encodes a mitochondrial leucyl-tRNA synthetase [7], and sod-2, which encodes the primary mitochondrial superoxide dismutase [8]. In addition, an unbiased RNAi screen for genes that increase lifespan identified many genes involved in mitochondrial function [7].
The ability of decreasing mitochondrial function to extend longevity is not limited to C. elegans. As in worms, it has been shown that decreasing the expression of components of the mitochondrial electron transport chain in Drosophila increases lifespan [9]. Similarly, knocking down the expression of complex I subunit ND75 in Drosophila was also shown to extend longevity [10]. In mice, it has been shown that a mutation affecting the cytochrome c oxidase assembly factor SURF1 reduces mitochondrial function, but increases lifespan [11]. In addition, a heterozygous mutation in the mouse homolog of clk-1, Mclk1, causes mitochondrial dysfunction [12], and extended longevity [13]. Thus, the ability of a mild decrease in mitochondrial function to increase lifespan is conserved across species.
While a number of groups have studied the longevity resulting from impairment of mitochondrial function and identified factors involved [14–19], the mechanism by which decreasing mitochondrial function increases lifespan has not been fully elucidated. It is also uncertain to which extent the mechanism of lifespan extension is the same between different mitochondrial mutants, and the extent to which these mechanisms overlap with other pathways of lifespan extension.
One of the most well-studied pathways of lifespan extension is the insulin-IGF1 signaling pathway. The two first genes that were shown to increase lifespan in any species are part of this nutrient signaling pathway. Mutations of the insulin-IGF1 receptor daf-2 [20], or other genes in the signaling pathway [21], can double the lifespan of the worm. This increase in lifespan is completely dependent on the FOXO transcription factor DAF-16 [20,22,23]. Mutations in the daf-2 gene cause decreased insulin-IGF1 signaling, which leads to increased nuclear localization of DAF-16 [24] and altered expression of DAF-16 target genes [25–27]. At least some of these changes in gene expression are required for daf-2 longevity [25].
In addition to its role in the insulin-IGF1 signaling pathway, DAF-16/FOXO has been shown to respond to various forms of stress. The nuclear localization of DAF-16 can be induced by heat stress, anoxia, oxidative stress, starvation and exposure to pathogenic bacteria [24,28]. As daf-16 mutants show increased sensitivity to multiple types of stress [29–31], and overexpression of DAF-16 results in increased resistance to stress [24], it appears that DAF-16 also has an important role in stress resistance.
A number of previous studies have examined the interaction between mild impairment of mitochondrial function and the insulin-IGF1 signaling pathway using either genetic mitochondrial mutants [2,5], or RNAi against genes affecting mitochondrial function [7,32,33]. These studies have shown that loss of daf-16 decreases the lifespan of long-lived worms with decreased mitochondrial function. However, these experiments also found that decreasing daf-16 levels also shortens the lifespan of wild-type worms, making the role of daf-16 in the longevity of the mitochondrial mutants and mitochondrial RNAi difficult to interpret without further experimentation.
In this work, we examine overlapping changes in gene expression in three long-lived mitochondrial mutant strains to identify common mediators of longevity. We find that all three strains show a modulation of DAF-16 target gene expression. Our results suggest that elevated levels of ROS in these mitochondrial mutants cause the activation and nuclear localization of DAF-16. Importantly, we find that DAF-16 and multiple DAF-16 interacting proteins are required for the full longevity of these mutants. This work suggests that there are converging downstream mechanisms contributing to lifespan extension between different mitochondrial mutants, and between long-lived mitochondrial mutants and other pathways of lifespan extension.
To identify overlapping genetic pathways that contribute to longevity in long-lived mitochondrial mutants, we compared gene expression in clk-1, isp-1 and nuo-6 worms to wild-type worms using RNA sequencing (RNAseq). RNA was isolated from six biological replicates per strain and sequenced individually. We observed a striking degree of overlap in the patterns of gene expression (see S1 Table for a complete list of significantly upregulated and downregulated genes). Of all of the genes upregulated in these strains, 18% were upregulated in all three strains, while 40% were upregulated in at least two strains (Fig 1A). Similarly, we found that 7% of all of the downregulated genes exhibited decreased expression in all three strains, and 27% were downregulated in at least two of the strains (Fig 1B). Of the genes that were commonly upregulated, we observed a number of known DAF-16 target genes including sod-3, dod-3, mtl-1, sodh-1, ftn-1, gpd-2 and icl-1. Accordingly, we chose to focus the remainder of our study on the role of DAF-16 in the longevity of the long-lived mitochondrial mutants.
To confirm our observation that DAF-16 target genes are upregulated in long-lived mitochondrial mutants, we tested eight specific DAF-16 target genes using quantitative real-time RT-PCR. These genes were chosen from the list of the top DAF-16 target genes identified in a meta-analysis of previous DAF-16 gene expression studies by Tepper et al., 2013 [27]. As with the data from RNA sequencing, we found that all eight DAF-16 target genes were significantly upregulated in clk-1, isp-1 and nuo-6 mutants (Fig 1C–1J).
Since the long lifespan of the insulin-IGF1 receptor mutant daf-2 is thought to be mediated by DAF-16, we next sought to determine the extent to which gene expression changes present in the long-lived mitochondrial mutants overlapped with gene expression changes in daf-2 worms. Accordingly, we performed RNA sequencing on daf-2(e1370) worms and compared the changes in gene expression to the long-lived mitochondrial mutants. Of the genes that were found to be upregulated in clk-1, isp-1 and nuo-6 worms, 46%, 50% and 57%, respectively, were also upregulated in daf-2 mutants (Fig 2A), which represents a highly significant degree of overlap (Table 1). Of the genes that were found to be downregulated in clk-1, isp-1 and nuo-6 worms, 51%, 36% and 42%, respectively, were also downregulated in daf-2 mutants (Fig 3A), which also represents a statistically significant degree of overlap (Table 1). Thus, there is considerable overlap between the transcriptional response in the long-lived mitochondrial mutants and decreasing insulin-IGF1 signaling through daf-2 mutation.
A previous study analyzed all gene expression studies that compared conditions with different levels of DAF-16 to identify DAF-16 up-regulated and DAF-16 down-regulated gene sets [27]. Accordingly, we compared our RNA sequencing results with the top-ranked DAF-16-modulated genes identified in the previous study. We focused on genes identified in the study by Tepper et al. that were significantly modulated in our daf-2 RNAseq data. We found that the genes that are upregulated in clk-1, isp-1 and nuo-6 mutants are significantly enriched for the DAF-16 upregulated genes (Fig 2; Table 1) and that DAF-16-down-regulated genes are also down-regulated in the mitochondrial mutants (Fig 3; Table 1).
To ensure that the upregulation of DAF-16 target genes in the long-lived mitochondrial mutants is being mediated by DAF-16, and not by other transcription factors that can activate the same genes, we knocked down daf-16 expression in the mitochondrial mutant strains using RNAi, and compared gene expression to worms treated with an empty vector RNAi bacteria. We examined five DAF-16 target genes (sod-3, dod-3, mtl-1, sodh-1 and ftn-1) and found that the increased expression of these genes in the long-lived mitochondrial mutants was significantly reduced (sod-3, sodh-1) or prevented (dod-3, mtl-1, ftn-1) by knocking down DAF-16 (Fig 4A–4E). This indicates that DAF-16 is required or partially required for the upregulation of the DAF-16 target genes in clk-1, isp-1 and nuo-6 worms. A DAF-16-dependent upregulation of sod-3 in isp-1 worms was also previously observed by others [5]. As with the quantitative real-time RT-PCR results, we found that the long-lived mitochondrial mutants exhibit increased fluorescence from a Psod-3::GFP reporter strain [34], which is diminished by daf-16 RNAi (Fig 4F).
To further explore DAF-16 activation in the long-lived mitochondrial mutant strains, we examined the nuclear localization of DAF-16. When DAF-16 is activated it moves to the nucleus, and thus increased nuclear localization of DAF-16 is indicative of activation. Accordingly, we crossed clk-1 and isp-1 worms to a reporter strain expressing DAF-16 linked to GFP (zIs356[Pdaf-16::daf-16:GFP])[24]. To ensure that we could visualize all of the DAF-16 protein present, we performed these experiments in a daf-16 deletion mutant background, as has been done previously [35]. While control worms (daf-16;zIs356) exhibited diffuse cytoplasmic expression of DAF-16:GFP, both clk-1 and isp-1 worms showed increased nuclear localization of DAF-16:GFP compared to wild-type worms (Fig 4G; S1 Fig). Note that we could not test nuclear localization in nuo-6 worms because we were unable to generate nuo-6;daf-16;zIs356 worms due to the close proximity of nuo-6 and daf-16 on chromosome I. In addition, we did not explore the effect of nuo-6 RNAi on DAF-16 nuclear localization since the nuo-6 mutation and nuo-6 RNAi increase lifespan by independent mechanisms and exhibit different changes in gene expression [6]. Nonetheless, our results indicate that DAF-16 is activated in long-lived mitochondrial mutant strains, leading to increased nuclear localization and altered expression of DAF-16 target genes.
The loss of DAF-16 has previously been shown to completely prevent the long lifespan of daf-2 insulin-IGF1 receptor mutants [20] and glp-1 mutants [36–38]. Since DAF-16 is activated in the long-lived mitochondrial mutants, and the changes in gene expression present in these mutants exhibit a high degree of overlap with those of daf-2 worms, we examined the effect of knocking down daf-16 levels by RNAi on the lifespan of clk-1, isp-1 and nuo-6 worms. In each case, we found that daf-16 RNAi markedly decreased the lifespan of the long-lived mitochondrial mutant (Fig 5A–5C). Similarly, as had been previously observed [20,37], daf-16 RNAi completely prevented mutations in either daf-2 or glp-1 from increasing lifespan (Fig 5D and 5E).
To rule out the possibility that daf-16 RNAi was decreasing lifespan through off target effects, we also examined the effect of a daf-16 deletion mutation on mitochondrial mutant longevity. We chose to use the mu86 allele as this mutation affects all transcripts of daf-16. We measured the lifespan of clk-1;daf-16(mu86), isp-1;daf-16(mu86) and daf-2;daf-16(mu86) double mutants (we were unable to generate nuo-6;daf-16(mu86) worms due to the close proximity of nuo-6 and daf-16 on chromosome I).
As with daf-16 RNAi, we found that a null mutation in daf-16(mu86 allele) completely prevented the increase in lifespan observed in clk-1, isp-1 and daf-2 worms (Fig 6). Note that the effect of the daf-16(mu86) mutation on lifespan is dependent on the experimental conditions: at 20°C daf-16(mu86) worms exhibit a similar lifespan to wild-type worms (equal or mildly decreased), while at 25°C daf-16(mu86) worms show a marked decrease in longevity compared to wild-type worms (S2 Fig). We chose to perform these experiments at 20°C where the daf-16(mu86) mutation has either no effect on longevity (Fig 6) or causes a small reduction in lifespan (S2 Fig). In both cases, we found that deletion of daf-16(mu86) markedly reduced clk-1 and isp-1 lifespan (Fig 6, S2 Fig).
As previous work using a daf-16 point mutation (m26) reported milder effects of daf-16 mutation on mitochondrial mutant longevity [2,5], we sought to determine if the magnitude of lifespan decrease is related to the severity of the daf-16 allele. Accordingly, we compared the lifespan of isp-1 worms to isp-1;daf-16(m26) worms. As with the daf-16(mu86) allele, we found that the daf-16(m26) allele significantly decreased isp-1 lifespan (Fig 6). The percentage decrease was 48% for the daf-16(mu86) allele and 36% for the daf-16(m26) allele, indicating that the milder allele has a lesser impact on isp-1 longevity.
To further explore the role of DAF-16 in the lifespan of the long-lived mitochondrial mutants, we calculated the percentage increase in lifespan resulting from mutations in clk-1, isp-1 and nuo-6 under basal conditions and conditions in which daf-16 expression is decreased. We found that the average increase in lifespan resulting from the clk-1, isp-1, and nuo-6 mutation on EV RNAi was 48%, 72%, and 85% respectively. The average increase in lifespan resulting from the clk-1, isp-1, and nuo-6 mutation on daf-16 RNAi was 17%, 19%, and 34% respectively. In each case, the difference in magnitude of lifespan increase on EV RNAi is statistically different from daf-16 RNAi (S3 Fig). Similarly, the maximum lifespan resulting from the clk-1, isp-1, and nuo-6 mutations on EV RNAi was 45%, 99% and 90%, compared to 25%, 38% and 50% on daf-16 RNAi. These results indicate that the ability of the mitochondrial mutations to increase lifespan is greatly diminished when daf-16 is knocked down using RNAi. Similar results were obtained in analyzing the effect of the daf-16 mutation (S3 Fig).
We next examined the effect of increasing DAF-16 levels on longevity in the mitochondrial mutant strains. To do this, we crossed clk-1, isp-1 and nuo-6 worms to zIs356[Pdaf-16::daf-16:GFP] worms. It should be noted that previous work has shown that the lifespan of zIs356 worms is dependent on the concentration of FUdR in the media. While it was initially found that the lifespan of zIs356 worms is similar to wild-type worms when FUdR is absent [24], subsequent studies showed that the lifespan of zIs356 worms is increased compared to wild-type worms on plates containing FUdR [39,40]. We chose to complete our studies using conditions under which the daf-16 transgene increases lifespan (100 μM FUdR). As previously reported, we found that when FUdR is present, zIs356 worms exhibit increased lifespan (S4A Fig). Overexpression of DAF-16 also increased the lifespan of clk-1, isp-1 and nuo-6 worms, but not daf-2 worms (S4B–S4E Fig). The ability of DAF-16 to increase lifespan was greatest in clk-1 worms followed by isp-1 worms, nuo-6 worms and daf-2 worms (S4F Fig). This pattern is inversely related to the magnitude of upregulation of DAF-16 target genes in the single mutant strains (Fig 1; clk-1 < isp-1 < nuo-6 < daf-2). Combined these results suggest that there may be a threshold for DAF-16 activation with respect to longevity and that daf-2 and daf-2;zIs356 worms are at or above this threshold, while the individual mitochondrial mutants are all below the threshold such that lifespan continues to increase with further DAF-16 activation.
Having shown that DAF-16 is required for the longevity of the mitochondrial mutants as is true for daf-2 worms, we next sought to determine the extent to which there is a common underlying mechanism. Previous work has demonstrated that the levels of reactive oxygen species (ROS) are increased in clk-1, isp-1, nuo-6, daf-2 and glp-1 mutants, and importantly that this increase in ROS is required for their longevity [14,41–44]. In addition, a number of experiments have shown that DAF-16 can be activated by ROS [24,35,45,46]. Accordingly, we hypothesized that one of the common underlying mechanisms of lifespan extension in all of these mutants would be the activation of DAF-16 by ROS.
To test this hypothesis, we examined the effect of increasing ROS on the nuclear localization of DAF-16 and expression of DAF-16 target genes. We treated worms with two different compounds that increase intracellular levels of ROS: 4 mM paraquat or 300 μM juglone. As had been previously observed [24], we found that increasing ROS resulted in nuclear localization of DAF-16:GFP, similar to what we had observed in clk-1 and isp-1 mutants (Fig 7A, S1 Fig). Increasing the levels of ROS through treatment with 4 mM paraquat also resulted in the upregulation of DAF-16 target genes, including dod-3, mtl-1, sodh-1 and ftn-1 (Fig 7B–7E). Importantly, the upregulation of these genes was prevented by the daf-16(mu86) mutation (Fig 7B–7E). Note that we and others have previously shown that treating worms with either paraquat or juglone can extend longevity [14,41,47,48].
To further explore the role of elevated ROS, we examined gene expression in sod-2 mutants, since these worms have a decreased ability to specifically detoxify mitochondrial ROS and, similar to the mitochondrial mutants, have increased lifespan [8]. As with the long-lived mitochondrial mutants, we found that many DAF-16 target genes are upregulated in sod-2 worms (Fig 7F; see Fig 1C–1J to compare to results from mitochondrial mutants), and that the upregulation of these genes is dependent on DAF-16 (Fig 7G; see Fig 3A–3E to compared to results from mitochondrial mutants).
Next, we compared RNAseq results between sod-2 and daf-2 worms. Among the genes that are upregulated in sod-2 mutants, 36% are also upregulated in daf-2 worms (S5 Fig). Among the genes that are downregulated in sod-2 worms, 24% are also downregulated in daf-2 worms (S5 Fig). Thus, as with the mitochondrial mutants, there is considerable overlap in the transcriptional changes present in sod-2 and daf-2 worms. Since sod-2 can be transcriptionally regulated by DAF-16, it is possible that the overlap in gene expression between sod-2 and daf-2 worms could results from being in the same pathway (daf-2 → daf-16 → sod-2 → daf-16 target genes), but, given the function of sod-2, we believe it is more likely that the elevated ROS in sod-2 mutants causes DAF-16 activation. Finally, we examined the effect of daf-16 RNAi on sod-2 lifespan and found that the long life of sod-2 worms is completely abolished by daf-16 RNAi (Fig 7H). Combined, our data suggests that elevated ROS can cause the nuclear localization of DAF-16, activation of DAF-16 target genes, and a DAF-16-dependent increase in lifespan.
Having shown that increasing ROS levels can activate DAF-16 target genes, we next sought to determine whether elevated levels of ROS contribute to modulation of DAF-16 target genes in the long-lived mitochondrial mutants. First, we sought to confirm that ROS levels are increased in the mitochondrial mutant strains, and determine whether this increase is dependent on DAF-16. For this purpose, we stained wild-type, daf-16, clk-1, clk-1;daf-16, isp-1 and isp-1;daf-16 worms with the ROS-sensitive dye dihydroethidium (DHE). As we and others have previously observed [14,49], DHE fluorescence is significantly increased in clk-1 and isp-1 worms (S6 Fig). The increase in ROS levels in these mutants is independent of DAF-16, as clk-1;daf-16 and isp-1;daf-16 mutants both show elevated levels of ROS (S6 Fig). This indicates that DAF-16 activation does not contribute to the elevated ROS levels in these long-lived mitochondrial mutants.
Having shown that ROS levels are increased in the long-lived mitochondrial mutants, we next sought to determine whether these ROS activate DAF-16. If elevated ROS in the long-lived mitochondrial mutants are causing activation of DAF-16 leading to upregulation of DAF-16 target genes, then decreasing ROS levels through treatment with an antioxidant should prevent the upregulation of DAF-16 target genes. To test this we used the Psod-3::GFP reporter strain. We treated isp-1;Psod-3::GFP and nuo-6;Psod-3::GFP worms with three different antioxidants that have previously been shown to decrease levels of ROS in worms: ascorbic acid, butylated hydroxyanisole and sodium ascorbate [42,43]. In each case, we found that the antioxidant treatment decreased activation of the Psod-3::GFP reporter (Fig 7I), suggesting that reporter was being activated by elevated ROS. We previously showed that this same antioxidant treatment decreases the lifespan of isp-1 worms [47], clk-1 worms [50], and sod-2 worms [51]. While these results indicate that decreasing ROS levels with antioxidants, reduces the expression of the DAF-16 target gene sod-3, they do not exclude the possibility that other transcription factors, such as SKN-1, contribute to the ROS-mediated upregulation of sod-3 and other DAF-16 target genes.
It was recently shown that the deubiquitylase MATH-33/USP7 stabilizes DAF-16 during decreased insulin-IGF1 signaling by reducing DAF-16 ubiquitylation thereby preventing degradation by the proteasome [52]. As with mutations in daf-16, a loss of function mutation in math-33 was found to completely abolish the increased lifespan of daf-2 mutants. To determine if MATH-33 is also required for the longevity of the long-lived mitochondrial mutants, we knocked down math-33 expression using RNAi. We found that math-33 RNAi markedly reduced the lifespan of clk-1, isp-1 and nuo-6 worms (Fig 8A–8C). In accordance with previous work [52], math-33 RNAi shortened the lifespan of daf-2 worms (Fig 8D). Finally, we tested the role of math-33 in glp-1 germline ablation mutants and found that knocking down math-33 completely prevented their increase in lifespan (Fig 8E). To confirm the results of the RNAi experiments, we generated isp-1;math-33 and nuo-6;math-33 double mutants (note that we were unable to generate clk-1;math-33 double mutants possibly due to lethality). As with RNAi knockdown, deletion of math-33 also reduced the long life of both isp-1 and nuo-6 mutants (S7 Fig).
To test the extent to which math-33 is required for the upregulation of DAF-16 target genes in response to elevated ROS, we treated wild-type worms and math-33 deletion mutants with 4 mM paraquat and examined the expression of four DAF-16 target genes using quantitative real-time RT-PCR. We found that in each case, the ability of ROS to activate DAF-16 targets was diminished in math-33 mutants (Fig 8F and 8I).
The activation of DAF-16 target genes, especially those that are downregulated in response to decreased insulin-IGF1 signaling, have been shown to be co-regulated by the transcription factor PQM-1 [27]. PQM-1 appears to act in conjunction with DAF-16 to promote growth and development, or repair and maintenance depending on the environmental conditions. Accordingly, we tested the role of PQM-1 in the lifespan of the long-lived mitochondrial mutants. We found that knocking down pqm-1 expression using RNAi partially reduced the lifespan of clk-1, isp-1 and nuo-6 mutants (Fig 9A–9C). Similarly, pqm-1 RNAi also reduced the lifespan of daf-2, as previously noted [27], and glp-1 mutants (Fig 9D–9F).
Multiple proteins have been shown to interact with DAF-16 under conditions of elevated ROS to promote DAF-16 activation, nuclear localization of DAF-16, and changes in the expression of DAF-16 target genes. The protein kinase MST1 has been shown to phosphorylate FOXO in response to treatment with ROS, thereby disrupting its interaction with 14-3-3 proteins and promoting its nuclear localization [46]. Overexpression of cst-1, the C. elegans homolog of MST1, increases lifespan in a daf-16-dependent manner, while cst-1 RNAi reduces the lifespan of daf-2 worms [46]. Similarly, β-catenin has been shown to bind to FOXO in order to increase FOXO transcriptional activity in response to ROS. Loss of bar-1, the C. elegans homolog of β -catenin, results in decreased lifespan, and decreased upregulation of the DAF-16 target gene sod-3 in response to paraquat [45]. Finally, the nuclear import receptor transportin-1 (TNPO1) has been shown to form a disulfide complex with FOXO4 in response to ROS leading to nuclear localization of FOXO4 [35]. This mechanism is conserved in C. elegans as knockdown of the C. elegans homolog of transportin-1 imb-2 prevents the nuclear localization of DAF-16:GFP in response to paraquat-induced ROS [35].
Because CST-1/CST-2, BAR-1 and IMB-2 act with DAF-16 to upregulate DAF-16 target genes in response to ROS, we hypothesized that these genes might be important for the longevity of the long-lived mitochondrial mutants. Accordingly, we tested the effect of knocking down these three DAF-16 interacting proteins on the lifespan of clk-1, isp-1 and nuo-6 worms. We found that knocking down imb-2 or cst-1/cst-2 substantially decreased the lifespan of all three mitochondrial mutants (Fig 10A–10F). RNAi against imb-2 or cst-1/cst-2 also decreased the lifespan of daf-2 and glp-1 mutants (S8 Fig). However, it should be noted that the magnitude of the effect on daf-2 and glp-1 lifespan was less than in the mitochondrial mutants. Finally, we found that knockdown of bar-1 resulted in a small but significant reduction in clk-1, isp-1 and nuo-6 lifespan (Fig 10G–10I), but did not affect the longevity of daf-2 mutants (S8 Fig).
While the effect of knocking down bar-1 on mitochondrial mutant longevity was milder than knockdown of imb-2 or cst-1/2, it should be noted that the RNAi paradigm used was different. Because knockdown of bar-1 throughout development causes protruding vulva and externalization of internal organs even in wild-type worms, we began RNAi knockdown at the L4 stage, whereas imb-2 and cst-1/2 RNAi was begun before development. Nonetheless, the fact that knocking down any of these genes significantly decreases lifespan in all three mitochondrial mutants demonstrates that proteins that facilitate the activation and nuclear localization of DAF-16 in response to ROS are required for the full lifespan of the long-lived mitochondrial mutants.
The Free Radical Theory of Aging proposes that ROS produced by normal metabolism are one of the primary causes of aging. While it is clear that high levels of ROS are toxic and can decrease lifespan, it has been shown that mildly increasing ROS can be beneficial. Deletion of sod-2, the primary mitochondrial sod gene, causes increased lifespan [8], which is dependent on elevated ROS, since treatment with antioxidants reduces sod-2 lifespan [41,51]. The ability of ROS to increase lifespan depends on the location of ROS within the cell: while mitochondrial superoxide increases lifespan, cytoplasmic ROS decreases it [50,53]. Similarly, it has been shown that treating worms with ROS-generating compounds, either paraquat or juglone, both result in increased lifespan [14,41,47,48]. In addition, there have been multiple examples of genetic mutations or interventions that increase lifespan in which the increase in lifespan has been shown to be at least partially caused by elevated ROS [54–59], including the long-lived mitochondrial mutant strains studied here.
clk-1 worms have been shown to have increased levels of ROS [14,41,60,61] and this increase in ROS is required for their longevity [50]. Similarly, isp-1 worms exhibit increased levels of mitochondrial ROS [14,41,62], and decreasing this ROS through treatment with an antioxidant limits their longevity [41,47]. Finally, nuo-6 worms have elevated levels of mitochondrial superoxide [41], which is also required for their long lifespan [41].
While the mechanism underlying the longevity of daf-2 mutants was thought to be through a direct effect of the daf-2 mutation decreasing insulin-IGF1 signaling and cytoplasmic retention of DAF-16, it was demonstrated that elevated levels of ROS also contribute to the long lifespan of daf-2 worms [42]. Combined, this suggests that DAF-16 is activated by two routes in daf-2 mutants: decreased insulin-IGF1 signaling and increased ROS. Similarly, the mechanism underlying lifespan extension in glp-1 germline ablation mutants was thought to be mediated by DAF-16 as deletion of daf-16 completely prevents the increased lifespan of glp-1 mutants [36–38,63]. More recent evidence suggests that elevated ROS also contribute to the longevity of glp-1 mutants. These worms were shown to have increased levels of mitochondrial ROS, and quenching ROS through treatment with antioxidants reduced their lifespan [43]. Combined these results indicate that elevated ROS contribute to the longevity of mutants in multiple different pathways of lifespan extension.
It is important to note that the ability of ROS to increase lifespan is conserved across species. In yeast, increasing mitochondrial ROS, directly through treatment with menadione, or indirectly with rapamycin, causes increased chronological lifespan [64,65]. In flies, expression of the NADH dehydrogenase NDI1 [66], or knocking down the expression of complex I subunit ND75 [10], causes an increase in ROS that leads to extended longevity. Similar to clk-1 mutant worms, which have increased ROS that contributes to their longevity [1,2], Mclk1+/- mice also exhibit increased ROS and long life [13]. Finally, D-glucosamine was shown to increase lifespan in both worms and in mice as a result of increased levels of ROS [58].
In addition to acting through DAF-16, it is likely that mild elevation of ROS also increase lifespan through NRF2/SKN-1. SKN-1 is activated in response to oxidative stress, and targets of SKN-1 have been shown to be activated in clk-1, isp-1, nuo-6, daf-2 and glp-1 mutants, as well as by treatment with paraquat [28,43,49,50,67]. SKN-1 has been shown to be required for the increased longevity observed in isp-1, daf-2 and glp-1 mutants [43,68]. Interestingly, there are multiple examples of lifespan extending pathways in which both DAF-16 and SKN-1 are involved [68–70]. Thus, an important area for future research will be to define how DAF-16 and SKN-1 interact in response to elevated ROS to promote longevity.
As the mechanisms by which elevated ROS extend lifespan have not been fully elucidated, we sought to gain insight into these mechanisms by examining overlapping changes in gene expression in three long-lived mitochondrial mutants. We found that there is a highly significant enrichment of DAF-16 target genes among the genes that are modulated in clk-1, isp-1 and nuo-6 worms and that daf-16 is required for their longevity. On the surface, our results are surprising as they contradict a well-established belief in the aging research field that mild-impairment of mitochondrial function increases lifespan in a DAF-16-independent manner. This conclusion was reached by multiple different studies [2,5,7,19,32,33]. In these previous studies, it was shown that mild impairment of mitochondrial function by genetic mutation or RNAi increased the lifespan of both wild-type and daf-16 mutants. From these data, it was concluded that the lifespan increase caused by mild impairment of mitochondrial function is DAF-16 independent, and this idea has persisted in the field [71]. However, the data from these earlier papers also clearly show that a mutation in daf-16 decreases the long lifespan induced by RNAi or genetic mutations affecting mitochondrial genes, which is entirely consistent with our data, and indicates that daf-16 is required for the full increase in longevity caused by mild impairment of mitochondrial function.
The interpretation of this data is complicated by the fact that daf-16 RNAi or a mutation in daf-16 can decrease lifespan in wild-type worms. As a result, more evidence is needed to either support or rule out a role for DAF-16 in the longevity resulting from mild impairment of mitochondrial function. Our new data showing (1) increased nuclear localization of DAF-16 in mitochondrial mutant strains, (2) upregulation of DAF-16 target genes in mitochondrial mutant strains, (3) a significant overlap in gene expression changes in the mitochondrial mutants and daf-2 worms, and (4) that multiple DAF-16 interacting proteins are also required for the long life of the mitochondrial mutant strains, provides strong support for the interpretation of the data in which the lifespan increase resulting from mild impairment of mitochondrial function is DAF-16 dependent, but that other DAF-16-independent pathways also contribute to their longevity.
In comparing our results to previous studies it should be noted that these previous studies primarily involved RNAi against different mitochondrial genes than those studied here using genetic mutations. In addition to the fact that different genes were targeted, it is also important to note that the mechanism of lifespan extension has been shown to be different between genetic mutations affecting the mitochondria and RNAi against genes encoding mitochondrial proteins [6]. Thus, additional studies would need to be performed to make firm conclusions about the DAF-16 dependency resulting from RNAi targeting mitochondrial proteins.
Similarly, while two previous studies have shown that the lifespan increase resulting from treatment with low concentrations of paraquat is reduced by a daf-16 mutation [14,41], further studies were not performed to determine the role of DAF-16 in increasing lifespan. Our current results show that paraquat treatment causes the nuclear localization of DAF-16, and a daf-16-dependent upregulation of DAF-16 target genes. Similarly, we show that sod-2 deletion mutants exhibit a daf-16-dependent upregulation of DAF-16 target genes and that deletion of daf-16 abolishes the lifespan increase in sod-2 worms. Combined, these results indicate that DAF-16 is required for the enhanced longevity that results from elevated ROS.
In previous studies involving genetic mitochondrial mutants, it was found that a mutation in daf-16(m26) only mildly reduced the maximum lifespan of clk-1(e2519) worms but did not reduce mean lifespan [2] and that a daf-16(m26) mutation reduced the lifespan of isp-1 worms [5], but to a lesser extent than we observed in our study. To determine whether the magnitude of lifespan decrease caused by daf-16(mu86) mutation in our study and this previous study stems from the specific allele of daf-16 that was studied, we examined the effect of the two different daf-16 mutations, mu86 and m26, on the lifespan of isp-1 worms. The mu86 allele that we utilized is a null allele resulting from a deletion that affects all of the daf-16 isoforms, while the m26 allele is a point mutation (G to A transition) that affects a splicing site. This mutation should not affect DAF-16 isoforms a, e and g, since these isoforms do not contain the exon bearing the m26 mutation. Importantly, daf-16 isoform a has been shown to regulate longevity [30,72]. In addition, two abnormally spliced species were detected in daf-16(m26) mutants [22] and this mutant fails to completely suppress dauer formation in daf-2 mutants [73]. While both daf-16 alleles decreased the lifespan of isp-1 mutants, the magnitude of this decrease was greater for the mu86 allele than the m26 allele, indicating that the impact on longevity is related to the severity of the daf-16 mutation.
Overall, this work indicates that there is considerable overlap between different pathways of lifespan extension, which were previously thought to be distinct. Just as ROS-mediated signaling now appears to play an important role in the lifespan extension resulting from mild impairment of mitochondrial function [41,50], decreased insulin-IGF1 signaling [42], germline ablation [43], dietary restriction [54] and transient hypoxia [74], our work demonstrates that DAF-16 is required for the long lifespan resulting from mild impairment of mitochondrial function, and, combined with previous work, indicates that DAF-16 is a common downstream mediator of longevity in multiple pathways of lifespan extension.
Strains were maintained at 20°C on nematode growth media (NGM) plates seeded with OP50 bacteria. The following strains were used in these experiments. N2(WT), clk-1(qm30), isp-1(qm150), nuo-6(qm200), sod-2(ok1030), daf-2(e1370), glp-1(e2141), CF1553 muIs84[Psod-3::GFP,rol-6], daf-16(mu86), math-33(tm6724), TJ356 zIs356[Pdaf-16:daf-16a/b::GFP+rol-6(su1006)], DR26 daf-16(m26), MQ1050 isp-1(qm150);daf-16(m26). The following mutant strains were generated as previously described [75]:
JVR456 isp-1(qm150); math-33(tm6724)
JVR457 nuo-6(qm200); math-33(tm6724)
JVR176 isp-1(qm150); muIs84[Psod-3::GFP+rol-6]
JVR181 clk-1(qm30); muIs84[Psod-3::GFP+rol-6]
JVR307 daf-2 (e1370); muIs84[Psod-3::GFP+rol-6]
JVR308 glp-1 (e2141); muIs84[Psod-3::GFP+rol-6]
JVR317 nuo-6(qm200); muIs84[Psod-3::GFP+rol-6]
JVR303 sod-2(ok1030); muIs84[Psod-3::GFP+rol-6]
JVR297 clk-1 (qm30); zIs356[Pdaf-16:daf-16a/b::GFP+rol-6(su1006)]
JVR299 nuo-6(qm200); zIs356[Pdaf-16:daf-16a/b::GFP+rol-6(su1006)]
JVR302 isp-1(qm150); zIs356[Pdaf-16:daf-16a/b::GFP+rol-6(su1006)]
JVR304 daf-2(e1370); zIs356[Pdaf-16:daf-16a/b::GFP+rol-6(su1006)]
JVR316 clk-1(qm30); daf-16(mu86)
JVR327 daf-2 (e1370); daf-16(mu86)
JVR380 isp-1(qm150);daf-16(mu86)
JVR333 daf-16(mu86) I; zIs356[Pdaf-16:daf-16a/b::GFP+rol-6(su1006)]
JVR334 clk-1(qm30); daf-16(mu86);zIs356[Pdaf-16:daf-16a/b::GFP+rol-6(su1006)]
JVR335 isp-1(qm150);daf-16(mu86);zIs356[Pdaf-16:daf-16a/b::GFP+rol-6(su1006)]
We were unable to generate nuo-6(qm200);daf-16(mu86) or nuo-6(qm200);daf-16(mu86);zIs356[Pdaf-16:daf-16a/b::GFP+rol-6(su1006)] worms because of the close proximity of nuo-6 and daf-16 on chromosome I. We were also unable to generate clk-1(qm30);math-33(tm6724) double mutants. clk-1;math-33+/- worms repeatedly failed to produce any clk-1;math-33 homozygous offspring suggesting that this combination might be lethal.
mRNA was isolated from well fed, synchronized, pre-fertile young adult worms as described previously [50]. For RNAseq experiments, mRNA was collected from six biological replicates. For quantitative real-time RT-PCR experiments, mRNA was collected from at least three biological replicates. The samples used for RNAseq analysis and quantitative real-time RT-PCR were separate, independent samples.
RNA sequencing and analysis was performed as described previously [49]. Six biological replicates per strain were sequenced individually on an Illumina NextSeq 500 sequencer. Read quality was assessed using FastQC v. 0.11.5 (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and one-pass aligned to WBcel235 C. elegans genome using STAR v. 2.5.2b [76] with default parameters and “—outReadsUnmapped None”. The STAR genome index was generated with the corresponding Ensembl WBcel235 build 89 GTF annotations integrated. Transcript abundances were quantified using the “—quantMode GeneCounts” option enabled during alignment. Differential gene expression analysis was performed using the quasi-likelihood framework in edgeR package v. 3.20.1 [77] in R v. 3.4.1. Gene names, gene IDs, and predicted function annotations were downloaded using the biomaRt package v. 2.34.0.
Venn diagrams of overlapping genes were generated using the free online tool BioVenn: http://www.biovenn.nl/. Heat maps were generated using the free online tool from the Broad Institute Morpheus: (https://software.broadinstitute.org/morpheus/).
Quantitative real-time RT-PCR (qPCR) was performed by converting mRNA to cDNA using a High-Capacity cDNA Reverse Transcription kit (Life Technologies/Invitrogen) according to the manufacturer’s directions. qPCR was performed using a FastStart Universal SYBR Green kit (Roche) in an AP Biosystems RT-PCR machine [51]. The threshold was set at 1.5 as this value routinely fell within the exponential portion of the graph of relative fluorescence versus cycle number. The amount of each mRNA was calculated as the relative copy number of the gene of interest compared to the relative copy number of act-3 mRNA, then expressed as a percentage of wild-type. The following primer sequences were used for qPCR:
sod-3 (AAAGGAGCTGATGGACACTATTAAGC, AAGTTATCCAGGGAACCGAAGTC)
dod-3 (AAGTGCTCCGATTGTTACGC, ACATGAACACCGGCTCATTC)
mtl-1 (ATGGCTTGCAAGTGTGACTG, GCTTCTGCTCTGCACAATGA)
sodh-1 GAAGGAGCTGGAAGTGTTGTTC, CTCCACGTATAGTGAGGTACTCCTG)
ftn-1 (GAGTGGGGAACTGTCCTTGA, CGAATGTACCTGCTCTTCCA)
acs-17 (GGAGACTATCACTGGAGAAGCTATG, GAACTGCTTCGTCTCCAAGAGTAG)
gpd-2 (CTCCATCGACTACATGGTCTACTTG, AGCTGGGTCTCTTGAGTTGTAGAC)
icl-1 (TGTGAAGCCGAGGACTACCT, TCTCCGATCCAAGCTGATCT).
Nuclear localization of DAF-16 was visualized using a reporter strain in which daf-16 is fused to GFP: TJ356 zIs356 [Pdaf-16::daf-16a/b:GFP+rol-6(su1006)] [24]. Worms were immobilized using levamisole on an unseeded NGM plate prior to visualization using a Nikon SMZ1500 fluorescence dissecting microscope. In the clk-1 and isp-1 strains, we performed experiments in a daf-16 mutant background so that the nuclear localization of the endogenous DAF-16 would not interfere with the nuclear localization of DAF-16:GFP, as others had done previously [35].
A Psod-3::GFP reporter strain was used to monitor the activation of DAF-16 target genes [34]. Quantification of reporter activity was performed as described previously [28].
Lifespan assays were completed at 20°C except where noted. Lifespan was measured on NGM plates containing 25μM FUdR except where noted. This concentration inhibits the development of progeny after the first transfer and has minimal effect on longevity compared to NGM plates with no FUdR [78]. Lifespans involved zIs356 worms were performed on plates containing 100 μM FUdR as these worms do not show extended longevity on NGM plates with no FUdR. For each lifespan assay, three biological replicates of at least 50 worms per replicate were performed. Each replicate was performed by a different experimenter. All of the experimenters were blinded to the genotype and or the treatment (RNAi knockdown) of the samples being tested. Worms that crawled off the plate, had internal hatching of progeny or expulsion of internal organs were censored.
RNAi was performed on NGM plates containing 1 μg/ml IPTG and 50 μg/ml carbenicillin. A single colony of RNAi bacteria from a freshly streaked LB-Tet-Amp plate was grown for 10–12 hours in LB media containing 50 μg/ml carbenicillin. Bacteria were concentrated 5X prior to seeding plates. Bacteria were allowed to dry and grow for 2 days before introducing worms. Three different paradigms were used to administer RNAi depending on the effect of the gene knockdown on fertility and development: RNAi treatment was begun either at the parental L4 generation, at the egg stage or at the L4 stage of the experimental animals. The paradigm utilized was chosen to allow worms to reproduce and to develop to adulthood. The majority of genes targeted by RNAi were knocked down by the standard L4 parental paradigm (daf-16 (R13H8.1), pqm-1 (F40F8.7), cst-1/2 (F14H12.4 and C24A8.4 both targeted by F14H12.4), and math-33 (H19N07.2). imb-2 and bar-1 knockdown using the L4 parental paradigm resulted in larval arrest, sterility, and/or proclivity to extrude the germline (resulting in high levels of censoring). To circumvent these phenotypes, yet allow strongest RNAi knockdown and assessment of lifespan, the following paradigms were implemented. For imb-2 (R06A4.4), RNAi treatment was initiated at the egg stage for the experimental animals. bar-1 (C54D1.6) knockdown was performed by directly plating L4 worms to RNAi plates.
The levels of ROS were measured using the ROS-sensitive dye dihydroethidium (DHE). Approximately 100 day 1 adult worms were picked into a 1.5 ml microcentrifuge tube and washed 3 times in PBS. On the final wash, the level of PBS was reduced to 100 μl, and 100 μl of 30μM DHE was added. Worms were incubated for 1 hour on a shaker at room temperature, washed 3 times in PBS, mounted on a 1.5% agarose pad and immobilized with 5mM levamisole. 30–40 worms were imaged for 3 biological replicates under the 40x objective of an upright Leica compound fluorescence microscope (DM5500B). Fluorescence intensity of ethidium labeled ROS was quantified in the anterior pharynx using a ROI (region of interest) method and ImageJ.
Statistical significance for lifespan assays was assessed using the log-rank test to compare survival curves using GraphPad Prism Version 5. For comparisons involving more than two genotypes or treatments, we used ANOVA to test the significance of differences followed by a Bonferroni posttest to identify specific differences. Error bars indicate standard error of the mean. To determine which genes from the RNAseq data are significantly modulated, we used the quasi-likelihood framework in edgeR package v. 3.20.1 [77] in R v. 3.4.1. To assess the significance of overlapping genes between two strains we used hypergeometric tests, via R, to determine if the number of overlapping up-regulated or down-regulated genes between the two strains was significantly greater than what would be expected by chance. A Benjamini-Hochberg false-discovery rate (FDR) adjustment was used to maintain a 5% FDR after multiple testing.
|
10.1371/journal.pgen.1003888 | tRNA Methyltransferase Homolog Gene TRMT10A Mutation in Young Onset Diabetes and Primary Microcephaly in Humans | We describe a new syndrome of young onset diabetes, short stature and microcephaly with intellectual disability in a large consanguineous family with three affected children. Linkage analysis and whole exome sequencing were used to identify the causal nonsense mutation, which changed an arginine codon into a stop at position 127 of the tRNA methyltransferase homolog gene TRMT10A (also called RG9MTD2). TRMT10A mRNA and protein were absent in lymphoblasts from the affected siblings. TRMT10A is ubiquitously expressed but enriched in brain and pancreatic islets, consistent with the tissues affected in this syndrome. In situ hybridization studies showed that TRMT10A is expressed in human embryonic and fetal brain. TRMT10A is the mammalian ortholog of S. cerevisiae TRM10, previously shown to catalyze the methylation of guanine 9 (m1G9) in several tRNAs. Consistent with this putative function, in silico topology prediction indicated that TRMT10A has predominant nuclear localization, which we experimentally confirmed by immunofluorescence and confocal microscopy. TRMT10A localizes to the nucleolus of β- and non-β-cells, where tRNA modifications occur. TRMT10A silencing induces rat and human β-cell apoptosis. Taken together, we propose that TRMT10A deficiency negatively affects β-cell mass and the pool of neurons in the developing brain. This is the first study describing the impact of TRMT10A deficiency in mammals, highlighting a role in the pathogenesis of microcephaly and early onset diabetes. In light of the recent report that the type 2 diabetes candidate gene CDKAL1 is a tRNA methylthiotransferase, the findings in this family suggest broader relevance of tRNA methyltransferases in the pathogenesis of type 2 diabetes.
| The inherited predisposition to type 2 diabetes is attributed to common variants in over 60 loci. Among these risk variants is CDKAL1, which has recently been shown to be a tRNA modifying enzyme (methylthiotransferase). Genetic variants of different severity can generate a spectrum of monogenic and polygenic forms of diabetes. Here we describe a new syndrome of young onset diabetes, short stature and microcephaly (small brain size) with intellectual disability in a large consanguineous family. By linkage analysis and whole exome sequencing we identified a nonsense mutation in TRMT10A, a gene that has hitherto not been studied in mammals. The yeast homolog TRM10 has been shown to be a tRNA modifying enzyme with methyltransferase activity. We demonstrate that TRMT10A mRNA and protein are absent in cells from the affected siblings. TRMT10A localizes to the nucleolus, where tRNA modifications occur. TRMT10A silencing induces cell death in insulin-producing pancreatic β-cells, suggesting that TRMT10A deficiency may reduce β-cell mass and the pool of neurons in the brain. This is the first study describing the impact of TRMT10A deficiency in man. Our findings may have broader relevance for the understanding of the pathogenesis of type 2 diabetes and microcephaly.
| Type 2 diabetes (T2D) is a heterogeneous polygenic disease with dramatically increasing worldwide incidence as a consequence of the obesity epidemic [1]. Environmental factors (energy dense diets rich in saturated fat and sedentary lifestyle [2], [3]) and genetic predisposition contribute to its pathogenesis. T2D develops when β-cells fail to compensate for peripheral insulin resistance by increasing insulin secretion [4], [5] as a consequence of β-cell dysfunction and reduced β-cell mass. Genome-wide association studies have identified a number of loci where genetic polymorphisms associate with T2D [6]. Inherited mutations in genes at some of these loci have been shown to cause monogenic forms of diabetes, indicating that genetic variants of different severity can generate a spectrum of monogenic and polygenic forms of diabetes [7]. An example of a T2D risk gene is CDK5 regulatory associated protein 1-like 1 (CDKAL1). Polymorphisms in this gene have been associated with T2D across ethnic populations [8]. CDKAL1 encodes a transfer RNA (tRNA) methylthiotransferase that catalyzes the methylthiolation of tRNALys(UUU) [9]. Cdkal1-deficient β-cells have impaired glucose-induced insulin secretion, and Cdkal1 knockout mice develop glucose intolerance due to aberrant insulin synthesis [9].
tRNAs undergo modifications of their bases or sugar moieties that are crucial for proper cellular function. Mammalian cells have an average of 13–14 modifications per tRNA [10]–[12], methylation being the most common one [12]. Chemical modifications of nucleotides surrounding anticodons of tRNAs are important to preserve translational efficiency and fidelity [13], modifications in the main body of the tRNA affect its folding and stability, and other modifications at various positions influence tRNA identity [14], [15].
Here we identified a nonsense mutation in TRMT10A (also called RG9MTD2) in a new syndrome of young onset diabetes and microcephaly. The TRMT10A yeast ortholog YOL093w codes for the protein TRM10 that has tRNA methyltransferase activity. TRM10 specifically methylates tRNA-Arg, -Asn, -Gln, -Thr, -Trp, -Met and -Lys at position 9 (m1G9), using S-adenosylmethionine (SAM) as methyl donor [16]. TRM10 was shown to be the major if not the only m1G9 methyltransferase in yeast, but its knockout did not alter cell survival or growth [16]. Mutational analysis in yeast revealed potential interactions between TRM10, TRM8/TRM82, and TRM1 [17]. These latter proteins have tRNA methyltransferase activity towards m7G46 and m22G26, respectively [12]. The concomitant deletion of TRM10 with TRM8, TRM82 or TRM1 induced growth arrest in S. cerevisiae exposed to high temperature, suggesting enhanced tRNA instability [17].
Here we describe the affected siblings and the identification of the TRMT10A mutation. We followed this up with studies of TRMT10A expression in tissues and subcellular localization, and interrogated the functional consequences of TRMT10A deficiency.
The proband was born to consanguineous parents of Moroccan origin, her paternal and maternal grandmothers being sisters (Figure 1). Head circumference, weight and length at birth are unknown. At age 26 years she had short stature (143 cm), microcephaly (adult head circumference 49 cm, -5SD) and intellectual disability, with a history of petit mal epilepsy in adolescence. Magnetic resonance imaging of the head showed a small brain with no malformation or other abnormality (Figure 1). She had developed diabetes at the age of 22 years. At diagnosis her body mass index (BMI) was 26.9 kg/m2; plasma glucose was 176 mg/dl and HbA1c 11.3%. Other features were a short neck, wide nose, low hairline, buffalo hump, retraction of the right 5th toe, scoliosis, and joint laxity. She also had osteoporosis, with dual-energy X-ray absorptiometry T-scores of -2.7 and -3.5 at the lumbar spine and femoral neck, respectively. A skeletal survey revealed no epiphyseal dysplasia or other bone abnormality (e.g. normal X-ray of the hands, Figure 1). Her sister had short stature (154 cm), microcephaly (adult head circumference 51 cm, -3SD) and intellectual disability (IQ 69). She developed diabetes at the age of 19 years, presenting with a fasting glucose of 365 mg/dl and HbA1c 13.2%. Her BMI was 21.7 kg/m2. A younger brother had short stature (141 cm at age 14 years and final height of 157 cm at 21 years), microcephaly (head circumference 51 cm, -3SD) and mental retardation (IQ 52). His head circumference at birth was reportedly normal (36 cm). He was diagnosed with diabetes at 14 years of age, with a plasma glucose of 251 mg/dl and HbA1c 11.1%. His BMI was 20.6 kg/m2. None of the patients had ketoacidosis and all three were treated with insulin at diagnosis. They were negative for anti-insulin, anti-GAD65, anti-IA2 and islet cell autoantibodies and had a HLA genotype that did not confer risk for type 1 diabetes. Endogenous insulin secretion persisted, shown by C-peptide measurements for up to 20 years of follow-up. The insulin requirements were moderate with an average insulin dose of 0.4–1.2 U/kg/day; glycemic control ranged from good to insufficient (HbA1c 6.5–8.5%). After 18 years of diabetes, the proband's ophthalmologic examination revealed bilateral diabetic retinopathy and cortical cataract. The parents and non-affected siblings had normal size (parents 166 and 157 cm, siblings 160, 175, 183 and 159 cm) and head circumference (both parents 58 cm, P97). The parents developed diabetes at age 58 years (BMI 30.9 and 31.6 kg/m2, plasma glucose 124 and 169 mg/dl and HbA1c 8.3 and 7.6% in the mother and father, respectively) and were treated with metformin and a sulphonylurea. One grandfather and two aunts had adult onset diabetes (Figure 1). One sister had gestational diabetes at the age of 22 years; her fasting plasma glucose was normal (90 mg/dl) at age 30 (Figure 1).
The GeneChip SNP array analysis identified only one large (>3 cM) homozygous genomic region that was common to the three affected siblings. It was located on chromosome 4q22-23 and spanned 12.4 Mb between heterozygous SNPs rs4128340 and rs10516462. In this segment, we genotyped 15 microsatellite markers, which confirmed homozygosity and biparental inheritance of a haplotype shared by both parents (Figure S1). The multipoint LOD score was 3.0. Microsatellite analysis in the unaffected sister with a history of gestational diabetes (Figure 1) showed inheritance of the non-mutated maternal haplotype and of the mutated paternal haplotype. In an additional unaffected brother with normal fasting plasma glucose (84 mg/dl) and HbA1c (5.1%) at age 23 years, we observed a critical meiotic recombination event, resulting in homozygosity for all disease-associated markers except those distal to microsatellite D4S1628. This recombinant chromosome reduced the critical linkage region to a 3.1 Mb segment at 4q23.
We initially sequenced the following genes located in the 3.1 Mb segment and considered as candidates: H2AFZ (H2A histone family, member Z), LAMTOR3 (late endosomal/lysosomal adaptor, MAPK and MTOR activator 3), DDIT4L (DNA-damage-inducible transcript 4-like), RAP1GDS1 (RAP1, GTP-GDP dissociation stimulator 1) and METAP1 (methionyl aminopeptidase 1), but no mutation was identified.
Exonic sequences-enriched DNA (whole exome) sequencing was performed in one proband and results were analyzed for variants that were not found in: dbSNP135 database, the Thousand Genomes database, the Exome Variant Server, or in-house exome sequencing on 51 individuals. There was only a single candidate mutation in the 3.1 Mb critical linkage segment, a homozygous G to A transition in exon 4 of gene TRMT10A (tRNA methyltransferase 10 homolog A (S. cerevisiae) at position 379 of the coding DNA sequence, predicted to replace an Arginine residue with a premature termination codon at position 127 of the polypeptide (c.379 G>A; p.Arg127Stop). Sanger sequencing confirmed the mutation (Figure 2A), which was homozygous in the three affected patients and heterozygous in both parents as well as in the unaffected brother with the critical recombination event. A comparison across species shows that Arg127 and the surrounding region are highly conserved (Figure S2). Outside the linkage region, exome analysis in the proband identified biallelic, potentially damaging mutations in the six following genes: BCLAF1; CES1; EVC2; PTPN22; ST13; ZNF626. As none were concordant in the three affected siblings, we rejected them as candidate mutations.
We sequenced the 8 exons and flanking intronic sequences in 20 patients with a similar phenotype of young onset diabetes associated to intellectual disability, microcephaly, epilepsy, developmental delay and/or short stature, five of whom were born to consanguineous parents, but failed to identify another patient with biallelic disease-causing mutations. We furthermore sequenced TRMT10A in 26 patients with non-autoimmune diabetes with onset before 25 years and a positive family history of diabetes, in whom no mutation was identified in known MODY-associated genes, but did not identify any mutation in TRMT10A.
To examine the outcome of the TRMT10A nonsense mutation on TRMT10A protein and mRNA expression, we performed Western blot and real-time PCR on lymphoblasts from two patients, a heterozygous carrier of the mutation, and three healthy controls. TRMT10A protein was absent in lymphoblasts from patients homozygous for the Arg127Stop mutation (Figure 2B). TRMT10A mRNA expression was much reduced in patients, and intermediate in the carrier (Figure 2C). This finding is consistent with nonsense-mediated mRNA decay induced by the premature translation-termination codon (PTC) and/or by PTC-induced transcriptional silencing of the affected gene, a mechanism known to prevent the synthesis of potentially deleterious truncated proteins [18], [19].
We next evaluated TRMT10A transcript and protein expression in rat tissues. TRMT10A was ubiquitously expressed with similar mRNA abundance in liver, kidney, spleen, lung, fat, and brain. Heart and muscle showed lesser TRMT10A mRNA expression, while pancreatic islets were enriched in TRMT10A transcripts (Figure 3A). TRMT10A protein was ubiquitously present and 2- to 3-fold more abundant in brain and pancreatic islets compared to other tissues (Figure 3B–C).
In situ hybridization studies were performed in human embryonic brain samples at 8, 11, 17 and 19 gestational weeks (GW). TRMT10A was expressed throughout the whole thickness of the dorsal telencephalon (presumptive cerebral cortex) at 8 and 11 GW, with higher expression in the ventricular zone and marginal zone (Figure 4). The ventricular zone contains most neural progenitors at early stages of corticogenesis, while the marginal zone is the region where the first post-mitotic neurons migrate. At later stages TRMT10A expression was not detected in the dorsal telencephalon but was found in the cerebellar cortex and cerebellar nuclei (Figure S3 and data not shown).
To examine TRMT10A subcellular localization we first performed in silico TRMT10A topology prediction using PSORII and WoLF PSORT [20]. These softwares detected monopartite and bipartite nuclear localization signals in the first 89 amino acids of the protein. This was confirmed with cNLS Mapper [21], [22] suggesting predominant nuclear localization. To experimentally demonstrate the TRMT10A subcellular localization we took two approaches: 1) Expression of a fluorescent recombinant fusion protein, human TRMT10A (hTRMT10A)-humanized Renilla green fluorescent protein (hrGFP); 2) Detection of endogenous TRMT10A by immunofluorescence. Confocal analysis of clonal rat INS-1E β-cells expressing the TRMT10A-hrGFP fusion protein showed nuclear fluorescence with intense signal accumulation in nuclear regions of low Hoechst 33342 staining (Figure 5A). Cells expressing hrGFP alone showed homogeneous cytosolic and nuclear fluorescence. The identity of the recombinant fusion protein expressed in these cells was confirmed by Western blot (Figure 5B) using an antibody raised against purified recombinant hTRMT10A. Similar results were obtained in dispersed rat and human islet cells expressing the recombinant fusion protein (Figure S4). To identify the nuclear compartment enriched in TRMT10A, we performed immunofluorescence in rat and human islet cells using antibodies against hTRMT10A and fibrillarin, a nucleolar marker [23]. Immunostaining of endogenous TRMT10A (Figure 6, red) mimicked the fluorescence profile of recombinant TRMT10A-hrGFP. Fibrillarin immunolabeling showed a similar punctuate nuclear pattern (Figure 6, green). TRMT10A and fibrillarin images were superimposable (Figure 6, merge) indicating that TRMT10A expression is enriched in the nucleolus.
RNA interference technology was used to knock down TRMT10A in β-cells. TRMT10A mRNA and protein expression was reduced by 50% in INS-1E cells (Figure S5). TRMT10A silencing did not modify glucose-induced insulin secretion and insulin content (Figure S6), but enhanced total protein biosynthesis by 25% in clonal rat β-cells (Figure 7).
We next examined whether TRMT10A silencing affects β-cell survival. TRMT10A knockdown induced apoptosis in clonal and primary rat β-cells and dispersed human islets (Figure 8). TRMT10A deficiency further sensitized rat β-cells to oleate-, palmitate- and ER stress-induced apoptosis (Figure 8A–D). These results were confirmed by Western blot for cleaved caspase-3, showing increased caspase-3 activation basally and after palmitate and cyclopiazonic acid exposure (Figure 8E). High glucose-induced β-cell apoptosis was also increased by TRMT10A silencing (Figure 8A). We observed that TRMT10A expression in β-cells is modulated by ER stress. Exposure of rat or human β-cells to the saturated FFA palmitate, previously shown to induce ER stress [3], [24], [25], or to chemical ER stressors enhanced TRMT10A expression (Figure S7) to an extent that was correlated with the intensity of ER stress (measured by the expression of ER stress markers, Figure S8). TRMT10A expression was induced in a PERK- but not IRE1-dependent manner (Figure S9). TRMT10A silencing did not induce expression of the ER stress markers BiP, XBP-1s, ATF3 and CHOP (data not shown).
In a large consanguineous family of Moroccan origin, we identified a new syndrome of severe insulinopenic young onset diabetes and microcephaly with intellectual disability. We used linkage analysis and whole exome sequencing to identify the causal mutation. We found only one region of homozygosity by descent shared by the three affected patients, and only one potentially damaging rare genetic variant in this region, located in the TRMT10A gene, changing an arginine codon at position 127 of the protein into a stop codon (Arg127Stop). In the rest of the patients' exome, we found no potentially damaging, rare biallelic variants shared by the three patients that might have qualified for a causal mutation.
Among the family members, four were heterozygous carriers of a mutant allele. Of these, the parents developed diabetes in their fifties, one sister had gestational diabetes, and one brother had normal plasma glucose levels at the age of 23 (Figure 1). Other family members were not available for testing. It is possible that TRMT10A haploinsufficiency increases the risk for adult onset diabetes.
TRMT10A contains 8 exons, the first exon being non-protein coding. The mutated codon 127 is in exon 4. The protein environment of Arg127 is extremely conserved across species. Little is known about the role of TRMT10A in mammals. A single study suggested altered TRMT10A mRNA expression in colorectal cancer [26]. Blast analysis indicated that TRMT10A is the mammalian ortholog of S. cerevisiae TRM10, previously shown to be involved in guanine 9 tRNA methylation m1G9 [16]. TRMT10A has seven transcripts in the Vega database. Two of them are non-protein coding due to a retained intron, three contain 8 exons coding for identical proteins of 339 amino acids, and differ only in their untranslated regions. InterProScan analysis indicates that these three proteins have a tRNA (guanine 9-N1) methyltransferase domain as well as tRNA (guanine-N1) methyltransferase domain, both of them present in TRM10. The last two TRMT10A transcripts contain only 6 exons and code for shorter proteins of 200 and 206 amino acids. These two variants are truncated at the C-terminus and only have the tRNA (guanine-N1) methyltransferase domain. In rat only one isoform of TRMT10A containing both domains is found. Based on these analyses, we suggest that TRMT10A functions as a tRNA-modifying enzyme, but this remains to be experimentally confirmed.
The Arg127Stop mutation is predicted to block the expression of the five coding human TRMT10A isoforms. The nonsense mutation abolished TRMT10A protein expression, and also significantly reduced its mRNA expression (Figure 2), probably by nonsense-mediated decay and/or transcriptional silencing [18], [19]. We show that TRMT10A is ubiquitously expressed but enriched in brain and pancreatic islets (Figure 3), consistent with the tissues affected in this new syndrome of diabetes and microcephaly. In silico topology prediction indicates that the five human TRMT10A isoforms, as well as the rat enzyme, have predominant nuclear localization. This was confirmed by immunofluorescence and confocal microscopy, with TRMT10A mainly localizing in the nucleolus of β- and non-β-cells (Figure 5–6 and S4). tRNA transcription and early processing occurs in several subcellular compartments including the nucleus, cytoplasm and cytoplasmic surface of the mitochondria [14]. tRNA genes are recruited to the nucleolus for transcription [27], 5′ leader sequence removal and 3′ end modification, removal of the 3′ trailer and addition of the CCA, which is required for efficient tRNA nuclear export [28]. Mature tRNAs are exported to the cytosol for aminoacylation and function in translation. This transport is not unidirectional; cytosolic tRNAs can follow a retrograde transport to the nucleus (e.g. during nutrient deprivation), to be re-exported to the cytosol following nutrient availability [14]. Some tRNA modifications occur on initial tRNA transcripts, while others are introduced in end-matured tRNAs [29]. Since tRNA transcription and maturation occurs in the nucleus it is expected that the enzymes catalyzing these modifications have a nuclear localization. Studies in yeast confirmed that a subset of tRNA methyltransferases is located in the nucleus [28], [30], [31], with distinct subnuclear distribution, i.e. nucleolus, nucleoplasm, or inner nuclear membrane; the reason for these different localizations is not known [14], [31]. The predominant nucleolar localization of TRMT10A is consistent with its proposed tRNA modifying activity.
Alterations in tRNA modification are expected to affect protein translation. We showed that TRMT10A knockdown in rat β-cells enhances total protein biosynthesis (Figure 7). TRMT10A silencing does not impair glucose-induced insulin secretion or content in β-cells (Figure S6), suggesting that TRMT10A deficiency has no major impact on β-cell function. TRMT10A knockdown sensitizes β-cells to apoptosis in control condition and after exposure to FFAs, high glucose or synthetic ER stressors (Figure 8), conditions related to T2D. It has been proposed that mammalian cytosolic and mitochondrial tRNAs prevent apoptosis by blocking the binding of cytochrome c to Apaf-1, thus preventing the formation of the apoptosome [32], [33]. It is not known whether tRNA modifications affect this tRNA-cytochrome c interaction.
Primary microcephaly refers to a congenitally small but otherwise normally structured brain, with a head circumference later in life that remains 3 SD below the mean for age and gender. Primary microcephaly is a very rare disorder affecting approximately 1/100,000 live births, mainly inherited as an autosomal recessive trait, and is associated with a high rate of parental consanguinity [34]. Microcephaly and young onset diabetes co-segregate in the present family, as both features were present in the three affected siblings and absent in the six unaffected siblings, defining a new syndrome. Our linkage analysis identified a single region where all affected siblings were homozygous over a significant length of genomic DNA. It is hence likely that the whole phenotype results from pleiotropic effects of a single mutated gene.
Microcephaly in our patients was associated with intellectual disability and no other neurological feature, except for a history of petit mal seizures in the proband. This clinical presentation fits with the phenotype of primary microcephaly [35]. Primary microcephaly is vastly heterogeneous, and many genes that cause primary microcephaly play a role in mitotic spindle organization and/or DNA repair, presumably affecting the proliferation of neural progenitors and the generation of an adequate pool of neurons in the developing brain [36].
The expression pattern of TRMT10A in the ventricular zone of the developing cortex is consistent with its influence on neural progenitor properties, including control of survival that is known to affect brain size. In addition it may act in subsets of differentiated neurons, as suggested by its expression in cortical marginal zone and cerebellum.
Early onset diabetes has been associated with microcephaly in other genetic disorders. Homozygous mutations in the IER3IP1 gene encoding the immediate and early response 3 interacting protein 1 result in infantile diabetes and congenital microcephaly with simplified gyration, hypotonia, intractable seizures, and early death [37], [38]. Cases of microcephaly with severe neurological expression were also described in Wolcott-Rallison syndrome, which includes permanent neonatal diabetes, multiple epiphyseal dysplasia, osteoporosis and liver dysfunction. This syndrome is due to biallelic mutations in EIF2AK3 encoding translation initiation factor 2-α kinase-3 [39]. EIF2AK3 is activated upon the accumulation of unfolded proteins in the ER and inhibits protein translation initiation [40].
Other human diseases are caused by mutations in genes encoding tRNAs and tRNA modifying enzymes. Pontocerebellar hypoplasia, characterized by hypoplasia and atrophy of ventral pons, cerebellum and the cerebral cortex, is caused by mutations in genes encoding tRNA splicing endonuclease subunits (TSEN) or mitochondrial arginyl-tRNA synthetase (RARS2) [41]. Mutations in mitochondrial tRNA genes and in aminoacyl-tRNA synthetases cause myopathies and neurodegenerative diseases, sometimes in association with diabetes. Recently, a syndrome of mental retardation, microcephaly and short stature was described, caused by mutations in NSUN2, encoding a methyltransferase that catalyzes the intron-dependent formation of 5-methylcytosine at C34 of tRNA-leu(CAA) [42], [43]. NSUN2 is the ortholog of yeast TRM4. Wild-type NSUN2 localized to the nucleolus, whereas mutant NSUN2 accumulated in the nucleoplasm and cytoplasm [42]; other NSUN2 mutations resulted in nonsense-mediated mRNA decay [43]. Inactivation of the X-linked gene FTSJ1, another RNA methyltransferase and ortholog of yeast TRM7, gives rise to non-syndromic intellectual disability [44].
In addition to causing microcephaly and short stature, the TRMT10A mutation causes a severe form of diabetes, which was not reported for these other RNA methyltransferase mutations. This may be related to cell-specific requirements of RNA modifications. It is of particular interest that CDKAL1 polymorphisms predispose to insulin secretion defects and T2D [8]. CDKAL1 was recently shown to methylthiolate tRNALys(UUU) [45]. The β-cell-specific Cdkal1 knockout mouse develops impaired glucose tolerance, due to misreading of Lys codons in proinsulin, defective insulin biosynthesis and increased susceptibility to ER stress and high fat diet [9].
In conclusion, we describe a nonsense mutation in the TRMT10A gene in a new syndrome of young onset diabetes and microcephaly. Based on its cellular localization and by homology with its yeast counterpart, we propose that TRMT10A has methyltransferase activity. We show that TRMT10A is expressed in human fetal brain; TRMT10A silencing does not impair β-cell function but induces apoptosis, suggesting that TRMT10A deficiency may negatively affect β-cell mass and the pool of neurons in the developing brain. Our findings may have broader relevance for the understanding of the pathogenesis of T2D and microcephaly.
The ethics committee of the Erasmus Hospital, Université Libre de Bruxelles approved of the study. The three patients, their parents, and two unaffected siblings gave informed consent. Human fetal brain was collected and used according to the guidelines of the local ethics committees on research involving human subjects (Erasmus Hospital, Université Libre de Bruxelles and Belgian National Fund for Scientific Research). Adult male Wistar rats were housed and used following the rules of the Belgian Regulations for Animal Care, with approval of the ethics committee of the Université Libre de Bruxelles.
A peripheral blood sample was obtained for genetic analysis from the three patients, their parents, and two unaffected siblings. Leukocyte DNA was extracted using proteinase K digestion followed by phenol-chloroform extraction and ethanol precipitation [46] and samples were stored at 4°C in T10E1 buffer. We used Affymetrix 11K-GeneChip microarrays representing 10,000 autosomal single nucleotide polymorphisms (Affymetrix, High Wycombe, United Kingdom) to genotype the three patients' DNA (500 ng each) on an Affymetrix platform following the instructions of the manufacturer. Regions of homozygosity were delineated using the ExcludeAR algorithm [47]. In chromosomal regions with apparent homozygosity by descent, microsatellites were genotyped in individual subjects. Marker order was obtained from the University of California at Santa Cruz (UCSC) physical map (http://genome.ucsc.edu/cgi-bin/hgGateway). A multipoint LOD score was computed using the MAPMAKER/HOMOZ software [48] assuming a gene frequency of 0.005 and marker allele frequencies as observed in a series of control subjects, with a minimal minor allele frequency of 0.10.
Genomic DNA from the proband (Figure 1, arrow) was sonicated and enriched for exonic sequences by hybridization on an Agilent SureSelect All Exon v1 capture kit. Exon-enriched DNA was paired-end sequenced over 90 bp by an Illumina HiSeq2000 sequencer (Beijing Genomics Institute). An average of 55.6 million paired-end reads were filtered to eliminate reads with more than 6 undetermined nucleotides or 40 identical bases in tandem. The filtered reads were then aligned to the human genome GRCh36 assembly using the SOAPaligner 2.20 software [49] and the genotypes were called using the SOAPsnp program [50]. Resulting single nucleotide variants (SNVs) were filtered according to the following rules: base quality larger than 20, read depth equal to or larger than 4, and a distance between two variants larger than 4. Insertions and deletions were identified separately, through alignment to GRCh36 using the Burrows-Wheeler alignment tool [51], and detection using the Genome Analysis Toolkit [52]. SNVs and indels were annotated using the Ensembl V54 database. We considered SNVs and indels that were not found in the dbSNP135 database, nor in the Thousand Genome (www.1000genomes.org) database, nor in the Exome Variant Server (http://evs.gs.washington.edu/EVS/), and that were not found in our other in-house exome sequencing results.
PCR primers for all exons and flanking intronic sequences were designed using the Exonprimer software (http://ihg.helmholtz-muenchen.de/ihg/ExonPrimer.html). All exons and flanking intronic regions of the candidate genes were sequenced by the Sanger method using the Big Dye Terminator cycle sequencing kit v2 (Applied Biosystems, Foster City, California, USA), and analyzed on a 3130 Genetic Analyser sequencing machine (Applied Biosystems). Sequences were analyzed in silico for mutations using the SeqScape software V.2.0. (Applied Biosystems).
In situ hybridization was done on human fetal brain (GW 8,11, 17, 19) as previously described [53]. Riboprobe template was generated by PCR using TRMT10A specific pairs of primers: F: CCAAGCTAATACGACTCACTATAGGGAGATGTGAACCAATATCTAAACGACAAA – R: GGATCCATTAACCCTCACTAAAGGGAGAGATTTTCCTTATCCTGCTTTTCTTC.
Clonal rat INS-1E cells (a kind gift from Dr C Wollheim, Centre Médical Universitaire, Geneva, Switzerland) were cultured in RPMI medium as previously described [54], [55]. Tissues were obtained from adult male Wistar rats (Charles River Laboratories). Rat islets were isolated by collagenase digestion followed by hand picking under a stereomicroscope. Islets were dispersed and β-cells purified by autofluorescence-activated cell sorting (FACS, FACSAria, BD Bioscience) and cultured as described [56], [57]. Human islets from non-diabetic organ donors (n = 13, age 68±4 years, BMI 27±1 kg/m2) were isolated by collagenase digestion and density gradient purification [58]. The islets were cultured, dispersed and transfected as previously described [59]. The mean percentage of β-cells of the human islet preparations was 50±5%, as determined by insulin immunofluorescence [25], [60]. Human lymphoblasts from three control individuals, two patients and one heterozygous carrier of the mutation were cultured in RPMI 1640 medium supplemented with 20% FBS, 100 mU/ml penicillin and 100 mU/ml streptomycin.
hTRMT10A was amplified by PCR from lymphoblast cDNA using oligonucleotides spanning the start and stop codons of the TRMT10A open reading frame (ORF), using primers F CGGAATTCATGTCATCTGAAATGTTGCC and R CGCTCGAGGTGTGGCAGAGAGTTCACTG. The restriction sites EcoRI and XhoI (underlined) were added to facilitate the directional cloning into the expression vector pGEX-6P-1 (GE Healthcare). This vector allows the expression of recombinant proteins fused to glutathione-s-transferase (GST) at its N-terminus. E. coli BL21 cells were transformed with the pGEX-6P-1-TRMT10A plasmid by electroporation. Positive clones were selected by colony PCR and sequenced. For recombinant protein expression, a single colony was grown overnight at 37°C in LB medium containing 100 µg/ml ampicillin. Cells were then diluted 1∶50 in the same medium and grown at 37°C until an optical density of 0.6 at 600 nm was reached. Isopropyl-β-D-thiogalactoside (0.25 mM) was then added and cells were grown at 28°C for 3 h to induce recombinant protein expression. Cells were harvested by centrifugation at 3000×g for 10 min, lysed by sonication in 20 mM Tris buffer pH 8 containing 0.5% Triton ×100, 10 mM dithiothreitol, 0.1 mM PMSF and protease inhibitor cocktail (Roche), and centrifuged for 15 min at 15,000×g at 4°C. The supernatant was applied to 1 ml glutathione spin columns (Pierce) and washed with ice-cold lysis buffer. The recombinant hTRMT10A was separated from the GST moiety by in column site-specific proteolysis using PreScission protease (GE Healthcare) following the manufacturer's instructions. The purified recombinant hTRMT10A was used for rabbit polyclonal antibody production (Eurogentec).
hTRMT10A was amplified by PCR from HeLa cDNA using the oligonucleotides F 5′-AAAAAACCCGGGAATGTCATCTGAAATGTTG-3′ (start codon is indicated in bold), and R 5′-AAAAAAGGATCCTGAGTGTGGCAGAGAGTT-3′ in which the restriction sites SmaI and BamHI (underlined) were added to facilitate the directional cloning into the mammalian expression vector Vitality phrGFP-1 (Stratagene). The stop codon of the TRMT10A ORF was removed to allow the production of the recombinant TRMT10A fused to the N-terminus of hrGFP. The PCR product was purified using the Wizard SV Gel and PCR clean-up system (Promega), sequentially digested with SmaI and BamHI (New England Biolabs), and cloned into the Vitality phrGFP-1 vector digested with the same restriction enzymes. The plasmid was introduced into electrocompetent One Shot E. coli (Invitrogen) by electroporation, and positive clones were identified by colony PCR. A single colony containing hrGFP (empty vector) or TRMT10A-hrGFP plasmid was grown overnight at 37°C in LB medium with 100 µg/ml ampicillin. Plasmids were purified with PureYield Midiprep (Promega) and quantified by NanoDrop (Thermo Scientific). Expression of recombinant TRMT10A-hrGFP in rat β-cells was confirmed by Western blot.
Cells were transfected overnight with 30 nM of a control siRNA (Qiagen), or two single siRNAs targeting rat or human TRMT10A using Lipofectamine RNAiMAX (Invitrogen). siRNA-lipid complexes were formed in Opti-mem (Invitrogen) for 20 min as previously described [61].
hrGFP and TRMT10A-hrGFP plasmids were introduced by lipofection in INS-1E cells or dispersed rat and human islet cells using Lipofectamine 2000 (Invitrogen). siRNA sequences, plasmid and Lipofectamine concentrations are described in Tables S1 and S2.
TRMT10A subcellular localization was examined by expressing recombinant hTRMT10A fused to GFP, or by immunolabeling endogenous TRMT10A in INS-1E cells and dispersed rat and human islet cells. Cells plated on poly-lysine coated cover slips were transfected or not with hrGFP and TRMT10A-hrGFP plasmids, fixed with 4% formaldehyde [62], permeabilized with methanol, blocked with goat serum and incubated or not for 1 h with rabbit anti-hTRMT10A (1∶200, Eurogentec), mouse anti-human/rat fibrillarin (1∶200, EnCor Biotechnology) and mouse anti human/rat insulin (Sigma). Alexa Fluor 546 goat anti-mouse IgG (H+L), Alexa Fluor 488 goat anti-mouse IgG (H+L) and Alexa Fluor 546 goat anti-rabbit IgG (H+L) (1∶500, Molecular Probes, Invitrogen) were used as secondary antibodies. Nuclei were stained with Hoechst 33342. Slides were analyzed by inverted fluorescence microscopy (Zeiss Axiovert 200, Oberkochen, Germany). Confocal analysis was performed on a LSM510 NLO multiphoton confocal microscope fitted on an Axiovert M200 (Zeiss) [63].
Poly(A)+ mRNA was isolated from INS-1E cells, dispersed human islets, or primary rat β-cells using the Dynabeads mRNA DIRECT kit (Invitrogen). For total RNA purification, rat tissues and pancreatic islets were resuspended in RNeasy Minikit lysis buffer (Qiagen), homogenized using a T10 basic ULTRA-TURRAX disperser (IKA) or lysed by sonication in a Bioruptor NGS (Diagenode), respectively. Total RNA was purified with the RNeasy Minikit and quantified by NanoDrop. mRNA and total RNA were reverse transcribed as previously described [25], [56]. Real-time PCR was performed using Rotor-Gene SyBR Green on a Rotor-Gene Q cycler (Qiagen), or FastStart SYBR Green on the LightCycler (Roche Diagnostics) [62], [64]. Standards were prepared using suitable primers in a conventional PCR. Gene expression was calculated as copies/µl using the standard curve approach [65]. Expression values were corrected for the expression of the reference genes GAPDH, OAZ1 and/or β-actin, which were not modified by the experimental conditions. The primers are provided in Table S3.
Rat tissues and pancreatic islets were resuspended in ice-cold PBS containing protease inhibitor cocktail and homogenized as described above. Total protein was measured in the lysates using the Protein Assay Dye Reagent (BIO-RAD). INS-1E cells and human lymphoblasts were lysed with Laemmli buffer [59]. Cell lysates were resolved in 10 or 14% SDS-PAGE gels and transferred to nitrocellulose membranes. Immunoblotting was performed using antibodies against hTRMT10A, human cleaved caspase-3 (Cell Signaling), human α-tubulin (Sigma-Aldrich) or human β-actin (Cell Signaling). Protein detection was done using horseradish peroxidase-conjugated secondary antibodies and SuperSignal West Femto chemiluminescence revealing reagent (Thermo Scientific). Immunoreactive bands were detected with a ChemiDoc XRS+ system and with Image Lab software (BIO-RAD). Protein levels were corrected for α-tubulin and/or β-actin.
Insulin secretion was performed as previously described [61]. Briefly, 72 h after transfection, INS-1E cells were cultured for 1 h in RPMI without glucose, washed with modified Krebs-Ringer bicarbonate HEPES solution, and insulin secretion was induced by 30 min incubation with KRBH containing 1.67 or 16.7 mM glucose, alone or in combination with 10 µM forskolin. Insulin was measured by ELISA (Mercodia) in cell-free supernatants and acid-ethanol extracted cell lysates [61], [66], [67]. Total protein was measured in cell lysates as described above.
FFA treatment was performed in RPMI 1640 containing 0.75% FFA-free BSA (Roche). Oleate and palmitate (sodium salt, Sigma) were dissolved in 90% ethanol and diluted 1∶100 to a final concentration of 0.5 mM [25], [68]. The chemical ER stressors cyclopiazonic acid and thapsigargin (two SERCA pump blockers), tunicamycin (an inhibitor of N-glycosylation) and brefeldin-A (an inhibitor of ER-to-Golgi vesicle transport) were used at 25 µM, 1 µM, 5 µg/ml and 0.1 µg/ml, respectively. The IRE1 inhibitor 4μ8C was used at 25 µM [69] For all treatments the control condition contained the same dilution of vehicle.
Apoptotic cell death was detected and counted by fluorescence microscopy after Hoechst 33342 (5 µg/ml; Sigma-Aldrich) and propidium iodide (5 µg/ml) staining as described [25], [60], [62], [70]. Apoptosis was also examined by Western blotting for cleaved caspase-3.
72 h after transfection INS-1E cells were cultured for 2 h in Krebs-Ringer buffer containing 11 mM glucose, 1% BSA and 10 µCi/ml L-(3,4,5 3H)-leucine (Perkin Elmer). Cells were then washed with Krebs-Ringer solution containing 10 mM non-radioactive leucine. Cells were collected in ice-cold water and lysed by sonication. Total protein was precipitated with 10% trichloro-acetic acid. The content of 3H-labeled proteins was determined in a liquid scintillation analyzer (Packard) [70]. Protein biosynthesis was expressed per total protein content to correct for differences in cell number in the experimental conditions.
Data are presented as means ± SE. Non-normally distributed variables were log-transformed before statistical testing. Comparisons between groups were made by ANOVA followed by two-sided Student's paired t test with Bonferroni correction for multiple comparisons. A p value<0.05 was considered statistically significant.
|
10.1371/journal.ppat.1003290 | The Plant Actin Cytoskeleton Responds to Signals from Microbe-Associated Molecular Patterns | Plants are constantly exposed to a large and diverse array of microbes; however, most plants are immune to the majority of potential invaders and susceptible to only a small subset of pathogens. The cytoskeleton comprises a dynamic intracellular framework that responds rapidly to biotic stresses and supports numerous fundamental cellular processes including vesicle trafficking, endocytosis and the spatial distribution of organelles and protein complexes. For years, the actin cytoskeleton has been assumed to play a role in plant innate immunity against fungi and oomycetes, based largely on static images and pharmacological studies. To date, however, there is little evidence that the host-cell actin cytoskeleton participates in responses to phytopathogenic bacteria. Here, we quantified the spatiotemporal changes in host-cell cytoskeletal architecture during the immune response to pathogenic and non-pathogenic strains of Pseudomonas syringae pv. tomato DC3000. Two distinct changes to host cytoskeletal arrays were observed that correspond to distinct phases of plant-bacterial interactions i.e. the perception of microbe-associated molecular patterns (MAMPs) during pattern-triggered immunity (PTI) and perturbations by effector proteins during effector-triggered susceptibility (ETS). We demonstrate that an immediate increase in actin filament abundance is a conserved and novel component of PTI. Notably, treatment of leaves with a MAMP peptide mimic was sufficient to elicit a rapid change in actin organization in epidermal cells, and this actin response required the host-cell MAMP receptor kinase complex, including FLS2, BAK1 and BIK1. Finally, we found that actin polymerization is necessary for the increase in actin filament density and that blocking this increase with the actin-disrupting drug latrunculin B leads to enhanced susceptibility of host plants to pathogenic and non-pathogenic bacteria.
| The cytoskeleton is a dynamic platform for sensing and responding to a diverse array of biotic and abiotic stresses. The nature and timing of the changes in actin organization range from excessive bundling, to massive depolymerization, to new filament assembly, depending on the particular signal and the responding cell type. Here, we use the Arabidopsis–Pseudomonas pathosystem to dissect pathogen-derived cues that elicit changes in the plant host-cell cytoskeleton. Overall, we provide the first evidence that the actin cytoskeleton rearranges in response to a phytopathogenic bacterium and we quantified the temporal response of epidermal cells to Pseudomonas syringae pv. tomato DC3000 strains and susceptible Arabidopsis mutants, using a robust set of tools for measuring changes in actin organization. An immediate but transient increase in actin filament abundance was associated with pattern-triggered immunity. This response could be mimicked with microbe-associated molecular pattern peptide treatments. Second, we observed a late increase in actin filament bundling that appears to be part of effector-triggered susceptibility. We dissected the initial steps involved in the host-cell signaling pathway and demonstrated that FLS2, BAK1, and BIK1 were required for the actin response. Collectively, these findings demonstrate that rapid changes in host-cell cytoskeleton organization occur in response to receptor-mediated signaling during plant innate immunity.
| Actin filament arrays in plant cells undergo constant remodeling and can respond rapidly to a diverse array of extracellular stimuli. Even in unstimulated epidermal cells, architectural rearrangements occur within seconds due to myosin-dependent translocation, remarkably fast filament assembly, and the destruction of filaments by prolific severing activity [1], [2]. This incessant remodeling of the actin cytoskeleton expends an enormous amount of energy, yet the physiological relevance of this is poorly understood. The actin cytoskeleton is a major signaling target and changes dramatically in response to numerous abiotic and biotic stimuli; the responses however are quite varied, ranging from filament bundling, to massive actin depolymerization, to assembly of new filament arrays [2]–[4]. For example, cells that are gently prodded with glass or tungsten needles generate extensively bundled filament arrays directly under the site of mechanical stimulation; yet, once the stimulus is removed the bundling is abrogated [5]. This is thought to mimic the efforts of fungi and oomycetes to gain entry into plant cells and, as such, it has been commonly assumed that attempted or actual penetration is responsible for eliciting changes in the host-cell actin cytoskeleton, rather than activation of host-cell defense signaling following the recognition of ‘non-self’. Actin filament arrays undergo a markedly different response upon recognition of ‘self’ pollen grains by a flower's stigma. Poppy pollen, for example, initiates a self-incompatibility (SI) response, resulting in massive depolymerization of actin filaments within minutes of stimulus perception, effectively inhibiting pollen tube growth and blocking fertilization [6]. In contrast with this signal-mediated destruction of actin filaments, the interaction between mutualistic bacteria and plant cells generally results in the development of bright phalloidin-decorated spots in host cells [7] – suggestive of actin polymerization. Other changes to actin during mutualistic interactions have been described, including filament reorientation from longitudinal to transverse arrays and increased numbers of actin bundles at the tip of root hairs; these responses can be reproduced with the application of purified Nod-factors from mutualistic bacteria onto host plant cells [8], [9]. Furthermore, nap1 mutant root hairs, which are incapable of proper nodule formation, lack the ability to elicit changes to the actin cytoskeleton when Nod-factors are applied because these mutants are deficient for ARP2/3-dependent actin polymerization [10], [11]. On the other hand, certain signals from pathogenic fungi, like the Verticillium dahlia (VD) toxin, can stimulate dose-dependent destruction and relocation of cortical actin filaments to the perinuclear region [12]. Thus, biotic and abiotic signaling cascades produce a myriad of responses that can lead to dramatically different outcomes for actin organization and dynamics.
Plants are constantly exposed to a large number of fungal and bacterial microbes, however, most plants are immune to the majority of potential invaders due to a multilayered defense system. The initiation of plant immunity relies on structural defenses (i.e. the presence of trichomes, the closing of stomata to prevent bacterial entry, and the thickness and composition of the cell wall and cuticle) and inducible measures to guard the plant from various microbes [4]. These inducible processes can be classified in two nodes of defense signaling: pattern-triggered immunity (PTI) and effector-triggered immunity (ETI). PTI is a broad based immunity initiated through the host-cell recognition of conserved structural components, known collectively as microbe-associated molecular patterns or MAMPs, by cell-surface receptors [4], [13], [14], [15]. The recognition of microbes by the receptor kinase, FLAGELLIN-SENSING 2 (FLS2), is one of the best-studied PTI pathways in Arabidopsis. Upon perception of bacterial flagellin or the synthetic 22-amino acid peptide flg22, FLS2 associates with another receptor-like kinase, BAK1 (BRASSINOSTEROID INSENSITIVE1-ASSOCIATED KINASE 1); this association releases the cytoplasmic kinase BIK1 (BOTRYTIS-INDUCED KINASE 1) to induce down-stream defense signaling [4], [16]–[19], and ultimately prevents infection by non-adapted pathogens [4], [13], [14].
Pathogens elicit PTI in their respective host plants, and thus necessarily secrete or translocate various toxins and inject effector proteins into host cells to subvert PTI [14], [15]. For example, Gram-negative bacteria use a specialized type III secretion system (T3SS) to translocate collections of type III effector (T3E) proteins directly into host cells resulting in effector-triggered susceptibility (ETS) of the host plant [4], [13], [14]. The second node of immunity, ETI, relies on host-cell recognition and response to pathogen-specific effector proteins or their activities, to mount a defense response that is more rapid and more pronounced than PTI [4], [13], [14].
The plant actin cytoskeleton has been implicated in the generation and maintenance of many aspects of PTI. Major hallmarks of PTI in plant cells include endocytic uptake of receptors, changes to cytoplasmic streaming, activation of defense genes via mitogen-activated protein kinase (MAPK) signaling, recruitment of NADPH oxidase to the plasma membrane, an oxidative burst, directed trafficking of Golgi and endoplasmic reticulum to the site of attack, and callose deposition [3], [20]–[22]. The importance of an intact actin cytoskeleton for each of these responses has been demonstrated with pharmacological studies. The actin cytoskeleton is further assumed to play a central role in plant defense against microbes based on static images that show actin filament bundles impinging upon sites of both compatible and incompatible fungal or oomycete attack [3], [20], [21], [23]. Finally, disrupting the plant actin cytoskeleton with cytochalasin E treatment or by overexpressing actin-binding proteins allows penetration into plant cells and tissues by incompatible fungi and oomycetes [24]–[26].
Despite the growing body of evidence that suggests the involvement of the actin cytoskeleton in either PTI or ETI, no direct evidence linking specific aspects of either layer of immune signaling have been reported. However, a recent report shows that a T3E protein from Pseudomonas syringae, HopZ1a, targets the microtubule cytoskeleton to circumvent PTI [27]. Similarly, one report describes actin filament stabilization through monoubiquitination of actin during infection by either pathogenic or mutualistic bacteria, but not in response to stress or viral infection [28]. We hypothesize that the continuous rearrangements of the actin cytoskeleton in Arabidopsis epidermal cells represent a surveillance mechanism to external threats [1]; however, it is still unclear whether there are direct links between actin and PTI- or effector-mediated signal transduction cascades or which actin-binding proteins are involved. Significantly, changes in the expression of Actin Depolymerizing Factor (ADF) enhance non-host susceptibility in both fungal and bacterial pathosystems, but these changes do not alter the focal accumulation of profilin at pathogen-invasion sites or expression of defense genes [26]. Notably, the localization of Arabidopsis ADF4 to the nucleus correlates with the reduced expression of the hallmark PTI-defense gene, FRK1, which implicates the actin cytoskeleton in the early onset of PTI [29]. Additionally, this adf4 knockout mutant fails to activate ETI in response to P. syringae expressing the cognate bacterial effector gene AvrPphB [30]. Collectively, these results suggest that actin organization and dynamics are strictly regulated in both PTI and ETI. To date, the timing and nature of actin-based responses in host cells during bacterial pathogen attack have not been described.
Using a combination of bacterial mutants and advanced imaging of actin cytoskeleton organization in epidermal cells from Arabidopsis, we analyzed the host-cell response to the bacterial phytopathogen P. syringae pv. tomato DC3000. We quantified the nature of specific changes in actin array architecture over a time-course of infection with both pathogenic and non-pathogenic strains of bacteria. A transient increase in the density of actin filament arrays in the cortex of epidermal cells was identified, and we demonstrate that this change did not require either the T3SS or effector proteins. Moreover, we found that infiltration of leaves with MAMPs was sufficient to elicit an increase in actin filament density. Using reverse genetics, we have also begun to dissect the plant signaling pathways required to elicit actin rearrangement during PTI. Notably, we found that FLS2, BAK1 and BIK1 were required for the increase in actin filament density. When actin polymerization was blocked by treatment with latrunculin B, the increase in actin filament density did not occur and plants were more susceptible to infection with pathogenic and non-pathogenic bacteria. These data implicate the transient increase in cytoskeletal array density as a contributing factor during PTI and identify parts of the signal transduction machinery necessary for this response.
In this study, we focused on cytoskeletal responses in the Arabidopsis–Pseudomonas pathosystem and used seedlings expressing a well-characterized actin reporter, GFP-fABD2. Dip-inoculated cotyledons from wild-type Col-0 and transgenic plants expressing GFP-fABD2, a fusion protein between green fluorescent protein and the second actin-binding domain of Arabidopsis FIMBRIN1 [1], exhibited necrotic lesions when inoculated with pathogenic P. syringae pv. tomato DC3000 (hereafter referred to as DC3000; Figure S1A & C), whereas cotyledons inoculated with the non-pathogenic T3SS-deficient mutant hrpH did not (Figure S1B & D). Furthermore, cotyledons (Figure S1E) and rosette leaves (Figure S1F) inoculated with DC3000 had a higher bacterial load than those inoculated with hrpH at 4 days after inoculation. These results confirm that DC3000 can proliferate and cause disease symptoms on seedling cotyledons expressing GFP-fABD2, and that bacterial growth is not significantly different when bacteria are infected in cotyledons or rosette leaves.
To study the response of host-cell cytoskeleton during bacterial infection, we imaged actin filament arrays in cotyledons with spinning disk confocal microscopy (SDCM) at various time-points after dip-inoculation with DC3000 and hrpH (Figure 1). Epidermal pavement cells from Arabidopsis cotyledons, display two populations of actin filaments in the cortical cytoplasm—dynamic, faint structures that resemble single actin filaments; and, thick, bright actin filament bundles (Figure 1A & D). At 6 hours post inoculation (hpi), we observed an increase in the abundance of actin filaments in the cortical array of epidermal cells inoculated with DC3000 (Figure 1B) or hrpH (Figure 1C), compared to mock-treated material (Figure 1A). At 24 hpi, we noticed little difference between the mock control (Figure 1D) and hrpH inoculation (Figure 1F); however, obvious actin filament bundling occurred following DC3000 treatment (Figure 1E).
To further investigate the timing and nature of actin responses following DC3000 inoculation, we used a set of previously validated tools for measuring actin array organization [31]–[33]. We quantified and statistically compared maximum intensity projections generated from fields of Arabidopsis epidermal cells for changes in the extent of actin filament bundling (skewness) and percent occupancy (density) following microbial infection. The bundling parameter is based on the assumption that a population of individual actin filaments will have a Gaussian distribution of pixel intensities, which becomes skewed in favor of brighter pixels, when the array of actin filaments becomes more bundled [31]. The density metric is calculated as the percent occupancy of GFP-fABD2-containing pixels in each micrograph [31]. For these analyses, we performed a time-course from 0 to 36 h after DC3000 infection by sampling at 3-h intervals (Figure 2). We observed a transient increase in the abundance of actin filaments in host cells at 0–15 hpi following DC3000 treatment (Figure 2A) and this occurred as early as 15–30 min after inoculation (Figure S2). Actin filament abundance was elevated by as much as 16%, with a peak at 6–9 hpi, and then significantly decreased from 24–36 hpi onward (Figure 2A). We also detected significantly enhanced filament bundling at 18–36 hpi, with the most prominent bundling at 24–27 hpi (Figure 2B). Mock-treated seedlings had no significant changes in actin architecture compared to untreated seedlings (Figure S3). These results revealed two distinct and statistically significant changes in actin filament organization following infection with virulent pathogen, i.e. an early and transient increase in actin filament density as well as a late increase in the extent of actin filament bundling. These observations are consistent with the immediate perception of DC3000 and response of the plant immune system followed by a subsequent suppression of PTI by the pathogen.
If the rapid and transient increase in actin filament density in epidermal cells exposed to DC3000 is part of the PTI response, then we predict that the same cytoskeletal change will occur with various phytopathogens that are not adapted to Arabidopsis. To test this, we quantified actin array architecture in cotyledons at 6–9 h following infection with several non-adapted pathogens that trigger PTI in Arabidopsis, including the bean pathovar P. syringae pv. phaseolicola (Pph); Agrobacterium tumefaciens; and the rice-blast fungus Magnaporthe grisea (Figure 3). Following Pph inoculation, we observed a significant increase in actin filament density (Figure 3A), but no change in the extent of filament bundling (Figure 3B). Additionally, we observed increased actin filament density with A. tumefaciens (Figure 3C) and M. grisea (Figure 3E) treatments, but no change in bundling (Figure 3D & F). In summary, the density or abundance of actin filaments is elevated in cotyledons following treatment with various bacterial and fungal microbes and likely represents a broad-based PTI response.
Since the increase in actin filament abundance was observed in host cells in response to both adapted and non-adapted microbes (Figure 3), we predict that the increased actin filament density occurs independent of the T3SS or translocated effector proteins. Therefore, we used genetic mutants to dissect the molecular nature of P. syringae's ability to elicit the host-actin response. First, we quantified actin filament architecture in cotyledons treated with the T3SS-deficient mutant hrpH over a full time-course and observed a significant increase in percent occupancy from 0–15 hpi following inoculation (Figure 2A). Actin filament density peaked at 6–9 hpi, similar to DC3000; however, no decrease in density at 24–36 hpi was observed (Figure 2A). In contrast to DC3000, no increase in filament bundling was observed at any time-point following hrpH inoculation (Figure 2B). The similar responses to DC3000 and hrpH inoculations at 0–15 hpi further support the argument that the transient increase in actin filament density is PTI-based. Moreover, because hrpH does not induce bundling in epidermal cells whereas DC3000 does, it is likely that bundling is associated with effector-triggered susceptibility (ETS). To further dissect whether the actin organization changes were part of a general response to bacteria or can also be influenced during ETI, we quantified the actin filament architecture over a full time-course in Arabidopsis cotyledons following inoculation with P. syringae DC3000 expressing the YopT homolog, AvrPphB ([34]–[36]; Figure S4). Similar increases in actin filament density were observed with DC3000 and DC3000 expressing AvrPphB inoculations at 0–15 hpi (Figure S4A). In contrast, filament bundling was not as pronounced with DC3000 expressing AvrPphB inoculation compared to DC3000 (Figure S4B) and the filament density did not decline at 24–36 hpi, suggesting that these later changes in filament array architecture are part of a gene-for-gene response or ETI. Finally, we quantified cortical actin architecture in cotyledons treated with another T3SS-deficient mutant, hrcC [37], and a D28E mutant that expresses the T3SS but lacks most T3E genes ([38]; Figure 4). Following hrcC or D28E inoculation, we observed a significant increase in percent occupancy similar to treatments with hrpH and DC3000 (Figure 4A & C), but no change in filament bundling at 6–9 hpi (Figure 4B & D). Sensu stricto, these results demonstrate that the early increase in actin filament density is associated with PTI.
As an alternative or in addition to T3SS activity, P. syringae might alter the host-cell actin cytoskeleton through the secretion of pathotype-specific toxins, such as coronatine, a jasmonate mimic synthesized by DC3000 [39]. Following inoculation of cotyledons with the coronatine-deficient mutant, COR- [37], we observed a significant increase in filament density in host cells (Figure 4E) but no change in filament bundling (Figure 4F). This confirms that less virulent bacteria still elicit an increase in actin filament density and that coronatine is not necessary for DC3000 to elicit changes in the cytoskeleton of epidermal cells.
To further examine the mechanism of increased actin filament density during PTI, we inoculated cotyledons with a mutant of P. syringae that has the flagellin gene deleted (ΔfliC; [40]) and quantified changes in actin organization. We observed a significant increase in filament density (Figure 4G) and no change in bundling (Figure 4H). Although this increase in filament abundance was significantly less than the increase associated with DC3000 treatment, flagellin is not strictly necessary to elicit the increase in actin filament density. These results imply that the perception of other MAMPs can also lead to changes in actin organization.
To investigate whether MAMPs are sufficient to elicit a change in actin organization, we challenged plants with synthetic MAMP peptides, as well as with the fungal glucosamine polymer, chitin, and monitored changes in actin architecture. We used flg22, the N-terminal twenty-two amino acids from Pseudomonas flagellin [41], [42]; elf26, a twenty-six amino acid peptide from bacterial EF-Tu [43]; and flgAt, the amino terminal sequence from Agrobacterium flagellin which does not elicit a response in Arabidopsis [42]. These peptides were introduced at various concentrations into mature leaves and the actin responses at 0–3 h after infiltration were quantified (Figure 5 & S5). With flg22, we detected a rapid and dose-dependent increase in actin filament density (Figure 5A & S5A); however, no change in the extent of filament bundling was detected at any concentration tested (Figure 5B & S5B). We detected similar changes to actin filament arrays in plants infiltrated with chitin (Figure S5G & H). In contrast, infiltration with elf26 or flgAt was indistinguishable from mock at all concentrations tested (Figure 5 & S5C–F) even though, all three MAMP peptides were able to stimulate a PTI-based defense response as demonstrated by activation of FRK1 transcripts (Figure S6). In conclusion, treatments with flg22 or chitin are sufficient to induce rapid changes in actin filament organization.
To examine the role of early cellular signaling pathways during PTI, we performed actin architecture analysis on several Arabidopsis knockout mutants, including a susceptible Arabidopsis ecotype, infiltrated with either 1 µM flg22 or 1 µM chitin (Figure 6). Wild-type Col-0 plants showed significantly enhanced filament abundance following treatment with flg22 (Figure 6B & D) or chitin (Figure 6C & D), compared with mock treatment (Figure 6A & D). The homozygous flagellin receptor mutant, flagellin sensing-2 (fls2), in the Col-0 background, as well as the Fls2-deficient ecotype Wassilewskija-0 (Ws-0), both lacked a significant increase in filament abundance following treatment with flg22 (Figure 6G & I and Figure 6L & N, respectively). Whereas with chitin treatment, actin filament abundance was significantly increased following treatment of epidermal pavement cells in either fls2 (Figure 6H & I) or Ws-0 (Figure 6M & N) plants. To further test which signaling pathways may involve actin, we performed actin filament density analysis on two additional DC3000-susceptible Arabidopsis knockout mutants, brassinosteroid insensitive1-associated kinase1 (bak1-4) and botrytis-induced kinase 1 (bik1), which are both known to associate with the FLS2 receptor [17], [18]. Neither bak1-4 (Figure 6P–T) nor bik1 (Figure 6U–Y) homozygous mutant plants showed significant changes from mock following treatment with flg22 or chitin. Additionally, no significant changes to filament bundling were observed following treatment with either flg22 or chitin for any of the plant lines tested. Since flg22 peptide and chitin are sufficient to elicit an increase in actin filament abundance, we tested whether Pseudomonas bacteria could elicit a similar response in Arabidopsis lines with altered defense signaling. Wild-type Col-0, the Ws-0 ecotype, and the fls2 knockout mutant were dip-inoculated, individually, with multiple Pseudomonas mutants and changes to the host cytoskeleton at 6–9 hpi were investigated (Figure S7). Notably, the actin response was reduced in both the fls2 mutant (Figure S7C) and the Ws-0 ecotype (Figure S7E), but not completely ameliorated, indicating that some changes to the actin cytoskeleton are independent of FLS2. However no significant changes to the extent of actin filament bundling were observed with any Pseudomonas strain or host-plant at this time-point (Figure S7B, D & F).
Latrunculin B (LatB) is a macrolide compound from marine sponges that inhibits actin polymerization by binding to monomeric actin and preventing its assembly onto filament ends [44]. To test whether actin polymerization is necessary for the increase in filament density during the initial response to phytopathogens, we co-infiltrated Arabidopsis leaves with various concentrations of LatB and DC3000 or hrpH Pseudomonas strains. At 6–9 hpi, actin filaments in epidermal pavement cells appeared to be markedly reduced following infiltration with LatB alone (Figure 7A, D & G) or LatB co-infiltrated with either DC3000 (Figure 7B, E & H) or hrpH (Figure 7C, F & I). As shown previously [1], short-term treatments with low doses of LatB primarily affected the dynamic actin filament arrays and individual filaments, whereas only modest effects on filament bundles were observed (Figure 7D–I). Using the metrics described earlier, we measured a significant decrease in actin filament density following LatB infiltration in mock-treated plants compared to the 0 µM control (Figure 7J). Further, we measured a significant and dose-dependent reduction in actin filament abundance in plants that were co-infiltrated with LatB and DC3000 or hrpH (Figure 7J). However, no significant changes to the extent of filament bundling were observed at this early stage of infection (P-value = 0.49, ANOVA; data not shown). This demonstrates that actin polymerization is necessary for the increase in filament density following infection with pathogenic and non-pathogenic bacteria during the PTI response.
If actin polymerization is an important component of PTI, then blocking this aspect of the host-cell response should enhance susceptibility to pathogens. To test this, we again co-infiltrated Arabidopsis plants with various concentrations of LatB and either DC3000 or hrpH and measured bacterial growth at 0, 6, 12, 24, and 48 h after treatment. As predicted, bacterial growth was significantly increased in plants co-infiltrated with DC3000 and 1 or 10 µM LatB compared to 0 µM LatB treatments (Figure 7K). We also measured a significant increase in bacterial growth following co-infiltration with the non-pathogenic strain, hrpH, and LatB (Figure 7K) indicating that the host-actin cytoskeleton and actin polymerization are necessary for innate immunity.
Using robust tools for measuring actin organization [31]–[33], we quantified temporal changes to actin arrays in Arabidopsis seedlings infected with the phytopathogenic bacterium P. syringae pv. tomato DC3000. Here we provide the first report that changes in plant actin array organization occur in response to pathogenic bacteria. We observed an early transient increase in the density of actin filament arrays that was corroborated by treatments with flg22 MAMP peptide, and chitin, as well as with several non-host adapted pathogens. We also demonstrated that several components of host-cell signaling (i.e. FLS2, BAK1, and BIK1) were necessary for this response, providing the first evidence for receptor-mediated signaling to the actin cytoskeleton during plant innate immunity. Finally, we used the pharmacological agent LatB to demonstrate that actin polymerization and an increase in filament abundance are necessary for innate immunity. Collectively, these observations demonstrate unambiguously that the rapid increase in actin filament abundance is a component of PTI, requiring recognition of a bacterial MAMP by an Arabidopsis MAMP-receptor complex. Secondly, we observed a decrease in actin filament abundance and a late increase in filament bundling that was only associated with pathogenic DC3000 in a T3SS-dependent manner. Therefore, we suggest that these changes are a consequence of the action of T3E proteins.
The increase in actin filament density observed in Arabidopsis cotyledons is associated with a rapid PTI response; it could be detected as early as 15–30 min after inoculation with DC3000. This response peaked at 6–9 hpi and was abrogated by 15 hpi. Further, this increase occurred in cotyledons inoculated with pathogenic and non-pathogenic Pseudomonas strains and mutants. It also occurred when seedlings were inoculated with Agrobacterium tumefaciens, and the non-host fungus Magnaporthe grisea, which indicates that the increase in actin filament density is a broad-based response to microbial perception. The increase in actin filament density is not the result of mechanical stimulation by the T3SS, as two T3SS-deficient mutants (hrpH and hrcC) still elicited the early cytoskeletal response. Although we observed significantly less actin filament percent occupancy following D28E inoculation compared to DC3000, the density of filaments was still significantly higher than the mock-treated control. Since each of the P. syringae mutants utilized herein is thought to contain the same suite of MAMPs (with the notable exception of ΔfliC), this significant reduction in filament abundance may imply a reduction in bacterial growth or indicate the action of T3E proteins in this response. The latter possibility seems unlikely as plants treated with P. syringae DC3000 expressing AvrPphB led to an similar actin response at early time-points and a reduced filament and bundling response at late time-points.
Changes to actin filament density, usually described as increases in polymerization [9] or an increase in phalloidin-stained spots [7], are also observed with mutualistic bacteria as well as with compatible and incompatible fungal and oomycete interactions [3], [20]–[23]. Despite the commonly accepted dogma that actin responses are caused by the mechanical stress of invading pathogens [5], [21]–[23], our data indicate that host-cell penetration by an invading pathogen is not necessary to elicit changes to the actin cytoskeleton. Here, we show that Magnaporthe spores, which do not attempt to penetrate Arabidopsis epidermal cells, still elicit an increase in actin filament abundance. Previously, a Magnaporthe mutant that is deficient for the ability to penetrate cells could also stimulate rearrangement of the host cytoskeleton in onion epidermal cells, leading the authors to speculate that this was due to sensing fungal cues other than the penetration peg [45]. Unlike fungi and oomycetes which use specialized structures to penetrate host cells, phytopathogenic bacteria reside in intercellular spaces; as a result, signals indicating the presence of these bacteria may affect multiple cells, rather than a single point of invasion. Therefore, mounting a host-defense response likely requires a broad and non-localized defense mechanism using cell surface receptors. Taken together with our quantitative results, it is likely that the increase in actin filament density is a conserved, receptor-mediated response to the perception of microbes by host plant cells.
Significantly, both bacterial and fungal MAMPs were capable of eliciting the increases in actin filament density in host plants. In this study, the flg22 peptide mimic of bacterial flagellin was sufficient to elicit a dose-dependent increase in actin filament density as early as 0–3 hpi. This fast response was not observed with a peptide mimic of the bacterial elongation factor EF-Tu (elf26), which indicates that the actin response may be specific to particular MAMPs, or could represent differential expression of MAMP receptors in organs, tissues and specific cell types [46]. For example, the EF-Tu receptor, EFR, may not be expressed in epidermal cells of cotyledons or rosette leaves but is still expressed in the whole organ. This could lead to lack of a detectable actin-based response in epidermal cells, whereas transcriptional hallmarks of PTI are still present in the whole organ. Another alternative is that the actin-based response in epidermal cells occurs rapidly and goes undetected over the timescales we are able to measure by SDC microscopy. In this way, a fast response would become “averaged-out” at the earliest timepoints measured. The increase in actin filament density still occurred in response to the ΔfliC mutant, indicating that additional MAMPs also trigger this response or that the presence of flagella is not completely abrogated in the ΔfliC mutant, despite lack of mobility. This increase in actin abundance occurs independently of the FLS2 receptor following DC3000 or ΔfliC treatment. Since, multiple Pseudomonas mutants are still capable of eliciting the increase in actin filament abundance 6–9 hpi in the absence of components of the FLS2 receptor complex, this further indicates that multiple or additional MAMPs are capable of altering host-actin architecture. Additionally, the application of fungal chitin on leaves (this study) or bacterial Nod factors on root hairs [9] also stimulated an increase in actin filament density, which further indicates that the host-actin cytoskeleton plays a general role in the perception of beneficial and pathogenic microbes.
A main hallmark of PTI is signaling through MAPK and CDPK phosphorylation cascades after host-perception of various MAMPs [4], [16]–[19]. For example, once flagellin or flg22 has bound the FLS2 receptor, BAK1 associates with FLS2, and following this association, the cytoplasmic kinase BIK1 dissociates from the receptor complex initiating host-defense signaling [16]–[19]. As expected, Arabidopsis mutants or ecotypes with deficiencies in the flagellin-sensing pathway (i.e. fls2, Ws-0) did not display changes actin filament architecture following treatment with the flg22 MAMP peptide. However, actin architecture changes still occurred following chitin treatment, indicating that chitin-induced signaling is still intact. Further, mutants in the shared signaling nodes between the flagellin- and chitin-signaling pathways (i.e. bak1-4 and bik1) did not display any significant changes to actin architecture following either bacterial or fungal MAMP treatment, indicating that actin rearrangements are conserved in several common immune pathways.
Our data provide evidence that the host-actin cytoskeleton plays an important role in innate immunity because the actin polymerization inhibitor LatB promotes the growth of P. syringae DC3000 on Arabidopsis leaves. Further this growth advantage is conferred specifically during PTI as the T3SS-deficient mutant hrpH also exhibited significantly enhanced growth. It is well known that an intact actin cytoskeleton is required for receptor-mediated endocytosis of ligands including the flagellin receptor FLS2 [47], as vesicle dynamics are reduced following treatment with either LatB or the actin stabilizer endosidin1 [48]; however, the actin cytoskeleton almost certainly plays additional roles during response to microbes. The requirement of the actin cytoskeleton for activation of NADPH oxidase at the plasma membrane, as well as Golgi, peroxisomes and endoplasmic reticulum trafficking toward sites of fungal and oomycete penetration, has been demonstrated through pharmacological studies [20]–[23]. Presumably, the trafficking of Golgi and ER is important for the production and deposition of antimicrobial compounds and fortification of the cell wall [20]–[23]. The specific targeting of defense proteins to the cell membrane is also an actin cytoskeleton-dependent process. The fungal resistance protein RPW8.2 prevents haustorium development and reduces oxidative damage to host cells by generating a unique membrane that fuses to the extrahaustorial matrix (EHM; [49]). Targeting of RPW8.2 to the membrane is disrupted with cytochalasin E treatment; plants susceptible to the powdery mildew fungus had less EHM localization and are unable to activate the same proteins as resistant plants [49]. In contrast, one actin-independent mechanism in plant defense is the accumulation of a SNARE involved in membrane fusion events at the plasma membrane, PEN1, at the fungal penetration site [20], [23], [50]. This differs from PEN2 and PEN3, which are implicated in callose deposition and require the actin cytoskeleton for proper localization during fungal infection [22], [50], [51]. Finally, perturbations to the actin cytoskeleton using drugs and toxins have been shown to trigger or alter programmed cell death in plant cells [2]. Taken together, these observations speak to the involvement of the actin cytoskeleton at various time-points, from minutes to hours, during common biotic stress events.
The regulation and turnover of the actin cytoskeleton requires the concerted activities of hundreds of actin-binding proteins that can respond to signals to polymerize or destroy actin filament networks. The growth of individual actin filaments in the cortical array of Arabidopsis epidermal cells is extremely fast, ∼2 µm/s, and most filaments exist for less than 30 s before being destroyed by prolific severing activity [1], [33], [52]. This constant formation and destruction of actin networks requires a huge expense of energy—on the order of millions of ATPs per second—and is thought to represent a surveillance mechanism to various biotic and abiotic stresses [1], [12]. It is easy to imagine that changes to any number of actin-binding proteins involved in actin dynamics could result in altered filament arrays; and that targeting specific aspects of the cytoskeleton would be an excellent opportunity for successful pathogen attack.
Plant actin-binding proteins respond to a plethora of second messengers in signaling cascades, including Ca2+, phospholipids and pH [2], [53], [54]. A potential link between the actin cytoskeleton and specific cytosolic Ca2+ signatures following microbial infection requires additional study [2], [55]. Further, calcium and pH fluctuations are known to occur in Arabidopsis during defense responses [56], [57]. Several plant actin-binding proteins have different activities as pH fluctuates from alkaline to acidic [58], [59]. Finally, there is a long history of alterations to actin filament arrays through actin-binding proteins sensing changes to concentrations and types of phospholipids like phosphatidylinositol (4,5)-bisphosphate (PIP2) and phosphatidic acid [60]. Phosphatidic acid is also a second messenger for plant defense responses that can activate MAPK signaling and defense genes [61]–[63] and accumulates upon treatment with various MAMPs [63]–[66].
The initiation of immunity in plants requires the concerted effort of both PTI signaling and the recognition of microbe-derived proteins evolutionarily adapted to circumvent innate immunity. A hint that the host cytoskeleton is a target for effector proteins, comes from the use of the Harpin elicitor, which triggers defense responses in host and non-host plants [67]. Specifically, Harpin elicitor treatment of grapevine cells triggers host-microtubule depolymerization within 3 hours, but has variable effects on the actin cytoskeleton [67]. The first example of a bona fide phytopathogenic effector protein specifically targeting the plant cytoskeleton is HopZ1a, which depolymerizes microtubules thereby disrupting the plant secretory pathway and suppressing cell wall-mediated defenses [27]. The involvement of microtubule rearrangements during PTI that results from recognition of DC3000 is unclear since this particular pathovar of P. syringae does not elicit changes in microtubule organization [27] and lacks HopZ1. Additionally, which cytoskeleton is targeted first remains an unanswered question, as DC3000 expressing HopZ1a did not disrupt the actin cytoskeleton at 16 hpi [27] and our data show changes to the actin cytoskeleton as early as 15 min after inoculation. Effector proteins likely target the actin cytoskeleton, as inoculations with the T3SS-deficient hrpH did not elicit the increased bundling that was obvious with DC3000 treatment. A role for actin in ETI is indicated by data from adf4 knock-out Arabidopsis plants, which are unable to elicit a hypersensitive response and are susceptible to P. syringae expressing AvrPphB [30]. Although the exact mechanism by which ADF4 mediates resistance to bacteria carrying AvrPphB is still unknown, it has been demonstrated recently that ADF4 is required for activation of resistance to DC3000 expressing AvrPphB through control of expression of the R-gene RPS5 [29]. Furthermore, this work correlated changes in the localization of ADF4 with the reduced expression of FRK1 and MAPK signaling, further implying a dual role for the actin cytoskeleton in the host response to phytopathogens.
In summary, we monitored the nature and timing of changes to the actin cytoskeleton in Arabidopsis during microbial infection. We quantified two distinct actin responses—a rapid transient increase in actin filament density and a late increase in filament bundling. We demonstrate that the early transient increase in actin filament density is associated with PTI by using adapted and non-adapted microbes and treatments with MAMPs. We also established the requirement of host-cell signaling machinery including the flagellin receptor complex, FLS2, BAK1 and BIK1, for the increase in actin filament abundance. This is the first evidence for temporal changes in actin cytoskeleton organization during PTI elicited by a phytopathogenic bacterium, and uncovers the initial MAMP signaling cascade responsible for altering the cytoskeleton.
Arabidopsis thaliana Ws-0, fls2 (SALK_062054), bak1-4 (SALK_116202), bik1 (SALK_005291) were transformed with GFP-fABD2 [68] using the floral dip method described previously [69]. T1 plants were screened on appropriate antibiotics and by fluorescence microscopy. Multiple T2 plants (n≥9) for each genotype were used for actin architecture analysis. Wild-type A. thaliana Col-0, Col-0 expressing GFP-fABD2 [1], and T2 plants expressing GFP-fABD2 were sown onto soil and stratified for 3 d at 4°C. Flats were transferred to a growth chamber and plants grown under long-day conditions (16 h light, 8 h dark) at 21°C for 10 or 24 d.
Information about the Pseudomonas mutants and strains, as well as Agrobacterium and Magnaporthe strains, used in this study are found in Table S1 in Text S1. Various bacterial strains were grown on NYGA media (0.5% [w/v] Bacto Peptone, 0.3% [w/v] yeast extract, 2% [v/v] glycerol, 15% [w/v] Bacto agar) and diluted with 10 mM MgCl2 to a concentration of 3×107 colony-forming units (CFU) mL−1 [70]. Ten day-old Arabidopsis seedlings were infected by gently agitating inverted seedlings in bacterial suspensions supplemented with of 0.02% [v/v] Silwet. MAMP peptides, flg22 [41], elf26 [43], and flgAt [41] all from NeoBioSci (Cambridge, MA), and chitin (Sigma-Aldrich, St. Louis, MO) were diluted in 10 mM MgCl2 at various concentrations. For hand infiltration of peptides, and latrunculin B co-infiltration with Pseudomonas strains, leaves of similar size from 24 d-old plants were designated for analysis with a marker. Leaves were gently infiltrated with an inoculum of 1×105 CFU/mL using a 3 mL needle-less syringe until intercellular spaces were filled with solution (∼300–400 µL per leaf). After infiltration, inoculated plants were covered for 30 min and returned to the growth chamber prior to imaging.
All image collection and data analyses were performed as single-blind experiments. Actin filament bundling and percent occupancy were measured using two metrics: skewness, based on the assumption that a population of actin filaments exhibits enhanced pixel intensities when bundled; and, density, an estimation calculated as the percent occupancy of signal (actin filaments) separated from background by setting a minimal threshold to include all actin filaments [31]. We imaged fields of epidermal pavement cells with spinning disk confocal microscopy (SDCM) by collecting 24 steps of 0.5 µm each starting at the plasma membrane. Spinning disk confocal microscopy was performed using a Yokogawa CSU-10 mounted on a Zeiss Observer Z.1 equipped with a 100X/1.46 NA PlanApo objective. Illumination was from a solid-state 50-mW laser with AOTF control over excitation wavelength (Intelligent Imaging Innovations, Denver, CO). The 488-nm laser emission was captured with an Evolve512 EMCCD camera (Photometrics, Tucson, AZ). The SDCM was operated using SlideBook software (version 5.0.031; Intelligent Imaging Innovations). A fixed specimen exposure time and gain setting were selected such that individual actin filaments could be observed, but actin filament bundles were not saturated. Maximum-intensity projections of z-series stacks were analyzed in ImageJ using algorithms described previously [31]–[33]. Gaussian blur and high-band pass filters were applied to projections prior to density analysis. No image processing was applied to maximum-intensity projections that were analyzed for skewness. At least 90 images were analyzed per time-point per treatment, from at least 30 individual seedlings for each measurement. Statistical comparisons and ANOVA test with Tukey HSD post-hoc analysis were carried out using KaleidaGraph (version 4.1.3b1; Synergy Software, Reading, PA).
|
10.1371/journal.pntd.0001467 | Should I Get Screened for Sleeping Sickness? A Qualitative Study in Kasai Province, Democratic Republic of Congo | Control of human African trypanosomiasis (sleeping sickness) in the Democratic Republic of Congo is based on mass population active screening by mobile teams. Although generally considered a successful strategy, the community participation rates in these screening activities and ensuing treatment remain low in the Kasai-Oriental province. A better understanding of the reasons behind this observation is necessary to improve regional control activities.
Thirteen focus group discussions were held in five health zones of the Kasai-Oriental province to gain insights in the regional perceptions regarding sleeping sickness and the national control programme's activities.
Sleeping sickness is well known among the population and is considered a serious and life-threatening disease. The disease is acknowledged to have severe implications for the individual (e.g., persistence of manic periods and trembling hands, even after treatment), at the family level (e.g., income loss, conflicts, separations) and for communities (e.g., disruption of community life and activities). Several important barriers to screening and treatment were identified. Fear of drug toxicity, lack of confidentiality during screening procedures, financial barriers and a lack of communication between the mobile teams and local communities were described. Additionally, a number of regionally accepted prohibitions related to sleeping sickness treatment were described that were found to be a strong impediment to disease screening and treatment. These prohibitions, which do not seem to have a rational basis, have far-reaching socio-economic repercussions and severely restrict the participation in day-to-day life.
A mobile screening calendar more adapted to the local conditions with more respect for privacy, the use of less toxic drugs, and a better understanding of the origin as well as better communication about the prohibitions related to treatment would facilitate higher participation rates among the Kasai-Oriental population in sleeping sickness screening and treatment activities organized by the national HAT control programme.
| Active screening strategies are common disease control interventions in the context of poor and remote rural communities with no direct access to healthcare facilities. For such activities to be as effective as possible, it is necessary that they are well adapted to local socio-economic and cultural settings. Our aim was to gain insight into the barriers communities in the Kasai-Oriental province of the Democratic Republic of Congo experience in relation to their participation in active screening activities for African sleeping sickness. Participation rates seem to be especially low in this province compared to other endemic regions in the country. We found several important factors to be in play, a number of which could be addressed by adapting the operational procedures of the mobile teams that perform the active screening activities (e.g., improved confidentiality during the screening procedure). However, more profound considerations were found in the form of regional beliefs related to the treatment of the disease. Although not based on rational grounds, these prohibitions seem to pose a significant barrier in a person's decision to seek diagnosis and treatment. A better understanding of these prohibitions and their origin could lead to improved participation rates for sleeping sickness screening in Kasai-Oriental.
| Human African Trypanosomiasis (HAT) or African sleeping sickness is a parasitic disease unique to sub-Saharan Africa. It is caused by protozoa of the Trypanosoma Genus of which the species T. brucei gambiense and T. brucei rhodesiense cause disease in humans. According to WHO figures [1], [2] the disease is present in 36 sub-Saharan countries, where 60 million people are at risk of which less than 4 million are under surveillance. In 2006 the annual number of new cases was estimated to be between 50,000 and 70,000, but only a fraction of that number of cases is reported. Between 1998 and 2009 the number of annually reported HAT cases has dropped from 37,991 to 9,878 [2], [3]. T. brucei gambiense causes the chronic form of the disease which is endemic in central and western Africa, while the acute form caused by T. brucei rhodesiense is found in East-Africa [4].
HAT control strategies are based on early case detection and treatment and vector control [1], [5]–[7]. Active screening strategies conducted by mobile teams which travel from village to village have substantially lowered the case load in several African countries, most notably in Uganda and Sudan [8]–[10]. In the Democratic Republic of Congo (DRC), where T. brucei gambiense is endemic, this control strategy has led to a considerable decrease in case numbers throughout the whole country, albeit with a large variability among the endemic provinces. In the North Equator and Kinshasa provinces a significant decrease in prevalence was observed. However, the number of cases detected in two other provinces, Bandundu and Kasai-Oriental, has remained unabated notwithstanding an increased screening effort [11], [12]. Many factors could explain this intervariability in HAT prevalence among provinces which otherwise are all subject to the same HAT control strategy. Amongst others, such factors could be related to a low coverage of the population at risk, low community participation rates in active screening and treatment activities, or an inefficient management and coordination of the mobile teams. Treatment failure could also play an important role, as studies have shown significant treatment failure rates across the country's provinces, from 5 to 10% in Bandundu, to 25% in North Equator, and possibly as high as 50% in Kasai-Oriental [13], [14]. In 2007 adherence to treatment in DRC was at 91.6% [15], with very little variation between endemic provinces. A recent analysis of the operational effectiveness of HAT screening and treatment activities in Kasai-Oriental (stagnation of infection rates) and North Equator (decrease of infection rates) in 1998, 2001 and 2005 by Lumbala (unpublished, manuscript in preparation) shows that low community participation rates in active screening activities and coverage of the population at risk could well explain the differences between these two provinces. While issues related to the coverage of the population at risk lie with the management and coordination of the national control programme's activities (availability of resources, control strategy), community participation rates reflect the level of health service utilization which can be influenced by various social, economical and cultural factors.
It is thought that the low participation rate of the rural population in the active screening activities is one of the main reasons for the continued HAT transmission in Bandundu and Kasai-Oriental. In a study on the effectiveness of active population screening and treatment for sleeping sickness in DRC, Robays et al [12] observed that the overall mean participation rate in the provinces of Equateur, Bandundu and Kasai-Oriental was 74%. However, important variability of the attendance rates between villages, between mobile teams and between provinces were observed. In Bandundu and Kasai-Oriental, participation rates as low as 64% and 50% were respectively found.
The reasons for this low rate of community participation in screening activities in these two provinces has been little investigated, although a better understanding of relevant cultural and socio-economic barriers could significantly improve the effectiveness of HAT control programmes. Robays et al [16] showed for example that in DRC's Bandundu province the fear of drug toxicity, financial barriers and the lack of confidentiality during screening were the most important obstacles for participation in the HAT campaign in that province. As a result of those findings, the DRC's HAT control program abolished the nominal user fee for screening. Although this fee was minimal, it still presented a hurdle for large families with no or little income. These findings might be considered indicative for other provinces of DRC, but care should be taken not to generalise as socio-economic and cultural contexts are very heterogeneous across the country. Therefore, this study aims to document in a qualitative manner the economic and socio-cultural factors which may influence community participation in active HAT screening and treatment activities in the Kasai-Oriental province, where the disease puts not only rural communities at risk but also a large group of workers who are engaged in diamond mining activities. Additionally, as is illustrated by the difference in local languages (Tshiluba in Kasai-Oriental, Kikongo in Bandundu), the Kasai-Oriental and Bandundu provinces differ distinctly on a cultural level, which may influence local disease perceptions and health seeking behaviour habits.
We performed a transversal descriptive qualitative study using the focus group discussion (FGD) technique in five health zones in the Kasai-Oriental province of DRC: Miabi, Tshilenge, Tshitengue, Kasansa and Mukumbi. Figure 1 shows the geographic locations of these health zones.
These health zones were selected on the basis of their location in a historic and currently active focus of HAT near the Kasai-Oriental capital of Bakwanga (Kalelu- Lubilashi) which represents almost 70% of the total number of reported cases in the province [17]. Table 1 shows the HAT prevalence data for all five health zones in 2005 and 2007. Along with the region's linguistic and cultural characteristics, the socio-economic setting in Kasai-Oriental is very diverse due to the presence of a diamond industry which employs many workers who live in encampments surrounding the diamond mines. In Kasai-Oriental the main populations at risk are diggers working in the diamond mines and farmers, who represent the most active groups in the population.
The data collection took place in January 2008. Thirteen FGDs were conducted. The focus groups were divided along three categories: gender, geographic characteristics (i.e. worker camp or village) and health zone. Table 2 shows the characteristics of each of the FGDs performed in this study.
The number of FGDs conducted in the different health zones reflects the “richness” of the information found in those health zones. We continued to hold focus group discussions in a particular health zone until data saturation was reached and no new information was coming out of them. The FGDs were stratified by gender to ensure the homogeneity of the groups and to promote openness during the discussions, as women in general do not speak freely in front of men in Kasai-Oriental. One of the focus groups in Mukumbi was mixed because we were not able to identify a minimum of six women willing to participate in an FGD in that specific location.
The question guide used in the FGDs was pre-tested and fine tuned in two focus group discussions performed in a health zone not included in the study area. The quality of these discussions was evaluated with the research team before finalizing the question guide. The following topics were covered: general knowledge of the disease; community practices regarding the disease; the community's attitude towards the mobile teams; the community's participation in screening and treatment activities; and community's expectations regarding HAT control services.
The FGDs numbered six to nine participants who were invited by the principal investigator (A.M.). They were selected at random based on their availability and whether they gave consent for their participation. Two exclusion criteria were used: (i) participants had to be resident of the local community for at least 2 years (a definition also used by the national HAT control programme in their population surveillance data); (ii) community leaders such as teachers, village chiefs and priests were excluded in order to avoid them dominating the discussion dynamics. The FGDs took place in a hut assigned by the community chief for this purpose. The average duration of the discussions was 45 minutes. They were held in the local language, Tshiluba. A local doctor and nurse, both native Tshiluba speakers, were trained to moderate and observe the FGDs. They switched roles for each discussion. Both had previously never been involved in the screening activities of the national HAT control programme. Their training, which was conducted by A.M., consisted of a two-day course during which they were briefed on the study objectives and taught how to conduct FGDs. Each discussion was evaluated and discussed post-hoc by A.M. and the two assistants. A.M. attended all the FGDs to supervise the process. All discussions were fully recorded on a digital audio recorder.
The FGDs were transcribed and translated into French by the two research assistants, who took turns in both tasks. A secretary prepared the transcripts in Microsoft Word. A.M. revised all the transcripts prior to analysis. QSR Nvivo8 was used to support the data analysis and identify trends in the data. This software allows researchers to organize and analyze complex and unstructured datasets by fragmenting and categorizing data whilst keeping a link with the source documents (transcripts of the FGDs in this case). The analysis itself is an inductive process which allows themes to emerge from the data. These themes are coded into categories which are continuously refined throughout the analysis. Finally, relationships between categories are created and inferences are made. The analysis and its process were discussed with the co-investigators.
The FGDs were coded, each element of the analysis representing an intervention by a focus group discussion participant. The codes were developed progressively and in an inductive manner, allowing relevant themes to emerge from the data. The coding book was discussed and refined with the co-investigators to assure the significance of the analysis.
The study protocol was approved by the thematic HAT institutional review board (IRB) in DR Congo and the IRB of the Institute of Tropical Medicine, Antwerp, Belgium. Local community authorities were asked for permission to perform the study in their villages. Focus group discussion participants were informed about the voluntary nature of their participation. Permission for tape recording the conversations was requested prior to starting each of the focus group discussions. Anonymity of the participants was guaranteed and no personal details were recorded. Oral consent was obtained and audio recorded before the start of each focus group discussion. Oral consent was preferred since regional literacy levels are low. This procedure for consent was approved by the IRB.
In this section we will present the various aspects of community perceptions related to HAT which were elaborated during the FGDs. Data is presented in the form of quotes from the FGDs. The quotes used in this section are illustrative of the sentiments reflected in the FGDs and were selected on the basis of their aptness and informative quality.
In general, sleeping sickness is well known in the region and is considered an affliction which has been around for many generations. In the local tongue it is referred to as ‘disama dya tulu’, which literally translates to ‘sleeping sickness’. “Sleeping sickness has been around for ages. We have heard people talking about it since our childhood. It is an ancient disease.” [FG10] Nevertheless, whilst many referred to specific cases from their direct environment during the discussions -“I know sick people. My mother has the disease and so does my sister.” [FG4]- , not all participants had personally been confronted with the disease in the past. “Well, I can't really say as I have never seen the disease. A typical example of a case? I haven't seen one yet.” [FG1] A number of people said they did not have any ‘scientific knowledge’ of the disease and were not completely at ease with the medical rational approach adopted by the health workers of the national control programme. “I know sleeping sickness exists, but the knowledge about how to avoid it is given by doctors who ask people to avoid this or that disease in this or that way.” [FG10]
When discussing the symptoms which they relate to the disease, behavioural problems, sleep, tiredness, fever and headaches were all commonly referred to in the FGDs. Behavioural problems and the linked personality changes on the one hand, and irregular sleeping patterns on the other, were considered the most tell-tale signs of sleeping sickness. Women seemed to be more knowledgeable about the symptoms and their relation to the disease than men. This could be explained by the women's traditional role as caregiver. When a family member becomes sick, it is generally the mother who accompanies them to the health services and who is briefed by the health staff about the disease and how to take care of the patient. Women are also more likely to participate in the disease screening activities and thus to interact with health workers, a trend which was also observed in other studies [18], [19]. The symptoms which are typical for the early stage of the disease, such as fever, headaches and tiredness, seem to be perceived less in the worker camps surrounding the diamond mines than in the villages. When the disease is in its late neurological stage the affected person is considered to be unaware of his state. Rather, it is his entourage which identifies him as abnormal and unable to take care of himself. “When a person has this disease, she is not able to reason for herself. Only those who are at her side can do it in her place.” [FG5]
Sleeping sickness is perceived as a severe illness since many people are affected by it and many patients die. “Many people have the disease, because we have seen with our own eyes how many people have died of it. We can't give an exact number, but many people have the sleeping sickness.” [FG4] The consequences of sleeping sickness are numerous and very visible within the communities, adding to the perceived severity.
The disease has serious repercussions on the patient, the family and the community. The high case fatality rate and the iatrogenic deaths induced by the toxic drug melarsoprol are very well known. Moreover, a large number of participants in our FGDs pointed towards the neuropsychiatric sequelae in those surviving, such as lunacy and trembling hands, which do not always regress after treatment. “This disease turns people into idiots. In order to control him, he needs to be restrained by force, even if he doesn't agree with that. He doesn't want a child to come near him. If it does happen, he might kill the child. It's the same with tall people. Afterwards, when he has been treated, his spirit never returns to normal. He remains confused and doesn't know how to do things.” [FG13] It is considered shameful to be affected by the disease and stigmatisation is common. This does however not imply a rejection by the community, but rather signifies a shift of the patient's place in the community. In other words, the social role of the patient in the community changes, together with the expectations, rights and obligations which go along with that role.
At the level of the family the repercussions are for one part socio-economic since when a family breadwinner becomes ill he no longer is able to work and provide for the family's needs. “The concern is that the sick person could be working to support the family. But now with him being sick, all those who depended on him share in his misfortune.” [FG5] The disease puts many additional strains on family ties and marriages. When the man is sick for example, the woman is often obliged to leave the house in order to avoid sexual relations with her husband, which is locally considered to be strictly forbidden for actual or recovering HAT patients. If it is the woman who is sick, the man in general goes in search of another partner. This can have drastic consequences for the family unit, as illustrated in the following quote: “Women can also get the disease, in which case the man doesn't wait. He takes another woman into his house and puts his partner out on the street together with the children, as they might be sorcerers, so they are banished. We see them every day. We call them ‘the children of the market’” [FG6]. In some instances a case of HAT can also lead to family conflicts caused by the search for a potential sorcerer considered to be at the origin of the disease. “Sometimes they say that the sick person's mother, paternal aunt or uncle is at the source of the disease. They seem to forget that the tsé-tsé fly is where the disease comes from.” [FG13] On the community level, an increase of HAT cases can have an impact on the general development of the village. “One can say that many people mainly work in the fields. When someone becomes sick, she no longer has the strength to cultivate the fields. That has a negative impact on the village because she no longer produces food for the village.” [FG5] In extreme cases this can even lead to an implosion of the local community. “When many people, or everybody, becomes sick, no work is done anymore. Many have become idiots, others behave like madmen. The village is dead.” [FG4] Such social and economic consequences can be very far reaching and eventually lead to forced migration to other villages.
In general, the vector of African sleeping sickness and its role in disease transmission is well known in the communities. The tsetse fly was elaborated upon in all FGDs and is locally known under various names: dibudu, bibuiba and kabwibwibwa. “The fly bites, she leaves behind this disease. She can be found in the villages, in the forests and near water. This insect that we call ‘dibubu’” [FG2] However, several other causes or modes of transmission were also stated. For example, traditional beliefs and sorcery were sometimes referred to when the spread of African sleeping sickness was discussed. “When the adults of the village address solemn words against this disease, it diminishes. So it is provoked by the realm of darkness.” [FG6] Some women talked about the role of the amaranth—a green plant cultivated in the region—and pigs as a source of the diseases. “When we arrived here, they told us that the amaranth gives people the sleeping sickness. When a person eats the amaranth, she will get the sleeping sickness.” [FG8] “I have already heard that the disease comes from the pigs here in the village. They say that bad pigs carry this disease.” [FG7] Such beliefs are likely grounded in the indirect role amaranths and pigs might play in the diseases, transmission, since tsetse flies seem to be more abundant in and around crop fields and animal pens.
Other discussants thought that the members of the mobile teams played an important role in spreading the disease, transmitting the disease while they perform the screening. “Observing how those who have the disease suffer, the population thinks that the nurses of the team carry the disease and transmit it to you the moment you stand in front of them.” [FG5] Finally, although rarely mentioned in the focus group discussions, contagion was also elaborated as a means by which the disease spreads. “Because it is a contagious disease. If we eat together with him, we might get contaminated.” [FG1]
In general, the communities talk about vector control activities as the most important way of prevention. This further reflects their general understanding of the tsetse fly's role in the transmission of the disease. “We take our machetes to cut down the trees, palms and others. Bibuibua will be scared and take flight.” [FG3] “How the disease can be avoided? By pouring medicine in the rivers and placing traps for the bibuibua.” [FG3] “That person should be educated and told to wear white clothes when she goes into the jungle.” [FG6] On the other hand, there is a feeling that prevention is in essence impossible because villagers cannot abandon their livelihood activities, which take place in locations, such as the fields, where the tsetse fly is found. “There is no way to avoid contact since when we go to the fields the flies bite us.” [FG6] The importance of raising awareness in communities is also referred to as an efficient way of preventing infections. The mobile teams are seen as an important channel in this respect. When they arrive in a village and before they start the screening activities, the nurses of the team give a lesson on how to avoid the disease. “They explain us how we can avoid getting the disease, how the disease enters the village and why we should avoid shabbiness around our houses.” [FG3]
The low attendance rates at the screening activities organised by the national HAT control programme pose a significant problem. When discussing the reasons for this observation during the FGDs, six main barriers were identified: giving priority to occupational activities; the toxicity of the drugs used in treatment; distrust towards the nurses of the mobile teams; fear of lumbar punctures; fear of unsolicited HIV/AIDS tests; and the lack of confidentiality during the screening procedure itself.
The population wakes early in the morning to leave for the diamond mines or crop fields. The mobile teams generally arrive in the villages later in the morning, with little or no prior notice, after most of the working populace have already left. Furthermore, people tend to avoid screening as long as they consider themselves to be healthy. Why risk being diagnosed and having to give up your livelihood activities for a prolonged period of time for treatment if you do not feel sick? “The people are afraid. Everyone reasons that if they catch me with sleeping sickness, I will no longer be able to do all my work. My activities won't be able to take place anymore. So it is better not to be tested as long as I don't feel sick. Once I do, then I go to the doctors. They will take care of me.” [FG12] This quote not only refers to the inability to work during treatment, but also for a significant period after treatment. The root of this logic lies in the regionally accepted notion that one must adhere to a number of prohibitions for six months after having been treated for HAT. Labouring is such a prohibition and is therefore considered to be strictly forbidden during the six-month rest period. These prohibitions are further elaborated below in the section on barriers to treatment.
Drug toxicity is generally considered an important barrier to participation in the active screening activities. Also in this context people do not feel compelled to participate in the screening process as long as they feel healthy and consider the risks of screening to outweigh the benefits. “Many people have died, even the one who has only been injected once. You see him die, and he wasn't even sick. People are frightened and think: ‘If I have myself tested, I might be giving up my activities for nothing and I might even die.’” [FG3]
As was previously indicated when discussing the perceived aetiology of HAT, there seems to be a degree of distrust from the community towards the nurses of the mobile teams. In several FGDs the latter were suspected of injecting the disease during the screening procedure. This idea arises from the perception that even people considered to be in good health are regularly diagnosed with the disease by the mobile teams. “The people refuse to have themselves tested because the nurses are going to inject them with the disease. They leave the insect of sleep behind through their injections, making us sick.” [FG7] Another example of this distrust which was voiced in several FGDs is the belief that an HIV test is part of the HAT screening procedure. There is a fear that one's potentially positive HIV status could be disclosed to the community, a risk people are not willing to take given the consequences. If someone is identified as HIV-positive, he or she becomes the focus of mockery and runs the risk of being rejected by the community. “They refuse to present themselves to the doctors because it is possible they are caught with AIDS. It is not unlikely as AIDS is present here.” [FG9]
The screening procedure itself was also criticized in the discussions. Especially the lack of confidentiality during the screening activities was considered to be an important issue. The procedure mostly takes place on a village square in plain sight of all those queuing up for the screening. It is considered embarrassing to be tested in public. Furthermore, people are afraid their disease status would become public knowledge. “Me, I feel shame about the possibility of being caught with this disease in public because I would be mocked. Therefore, I don't present myself for these tests.” [FG5] The lumbar puncture in open air, which is part of the screening procedure, is also considered as a significant barrier. “Others are afraid of the syringe as it hurts in the spine. We are afraid of it.” [FG5]
Cases confirmed by the mobile teams are referred to HAT treatment centres. However, also here several important barriers can be identified. The toxicity of the drugs, the financial inaccessibility, the prohibitions related to the treatment, the lacking geographic accessibility of the treatment centres, the sense of feeling healthy notwithstanding a positive diagnosis, and the fear of the regular lumbar punctures which are performed during the follow-up of the treatment are all elements which influence one's decision to seek treatment after positive diagnosis by the mobile teams.
Treatment of sleeping sickness in the region is free, but presents significant indirect costs to the patient and his escorts, mainly related to the travel to treatment centres on one hand, and nourishment on the other. “When I was caught with the disease, I left the village and left to Gandajika. I didn't have any family there, I was accompanied by my wife. When we ran out of food, we had to go back to our village for food before returning to the treatment center. This was a huge problem. That is why offering treatment in centres far away from the village brings along many difficulties to the sick. How will they get to the centre? How do they feed themselves? How can they be monitored by their relatives?” [FG13] This opinion, especially regarding the financial repercussions, was shared by many, though financial factors seemed to be less of a barrier for those living in the encampments surrounding the diamond mines.
One's perceived health state is not only a factor in deciding whether to participate in the screening activities, but also for the next step, when a confirmed HAT case is referred to a treatment centre. Some people diagnosed with sleeping sickness simply do not feel sick and do not see the need to go through the long and painful process of treatment. “They had diagnosed her with this disease. Up to today she has decided not to receive treatment and she still is as she always has been, in good health. So the people make mistakes with their tests, and therefore we can't have them done.” [FG5]
Of special interest are a number of generally accepted prohibitions linked to the treatment of sleeping sickness. Patients are expected to adhere to these prohibitions during a six month resting period after treatment. The following prohibitions, illustrated by the quotes in Table 3, were elaborated upon during the FGDs: no walking in the sun; no warm meals and hot spices; no alcohol consumption; no smoking; no heavy labour; and no sexual relations. The social and economic implications of these prohibitions form important barriers to HAT screening and treatment activities. The patient's chance of survival and the probability of making a full recovery are perceived as being directly linked to the degree to which the patients stick to the prohibitions. Treatment failure and other complications are blamed on the individual, reasoning they brought it upon themselves by not adhering to the prohibitions. Because of their importance, there is a strong element of social control involved, as the patient's entourage is mobilised to help him stick to the prohibitions. The communities are very much aware of these prohibitions. They were a recurring theme in most FGDs. Although the cited six month rest period is an existing guideline from the national HAT control programme [20], the specific prohibitions mentioned in the FGDs are not.
African sleeping sickness is well known and recognised as a serious disease in the communities of Kasai-Oriental province. In general the symptoms, vector, and treatment procedures for sleeping sickness are well known amongst the population. Fear of drug toxicity, lack of confidentiality during screening procedures and financial barriers were all elaborated upon in the FGDs as primary reasons for non-participation in the active screening activities organised by the national control programme.
These findings are in line with a similar study conducted in the province of Bandundu [16], although two additional important barriers came forward in our focus group study.
The first is the apparent incompatibility between the itineraries of the mobile screening teams and the population's livelihood activities. By the time the mobile screening teams arrive in the village, many workers have already set out to the diamond mines or crop fields. Part of this problem could be due to a lack of communication between the mobile teams and the communities. Improved planning of the screening activities to assure compatibility with the local population's habits would also be important in this respect. Factors such as daily routines and seasonal variations in the communities' activities should therefore be taken into account.
A second barrier which seems to be of particularly high importance in Kasai-Oriental is the generally accepted belief that a HAT patient must adhere to a number of prohibitions for a period of six months after receiving treatment. These hold important social and economic implications. For example, it is forbidden to perform heavy labour for the first six months after treatment. This prohibition makes it nearly impossible to make a living during this period and signifies an important loss of income. Additionally, by implying significant restrictions on participation in everyday activities, the prohibitions also lead to a degree of social exclusion and put a considerable amount of pressure on family relationships. Strong social control regarding the prohibitions is in place and victim blaming is common. When a person becomes sick, suffers a relapse or dies during the rest period, this is considered to be the result of non-adherence to the prohibitions. Given the profound impact of the six month rest period and the accompanying prohibitions, their observed role as a barrier to active HAT screening and treatment activities in Kasai-Oriental is not surprising. Even when a person is diagnosed with HAT they will refuse to go to treatment centers up until the time they become severely ill. For the national HAT control programme, a good treatment is one that is administered within 10 days after diagnosis. The problem is however that part of the population refuses to even participate in active screening activities, as they prefer not to know with certitude whether they have sleeping sickness, thus ensuring they are able to avoid treatment and all the prohibitions linked therewith. Although similar prohibitions were reported in the Bandundu study by Robays et al [16], in our study they seem to be a more profound element in the motivations of the community's actions as they constituted a recurring theme in almost all focus group discussions.
Notwithstanding the important role the prohibitions seem to take up in the communities' perception regarding HAT, their precise origin remains unclear. However, the fact that the prohibitions do not seem to have a medical basis does not make them any less significant as a barrier to HAT screening and treatment. On the contrary, a better understanding of their origin would be a first step in a process which could lead to a significantly higher participation rate in the national programme's screening activities in Kasai-Oriental. Further in-depth research is therefore necessary to document them. The focus of such research should not only be on the communities, but should also investigate the possible roles of the health workers in HAT treatment centers and the mobile teams.
The prohibitions related to nutrition may lie in traditional naturalistic or holistic beliefs about illness and health. Maintaining a form of natural balance is a central concept in such theories. Many Hispanic, African, Arab and Asian cultures incorporate elements of this basic idea in their understanding of disease. The Yin/Yang theory used in Asian cultures is a well-know example of such an approach. Another relatively common holistic framework used in traditional understanding of health and illness is based on a hot/cold dichotomy [21]–[23]. Some diseases are considered hot, whilst others are considered cold. The same dichotomy is applied to beverages, foods, herbs and medicines. Classification in either category is not necessarily based on physical characteristics of the item, but is more often according to their perceived effects on the body or their association with natural elements. There is a strong element of flexibility and subjectivity at play in the process of hot/cold classification. Regional and cultural variations in classification are therefore common. However, what is considered to be universal in this mode of thought is the practice of perceiving disease as a hot-cold imbalance which must be restored in order for the sick person to be cured. A similar logic might be at work behind the prohibitions which forbid warm meals, hot spices and possibly alcohol. If sleeping sickness is regarded as a hot disease, hot foods and beverages would be considered to upset the patient's hot/cold balance after treatment rather than allowing it to be fully restored. This imbalance, if not resolved, may then be perceived to lead to complications, relapse or even death.
It appears from our FGDs that patients often remain with neurological and psychiatric sequelae. Sleeping sickness is a chronic disease that evolves in two stages. Stage 1, or the hemolymphatic stage, is characterized by nonspecific symptoms and can even remain asymptomatic [24]. The treatment for early stage HAT is of low toxicity and recovery is complete. Unfortunately at this stage the patient often does not seek care as symptoms are mild or absent. When the patient does seek care, the non-specific symptoms are often confounded with malaria—which is also endemic in Kasai-Oriental—leading to misdiagnosis and a delay in receiving correct treatment. Stage 2 HAT is characterized by an infection of the nervous system which leads to neurological and psychiatric disorders. It is often only at this stage that a patient seeks care. The first-line treatment for second- stage HAT in DRC at the time of this study was Melarsoprol, a toxic organic compound of arsenic. Treatment with Melarsoprol can cause severe side-effects such as arsenical encephalopathy in 5 to 10% of cases, which can lead to neurological damage and death [25], [26]. In Kasai-Oriental an important population at risk consists of diamond diggers, an active portion of the population. These workers are very concerned about their professional activities to the point of ignoring their own health-related problems. As long as they have the strength to continue working in the diamond mines, they will not seek medical care, even if they are ill. Only when they become severely sick and are no longer able to work do they go to health centers. They therefore often present with well-advanced HAT infections. Treatment at such a late stage is problematic and mostly does not improve the neurological and psychiatric sequelae caused by the disease, even after cure. Melarsoprol treatment failure is high in Kasai-Oriental and—as was shown in this study—is often accredited to non-compliance to the 6 month resting period and the many prohibitions linked therewith. Patients are blamed and are themselves held responsible if they do not make a full recovery after treatment. Ultimately, the fear of possibly failing to comply with these prohibitions becomes a barrier to active HAT screening and treatment in itself.
The toxicity of the drugs used to treat HAT remains an important barrier in a person's decision to participate in the national programme's active screening activities. The adverse effects of Melarsoprol are well documented [26], and the communities are well aware of the risks associated to its use in treatment. In November 2009 the WHO started rolling out a new nifurtimox-eflornithine combination therapy (NECT) in DRC. NECT has shown very high cure rates and low adverse effect rates. Furthermore, it is relatively easy to administer, requiring only fourteen infusions over the course of ten days. Although many challenges remain [27], NECT has the potential to hugely improve the level of HAT care which is delivered to communities. However, it would be prudent to consider that it might take some time before the implementation of NECT positively influences the participation rates in the national programme's active screening activities. After all, perceptions and beliefs such as the fear of drug toxicity are deep-rooted and do not change with the wind. The legacy of melarsoprol will therefore most likely remain a barrier for some time to come. If NECT is to live up to its full potential, it is clear that communication and sustained community participatory approaches have important roles to play in the activities of national HAT control programmes, not just in the Democratic Republic of Congo, but also in all other HAT endemic countries. Additional socio-anthropological studies similar to the one reported in this paper could offer valuable insights in this respect.
|
10.1371/journal.pntd.0006562 | Diagnostic comparison between FECPAKG2 and the Kato-Katz method for analyzing soil-transmitted helminth eggs in stool | Over one billion people are infected with soil-transmitted helminths (STH), i.e. Ascaris lumbricoides, hookworm and Trichuris trichiura. For estimating drug efficacy and monitoring anthelminthic drug resistance, accurate diagnostic methods are critical. FECPAKG2 is a new remote-diagnostic tool used in veterinary medicine, which produces an image of the stool sample that can be stored on an internet cloud. We compared for the first time FECPAKG2 with the recommended Kato-Katz method.
Two stool samples were collected from adolescent participants (age 15–18 years) at baseline and 14 to 21 days after treatment in the framework of a randomized clinical trial on Pemba Island, Tanzania. Stool samples were analyzed with different diagnostic efforts: i) one or ii) two Kato-Katz thick smears from the first sample, iii) two Kato-Katz thick smears from two samples and iv) FECPAKG2 from the first sample. Parameters were calculated based on a hierarchical Bayesian egg count model.
Complete data for all diagnostic efforts were available from 615 participants at baseline and 231 hookworm-positive participants at follow-up. At baseline FECPAKG2 revealed a sensitivity of 75.6% (72.0–77.7) for detecting A. lumbricoides, 71.5% (67.4–95.3) for hookworm and 65.8% (64.9–66.2) for T. trichiura, which was significantly lower (all p<0.05) than any of the Kato-Katz methods and highly dependent on infection intensity. Despite that the egg counts based on FECPAKG2 were relatively lower compared to Kato-Katz by a ratio of 0.38 (0.32–0.43) for A. lumbricoides, 0.36 (0.33–0.40) for hookworm and 0.08 (0.07–0.09) for T. trichiura, the egg reduction rates (ERR) were correctly estimated with FECPAKG2.
The sensitivity to identify any STH infection was considerably lower for FECPAKG2 compared to Kato-Katz. Following rigorous development, FECPAKG2 might be an interesting tool with unique features for epidemiological and clinical studies.
| About 1.5 billion people are infected with soil-transmitted helminths (Ascaris lumbricoides, hookworm and Trichuris trichiura). Since morbidity correlates with the number of worms harbored by an infected individual, WHO aims to reduce moderate and heavy infections in pre- and school-aged children by 2020. The cornerstone of estimating the prevalence, assessing drug efficacy and monitoring drug resistance are accurate diagnostic tools. The currently recommended Kato-Katz, has some major disadvantages like a short processing window and low sensitivity and new diagnostic tools are needed. FECPAKG2 is an online, remote location tool developed for counting nematode eggs in sheep, cattle, equine and Camelids fecal samples. The output of the system is an image of the sample, which is saved and uploaded onto an internet cloud. This offers new options particularly for low resource settings. We tested FECPAKG2 for the first time for analyzing human stool in a randomized controlled trial. We observed a baseline sensitivity of 75.6% for detecting A. lumbricoides, 71.5% for hookworm and 65.8% for T. trichiura and an increased sensitivity for moderate infection intensities. Despite lower sensitivity and egg counts, FECPAKG2 was able to correctly estimate egg reduction rates. Following further development, FECPAKG2 might become an important tool for soil-transmitted helminth control programs, epidemiological and clinical studies.
| Approximately 1.5 billion people are infected with the soil-transmitted helminths (STH) Ascaris lumbricoides, hookworm and/or Trichuris trichiura [1]. While the majority of light infections remain asymptomatic, moderate and heavy infections are responsible for a considerable health burden, including growth stunting, intellectual retardation, cognitive and educational deficits, malnutrition and iron-deficiency anemia [2,3]. The estimated global STH burden was 3.3 million disability adjusted life-years in 2016 [4]. Large scale distribution of anthelminthic drugs (i.e. albendazole and mebendazole) to at-risk populations in preventive chemotherapy programs is the current strategy against STH infections [5]. The ultimate goal of the World Health Organization (WHO) is to reduce burden caused by moderate and heavy infections [5].
For estimating prevalence of soil-transmitted helminthiasis, assessing infection intensities, evaluating drug efficacy and monitoring drug resistance, accurate diagnostic methods are essential [5–7]. The currently recommended Kato-Katz method has already been in use for decades [8,9]. The advantages of Kato-Katz are its low cost, short sample preparation time, simple handling and the need of only basic equipment [8,10]. However, the method has a low sensitivity for low STH infection intensities, hookworm eggs disappear after one hour and samples and slides for hookworm cannot be stored [11–13]. The sensitivity can be improved by analyzing multiple Kato-Katz thick smears from several samples [12,14] or by analyzing an increased amount of stool as it is done by the FLOTAC (1 gram) or Mini-FLOTAC (2/10 gram) system [15,16].
Once the strategy is moving towards transmission control and STH elimination, an increased sensitivity of the diagnostic method of choice is crucial [6]. Nowadays, several molecular tools are available to diagnose STH infections. Although these tools show increased sensitivity, they are time consuming, require costly laboratory equipment and highly skilled laboratory technicians [17,18]. Therefore, the research on new diagnostic tools is necessary, with the aim of developing a fast, simple and cost-effective method for the diagnosis of STH infections. FECPAKG2 is an online, remote location, parasite diagnostic system used in veterinary medicine [19]. The first FECPAK system was originally established for counting nematode eggs in sheep fecal samples [20–22]. FECPAKG2 is based on the flotation-dilution principle, similar to the McMaster method [23]. The novelty of FECPAKG2 is the accumulation of parasite eggs into one viewing area within a fluid meniscus [24,25]. An image of the fecal sample is then captured, is stored offline on a computer and can be uploaded onto a cloud once connected to the internet. Subsequently, the image can be analyzed at any time by specialists around the world.
The aim of the study was to comparatively assess the sensitivity, the associated cure rates (CRs), the egg counts and their related egg reduction rates (ERR) based on FECPAKG2 and the Kato-Katz method (i.e. single, double and quadruplicate Kato-Katz). The diagnostic comparison was conducted in the framework of a clinical trial including different tribendimidine co-administrations against hookworm infections on Pemba Island, Tanzania [26].
In 2016, a randomized controlled, single-blind, non-inferiority trial evaluating the efficacy of tribendimidine co-administrations, was conducted in Tanzania and Côte d’Ivoire. The presented data on the diagnostic comparison is based exclusively on samples collected in Tanzania [26]. Ethical clearance was obtained from the Zanzibar Medical Research and Ethical Committee in Tanzania (reference ZAMREC/0001/APRIL/016) and the Ethics Committee of Northwestern and Central Switzerland (reference EKNZ UBE-15/35). This trial is registered with ISRCTN registry (number ISRCTN14373201). Written informed consent from parents or legal guardians and verbal assent from participants were obtained prior to the sample collection. At the end of the study, participants remaining positive for any STH were treated with a standard dose albendazole (400 mg) according to national guidelines [27].
The study was carried out during August and September 2016 on Pemba Island, Tanzania. Details of the clinical trial procedure are described elsewhere [26]. Briefly, adolescents (age 15 to 18) from four different secondary schools (Wingwi, Mizingani, Wesha and Tumbe) were asked to provide two stool samples at baseline. Hookworm positive participants were randomly allocated to the treatment arms: i) tribendimidine (400 mg), ii) tribendimidine (400 mg) plus ivermectin (200 μg/kg), iii) tribendimidine (400 mg) plus oxantel pamoate (25 mg/kg) and iv) albendazole (400 mg) plus oxantel pamoate (25 mg/kg). Another two stool samples were collected 14 to 21 days after treatment at the follow-up visit. Participants, laboratory and field technicians were blinded.
For each of the following diagnostic method i) one Kato-Katz thick smear of the first sample, ii) two Kato-Katz thick smears of the first sample, iii) quadruplicate Kato-Katz thick smears (two Kato-Katz thick smears of each sample) and iv) FECPAKG2 from the first sample, the sensitivity was determined for A. lumbricoides, hookworm and T. trichiura at baseline and follow-up. The sample size calculated for the clinical trial [26] was deemed sufficient for this diagnostic comparison.
A hierarchical Bayesian egg-count model as described by Bärenbold et al. [30] was applied to individual level data. The Kato-Katz counts were modelled with a negative binomial distribution depending on the daily egg density. The log of the mean egg density at the individual level was assumed to vary normally between days and the mean infection intensities to be gamma distributed in the population with a mean that reflects the mean infection intensity of an infected individual. The model was extended with a negative binomial process, to simulate the data obtained by FECPAKG2, with a linearly reduced daily egg density for the same individual compared to Kato-Katz and an independent over-dispersion parameter of the negative binomial distribution. Sample sensitivity of each test was calculated as the ratio between observed prevalence and estimated true prevalence. We assumed a specificity of more than 98% for Kato-Katz and set an uniform prior for the specificity of FECPAKG2.
The efficacy for each treatment arm in terms of CRs (percentage of egg-negative participants with a previous infection) and ERRs (percentage of arithmetic mean egg count reduction from baseline to follow-up) was calculated according to the four different diagnostic methods for all baseline positive children. CRs were calculated with imperfect diagnostic methods and an estimate for the true value based on the egg count model was given. Varying sensitivity between baseline and follow-up because of reduced infection intensity, show the following relation to the “true” CRs: (1−CRTrue)=(1−CRobserved)×sblsfu which follows from the definition of the cure rate under the assumption of no reinfections happening between baseline and follow-up (S1 Text). For the different diagnostic methods, the sensitivity-ratio between baseline and follow-up was calculated. In case the 95% confidence interval (CI) of the sensitivity-ratio included 1, the apparent CRs were not significantly different from the true CR.
Eggs per gram of stool (EPG) were calculated by multiplying the single and the average of two (duplicate) or four (quadruplicate) Kato-Katz thick smears with a factor of 24. For FECPAKG2 the egg counts were multiplied by a factor of 34. The true ERR was based on the reduction from baseline to follow-up of the mean infection intensity estimates from the model. The 95%-confidence intervals (CI) for the apparent ERRs of the treatments for each diagnostic method were obtained using a bootstrap resampling approach with 5000 replications [31].
For the statistical analysis, Stata version 14.0 (Stata Corporation; College Station; Texas, United States of America), OpenBugs version 3.2.3, Stan version 2.16.2, and R version 3.4.1 were used.
Stool samples from 1,005 participants were collected (Fig 1). Data of 391 participants were excluded: 142 provided only one stool sample, the sample of 105 participants were not analyzed with FECPAKG2 because of technical issues (ID mismatch or not sufficient stool) and FECPAKG2 images from 144 from participants were classified as insufficient quality. A total of 615 participants had complete baseline data and 384, 330 and 579 were infected with A. lumbricoides, hookworm and T. trichiura, respectively (Table 1). Only 25 participants were negative for any STH. From the participants with baseline data, 308 were treated, whereas 285 were hookworm negative and 22 were absent at treatment day. Of 308 participants randomized to treatment 13 participants were lost to follow-up,. from 21 participants the samples were not analyzed with FECPAKG2 because of technical issues and the data of 43 participants were excluded because of insufficient quality of the images. Complete follow-up data were available from 231 participants.
The estimated true baseline prevalence was 64.0% (95% confidence interval [CI] 62.2–67.1) for A. lumbricoides, 54.8% (53.1–57.9) for hookworm and 94.7% (94.0–96.0) for T. trichiura. At follow-up, prevalence values of 5.5% (4.0–8.5), 44.3% (39.4–50.5) and 52.0% (49.8–54.7) were estimated for A. lumbricoides, hookworm and T. trichiura respectively (S1 Table).
At baseline, the sensitivity of the quadruplicate Kato-Katz was significantly higher compared to any other method with 97.7% (93.1–99.9) for A. lumbricoides, 98.3% (92.7–99.9) for hookworm and 99.5% (98.1–99.9) for T. trichiura. In contrast, the sensitivity of FECPAKG2 was significantly lower than the single and duplicate Kato-Katz method (all p<0.05) with 75.6% (72.0–77.7) for detecting A. lumbricoides, 71.5% (67.4–95.3) for hookworm and 65.8% (64.9–66.2) for T. trichiura. The specificity estimated for FECPAKG2 was 96.9% (94.8–98.9) for A. lumbricoides, 91.3% (89.3–93.1) for hookworm and 95.3% (91.8–97.6) for T. trichiura. Estimated true prevalence, sensitivities, sensitivity-ratio and egg counts from the 231 participants with complete follow-up data is presented in S1 Table.
The sensitivity of FECPAKG2 was highly dependent on the infection intensity (Fig 2, S2 Table S). For an infection intensity of 100 EPG, the sensitivity of FECPAKG2 was as low as 42.9% (37.3–46.9) for A. lumbricoides, 56.3% (51.0–61.3) for hookworm and 22.2% (19.9–23.5) for T. trichiura. The estimated sensitivity increased for moderate infection intensity according to WHO cut-offs [8] and resulted in 82.0% (78.8–84.5) for A. lumbricoides (EPG 5000), 95.6% (94.1–97.3) for hookworm (EPG 2000) and 70.3% (67.6–73.9) for T. trichiura (EPG 1000).
The estimated true mean egg counts according to the model were 18125 EPG (15024–21724) for A. lumbricoides, 474 EPG (402–558) for hookworm and 1999 EPG (1762–2252) for T. trichiura at baseline (Table 1). Data from the follow up is presented in S1 Table. The EPGs based on FECPAKG2 were several times lower at baseline and follow-up compared to the different Kato-Katz sampling efforts. Relative to the Kato-Katz, the egg counts of FECPAKG2 were lower by an egg density-ratio (Fig 3, red line) of 0.38 (0.32–0.43) for A. lumbricoides, 0.36 (0.33–0.40) for hookworm and 0.08 (0.07–0.09) for T. trichiura.
The true CRs estimated by the model and the apparent CRs according to the different diagnostic methods are presented in Fig 4 and S3 Table. According to the sensitivity-ratio (SR), there was no noteworthy difference between the true estimated and the apparent CRs for the quadruplicate Kato-Katz (S3 Table). For FECPAKG2 the true estimated CRs for hookworm (SR 2.21, 1.88–2.63) and T. trichiura (SR 2.06, 1.83–2.36) differed significantly compared to the true estimated CRs. Since the CRs were generally high for A. lumbricoides (CR>93%) and most participants were cured, the sensitivity-ratio estimates had a higher uncertainty, included one and no differences among the diagnostic method were observed (SR 1.38, 0.98–2.28, S3 Table).
For tribendimidine or albendazole in combination with oxantel pamoate against hookworm, low true CRs were observed and the apparent CRs decreased with higher Kato-Katz sampling effort. The CRs according to FECPAKG2 compared to the true CRs were significantly higher for tribendimidine-oxantel pamoate (82.6%, 68.6–92.2 versus 46.3%, 35.2–52.6) and albendazole-oxantel pamoate (82.5%, 67.2–92.7 versus 49.2%, 36.7–56.2). Against T. trichiura, the difference was particularly pronounced for the treatment arm tribendimidine-ivermectin with a true CRs of 34.1% (25.7–37.7), followed by the quadruplicate (38.6%, 26.0–52.4) and duplicate Kato-Katz (50.9%, 37.3–64.4) and a significantly higher CR for FECPAKG2 (76.3%, 59.8–88.6). Similar, slightly less pronounced differences were found between the true and the FECPAKG2 CRs for tribendimidine monotherapy (5.5%, 1.6–8.5 versus 32.4%, 17.4–50.5) and tribendimidine-oxantel pamoate (66.8%, 58.1–71.1 versus 92.7%, 80.1–98.5).
No noteworthy difference was observed between the true ERRs and the arithmetic ERRs according to the four diagnostic methods (S4 Table, Fig 5). Despite lower EPGs for FECPAKG2 compared to any of the Kato-Katz methods, the ERRs and interval estimates remained similar with one exception. For tribendimidine monotherapy against T. trichiura, the true ERR (22.9%, 5.3–50.3) and the ERR determined by FECPAKG2 (29.4%, -38.3–66.7), were non-significantly higher compared to the ERRs based on the quadruplicate Kato-Katz (17.6%, -17.1–38.8).
New diagnostic tools are required to complement or replace the currently recommended Kato-Katz method [8]. FECPAKG2 is a remote-location, online parasite diagnostic system, which is used in veterinary medicine. This is the first study, which compared the FECPAKG2 method in human parasitology in the framework of a randomized, clinical trial on Pemba island, Tanzania [26]. We assessed for FECPAKG2 several different diagnostic parameters including prevalence, sensitivity and the associated CRs, egg counts, infection intensity and the related reduction in intensity after treatment.
For FECPAKG2, sensitivity was significantly lower compared to single, duplicate and quadruplicate Kato-Katz for identifying any of the STH at baseline and follow-up. However, a lower sensitivity was expected, since FECPAKG2 examines only 1/34 of gram of stool compared to 1/24 gram for the single, 1/12 gram for duplicate and 1/12 (day 1) plus 1/12 gram (day 2) for the quadruplicate Kato-Katz. For detecting moderate infection intensities, the FECPAKG2 sensitivity increased to 82.0% for A. lumbricoides, 95.6% for hookworm and 70.3% for T. trichiura. Similar characteristics have been shown for the Kato-Katz method, i.e. low sensitivity for low infection intensities and high sensitivity for moderate and heavy infections [12].
Since the CRs are a function of the sensitivity, and the sensitivity of FECPAKG2 was highly dependent on the infection intensity, the FECPAKG2 CRs and the true CRs were significantly different. For example, for tribendimidine-oxantel pamoate the T. trichiura infection intensity changed from baseline (true EPG~2000) to follow-up (true EPG~100), which led to a decreased sensitivity from 80.5% (baseline) to 22.2% (follow-up,). Therefore, the CR for FECPAKG2 (92.7%) was significantly overestimated compared to the true CR (66.8%) (S3 Table). These results indicate, that in the present form FECPAKG2 does not accurately estimate CRs, which was also true for the single and duplicate Kato-Katz.
While the lower sensitivity negatively influenced the CRs, the ERRs remained unchanged, which was already reported by Levecke and colleagues for different Kato-Katz sampling efforts [32]. Similarly, no differences among the diagnostic methods were shown in our study. For instance, the above-mentioned treatment example resulted in a true ERR of 94.3%, which was not significantly different from an ERR of 95.7% with FECPAKG2(S4 Table). While the egg counts with FECPAKG2 were generally lower compared to Kato-Katz, the ERRs remained equal. Thus, FECPAKG2 might be an interesting tool for monitoring anthelmintic drug efficacy [5].
A lower egg recovery rate from sheep or cattle fecal samples was already observed for the earlier FECPAK system in comparison with FLOTAC, Mini-FLOTAC and McMaster, however, no data about the performance of the new FECPAKG2 was available [20,21]. The lower recovery of eggs by FECPAKG2 might be due to the inability of detecting unfertilized A. lumbricoides eggs and a high extent of debris covering the eggs. To overcome the problem with high debris, a variety of different sized meshes for the FECPAKG2 cylinder are currently being tested. In addition, in the FECPAKG2 cassette the capillary rise of the aqueous saline generates an axisymmetric meniscus over the cylindrical rod, which converges the eggs on the top of the meniscus [29]. The accumulated eggs remain in a single microscopic field of view and a staged image of the meniscus is taken with the MICRO-I. For increasing the recovery, a vibration function in the MICRO-I might improve the egg accumulation, as suggested by Sowerby and colleagues [29]. Further optical and image processing improvements for the MICRO-I are under development. These improvements will speed up the processing capability of the device and will generate higher quality images that are expected to improve the egg recovery (sensitivity) and accuracy of the image mark-up.
Obviously, the examination of only one cassette and one stool sample with FECPAKG2 was a limitation of our study. The collection of two stool samples would account for the day-to-day variation and would increase sensitivity [30]. For example, in this study the sensitivity increased from one analyzed stool sample (single or duplicate Kato-Katz) to two stool samples (quadruplicate Kato-Katz) about 10%-points for A. lumbricoides and hookworm. The sensitivity-ratio indicated a weak dependence of the quadruplicate Kato-Katz on infection intensities, which did not induce a significant bias for this study, since the sample size was rather small and precision estimates wide. Nevertheless, the bias might become important in larger studies with higher accuracy. By collecting samples on several days, the sensitivity of FECPAKG2 for low infection intensities might improve, which would limit the bias introduced in CR estimates. Hence, the analysis of two cassettes and two stool samples with FECPAKG2, should be the subject of further studies. Additionally, the time for preparing one sample and the costs of FECPAKG2 should be compared against current established diagnostic methods.
Other limitations of this study were the loss of samples due to the mixing up of sample IDs, insufficient amount of stool and insufficient quality of many FECPAKG2 images. In more detail, a total of 144 (19.0%) samples at baseline and 43 (14.0%) samples at follow-up were excluded, because of insufficient filling of the cassette or problems associated with the capturing of the image (i.e. blurriness, stacking bands, cracked rods, debris, air bubbles etc.), which was detected only during the mark-up process of the images when sample analysis could not be repeated. With lower numbers of analyzed samples per day, larger number of laboratory technicians, better experience with handling of the FECPAKG2 the number of excluded samples might have been lower and hence these factors should be considered in future studies.
Despite the discussed limitations of FECPAKG2 at the current stage of development, several advantages are worth highlighting. The most innovative feature is the captured image, which is saved offline, uploaded online onto an internet cloud and analyzed at any later time point. In contrast, the major limitation of Kato-Katz is the disappearance of hookworm eggs one hour after the preparation [13]. Moreover, stool samples cannot be stored [11], which limits the time to control the diagnostic quality [28]. The storage of the FECPAKG2 images offers new options, especially for low resource settings. First, diagnostic results of STH can be stored for the first time, analyzed by trained technicians around the world and quality control is not restricted to time. Second, technicians can focus on processing the samples while analysis is done at a later time point, potentially leading to a faster turnaround in laboratories. Third, in case of identification discrepancies, specialist around the world can be consulted, which improves the diagnostic results. Research is ongoing to develop an image-analysis algorithm, which will automatically count the different helminth eggs in the future.
In conclusion, we have assessed for the first time the performance of FECPAKG2 in human parasitology, in the framework of a randomized controlled trial. Compared to different Kato-Katz sampling efforts, FECPAKG2 showed lower sensitivities and egg recovery rates. The sensitivity increased with higher infection intensities. Further research is required for increasing sensitivity and egg recovery to develop FECPAKG2 as a useful addition in the near future to the depleted diagnostic set of tools for STH infections.
|
10.1371/journal.ppat.1003016 | The Stringent Response of Staphylococcus aureus and Its Impact on Survival after Phagocytosis through the Induction of Intracellular PSMs Expression | The stringent response is initiated by rapid (p)ppGpp synthesis, which leads to a profound reprogramming of gene expression in most bacteria. The stringent phenotype seems to be species specific and may be mediated by fundamentally different molecular mechanisms. In Staphylococcus aureus, (p)ppGpp synthesis upon amino acid deprivation is achieved through the synthase domain of the bifunctional enzyme RSH (RelA/SpoT homolog). In several firmicutes, a direct link between stringent response and the CodY regulon was proposed. Wild-type strain HG001, rshSyn, codY and rshSyn, codY double mutants were analyzed by transcriptome analysis to delineate different consequences of RSH-dependent (p)ppGpp synthesis after induction of the stringent response by amino-acid deprivation. Under these conditions genes coding for major components of the protein synthesis machinery and nucleotide metabolism were down-regulated only in rsh positive strains. Genes which became activated upon (p)ppGpp induction are mostly regulated indirectly via de-repression of the GTP-responsive repressor CodY. Only seven genes, including those coding for the cytotoxic phenol-soluble modulins (PSMs), were found to be up-regulated via RSH independently of CodY. qtRT-PCR analyses of hallmark genes of the stringent response indicate that an RSH activating stringent condition is induced after uptake of S. aureus in human polymorphonuclear neutrophils (PMNs). The RSH activity in turn is crucial for intracellular expression of psms. Accordingly, rshSyn and rshSyn, codY mutants were less able to survive after phagocytosis similar to psm mutants. Intraphagosomal induction of psmα1-4 and/or psmβ1,2 could complement the survival of the rshSyn mutant. Thus, an active RSH synthase is required for intracellular psm expression which contributes to survival after phagocytosis.
| The stringent response is a bacterial response to a multitude of different environmental stress conditions which is characterized by the synthesis of the messenger molecules (p)ppGpp. There is now growing evidence that these molecules also play a key role for pathogens to switch between specific phenotypic states within the host. This seems crucial for the adaptation to different microenvironments encountered during infection for instance after uptake by phagocytes. Killing of phagocytes as well as survival within these cells was proposed as major mechanisms for the success of the human pathogen Staphylococcus aureus to spread within the body. In the current study we demonstrate the effect of the stringent response on global gene expression in S. aureus and its impact on intracellular survival in human neutrophils. We reveal that a stringent response is induced after uptake of S. aureus in neutrophils and RSH activity is crucial for intracellular induction of psm expression, coding for cytotoxic phenol-soluble modulins (PSMs). Finally we show that this in turn mediates bacterial survival and escape after phagocytosis. These findings contribute to the understanding of how and where PSMs can act as potent cytolytic molecules and emphasise the importance of (p)ppGpp as an intracellular signalling molecule.
| In most bacteria, nutrient limitations provoke the so-called stringent response, which is initiated by the rapid synthesis of the alarmones pppGpp and/or ppGpp, here referred to as (p)ppGpp. Under stringent conditions, (p)ppGpp results in the shut-down of proliferation-related activities, including the transcriptional repression of genes coding for major components of the protein synthesis apparatus (rRNA, ribosomal proteins and translation factors) as well as the inhibition of replication [1], [2], [3]. Typically genes that are presumed to be important for maintenance and stress-defence are activated under stringent conditions. However, the stringent phenotype resulting from (p)ppGpp synthesis seems to be bacteria species specific and may be mediated by fundamentally different molecular mechanisms [3].
The molecular mechanisms leading to the profound reprogramming of the bacterial cellular machinery under stringent conditions were mostly studied in Escherichia coli. Here, (p)ppGpp can be synthesized by either one of the two homologous enzymes: RelA and SpoT. The RelA-synthase is activated by sensing uncharged tRNAs that are bound to the ribosome. SpoT is a bifunctional enzyme that not only produces (p)ppGpp in response to diverse signals but also contains a (p)ppGpp hydrolase domain important for (p)ppGpp turnover. In E. coli, (p)ppGpp binds, with the help of the DksA protein, directly to the RNA polymerase (RNAP). However, even in this model organism, there is still much debate concerning how (p)ppGpp eventually leads to different promoter activities, how (p)ppGpp influences the stability of open complex formation at the initial phase of transcription and which of the promoters are indirectly regulated via secondary regulatory circuits such as alternative sigma factors or other transcription factors [1].
In other organisms, such as the gram-positive Bacillus subtilis, (p)ppGpp probably does not interact with the RNAP, and no DksA homologue is present. Here (p)ppGpp has been proposed to affect promoter activities only indirectly via changes of the intracellular nucleotide pool [4], [5], [6]. In this model, the nature of the initiation nucleotide (iNTP) determines whether genes are under positive or negative stringent response [4], [5], [7], [8]. (p)ppGpp synthesis is usually accompanied by a decrease in intracellular GTP concentration. rRNA promoters in B. subtilis initiate with GTP and a change of this base at position +1 results in a loss of regulation by (p)ppGpp and GTP. Furthermore, GTP can act as a co-factor for the repressor CodY, and thus the lower GTP levels imposed by the stringent response result in the de-repression of CodY target genes at least in some firmicutes, e.g., B. subtilis [9] or Listeria monocytogenes [10].
In most firmicutes, three genes coding for putative (p)ppGpp synthases are present. The bifunctional RSH (RelA/SpoT homologue) enzymes are typically composed of a C-terminal sensing domain and an N-terminal enzymatic domain with hydrolase and synthase functions. RelQ and RelP (also named SAS1 and SAS2, for single small alarmone synthase) are small proteins with only a putative (p)ppGpp synthase domain [11], [12].
Little is known about the stringent response in the human pathogen Staphylococcus aureus, most likely because of the essentiality of the major (p)ppGpp synthase/hydrolase enzyme RSH [13], [14]. Previously, we have shown that, at least under nutrient rich conditions, mutants defective only in the synthase domain of RSH (rshSyn) are not impaired in growth [13]. Furthermore, RSH is the only enzyme responsible for (p)ppGpp synthesis in response to amino acid starvation triggered by leucine/valine deprivation or mupirocin treatment. Preliminary characterisation of the rshSyn mutant revealed that part of its phenotype mainly the regulation of genes involved in amino-acid metabolism can be explained by the (p)ppGpp induced de-repression of the CodY regulon. CodY of S. aureus was previously shown to be a major regulator of virulence gene expression [15], [16]. Thus, the CodY regulon seems to be an integral part of the stringent response in S. aureus linking metabolic circuits and virulence. However, whether, and to what extent, other direct or indirect regulatory circuits are involved in RSH mediated stringent response is unclear. A CodY independent contribution of the stringent response to virulence could, so far, not be indicated. However, for many other pathogenic bacteria, a contribution of (p)ppGpp to bacterial virulence could be shown [17]. In particular, for many intracellular bacteria, (p)ppGpp is essential to survive and replicate in diverse host cells, i.e., epithelial and endothelial cell types and professional phagocytes.
Here, we delineate three (category I–III) main consequences of RSH-dependent stringent response in S. aureus. (I): Similar to other organisms, a severe down-regulation of the protein synthesis machinery was observed, and thus, the inhibitory effects of (p)ppGpp on gene expression seem to be highly conserved in bacterial species. (II): In contrast, genes that are activated upon (p)ppGpp induction are highly species specific. In S. aureus, they are mostly regulated indirectly via the de-repression of CodY. (III): Interestingly, seven genes, including those coding for the toxic phenol-soluble modulins (PSMs), were found to be up-regulated independently of CodY. We demonstrate that a stringent response is induced after uptake of wild-type bacteria in human neutrophils. Moreover, RSH activity is required for psmα1-4 and psmβ1-2 expression in neutrophils which in turn promotes survival after phagocytosis.
S. aureus is able to mount a stringent response upon amino acid starvation, which is characterized by the generation of (p)ppGpp [18], [19]. However, little is known about the impact of (p)ppGpp synthesis on global physiological processes in S. aureus. Mupirocin treatment, an antimicrobial agent that inhibits the bacterial isoleucyl tRNA synthetase and therefore mimics isoleucine starvation, results in profound reprogramming of gene expression [20], [21]. Parts of these effects are mediated by (p)ppGpp. However, at least some of them were also observed in an RSH-mutant, which is devoid of (p)ppGpp synthesis after mupirocin treatment [13].
Here, we aimed to delineate the impact of (p)ppGpp synthesis induced by amino acid starvation on gene expression in S. aureus under defined nutrient limitation rather than after the addition of inhibitors such as mupirocin. Previously we could show that transferring S. aureus into a chemical defined medium (CDM) lacking the two branched-chain amino acids leucine and valine results in growth inhibition [13]. A similar experimental setup was chosen here for a further in-depth analysis including transcriptional profiling by microarray: Bacteria were grown to mid-exponential growth phase in complete CDM, filtered and shifted into a medium lacking leucine and valine (−leu/val) (Fig. 1A, open arrow). After 30 minutes of incubation in CDM −leu/val the bacteria were harvested for RNA purification (Fig. 1A, filled arrow). At this time all strains showed a slight growth inhibition compared to bacteria grown in complete CDM, which is accompanied by a clear induction of the stringent response indicated by RSH dependent (p)ppGpp synthesis and reduction of the GTP pool [13]. The rshSyn mutant showed no accumulation of (p)ppGpp under these conditions which indicates that a contribution of the two putative (p)ppGpp synthases RelP and RelQ is negligible.
To analyze the effects mediated by (p)ppGpp synthesis we compared the transcriptional profile of wild-type strain HG001 with that of the rshSyn mutant under the indicated stringent conditions. Previously a close link between the stringent response and the activity of the repressor CodY was demonstrated in different firmicutes [10], [13], [22], [23]. Thus, to delineate which and how many (p)ppGpp regulated genes are influenced through CodY we also analyzed the RSH mediated stringent response in codY-negative S. aureus.
In accordance to previous results [13] we could confirm by northern blot analysis that under the selected conditions of leucine/valine starvation the wild-type strain showed a significant repression of genes of the translational apparatus, such as rpsB (ribosomal protein S2), infB (translation initiation factor B) and tsf (translation stable factor), whereas no repression occurred in the rshSyn mutant (data not shown). By contrast, typical CodY target genes like brnQ1 coding for an amino acid transporter were induced in the wild-type strain (Fig. 1B). The low expression of brnQ1 in the rshSyn mutant however, could be restored by additional mutation of codY, indicating that the rshSyn mutation affects brnQ1 transcription through CodY (Fig. 1B). To analyze whether this observation is due to a general response to amino acid limitation, we induced serine starvation by addition of serine hydroxamate (SHX). The same transcriptional pattern was observed when compared to conditions of leucine/valine starvation (−leu/val) (Fig. 1B).
Based on the results of the northern blot hybridizations, we performed microarray analysis of wild-type S. aureus HG001, rshSyn, codY and rshSyn, codY double mutants after leucine/valine depletion (Fig. 1A). Three major categories (I–III) of stringently regulated genes were observed: I) genes that are negatively regulated by (p)ppGpp independent of CodY. II) genes that are positively influenced by (p)ppGpp through CodY de-repression and III) genes positively influenced by (p)ppGpp independent of CodY.
Microarray analysis revealed that 102 genes were significantly (p-value cut-off, 0.05) down-regulated in the wild-type compared to the rshSyn mutant in the codY positive (WT<rshSyn) as well as in the codY negative background (codY<rshSyn, codY) (Fig. 2A, overlap blue Venn diagram). This indicates that these genes are negatively regulated by (p)ppGpp independent of CodY. Since in total 161 genes were significantly down-regulated in the codY positive background, we also verified the 59 missing genes in the codY negative background. All of these genes appeared also to be down-regulated, although the difference did not reach the significance level.
Most of these CodY-independent negatively regulated genes are hallmark genes of the stringent response in other bacteria [24], [25], [26]. They are mostly involved in the translation process (Fig. 2B), such as coding for ribosomal proteins (rps, rpl), for translation initiation factor (infB) and translation elongation factor (tuf) (Fig. 2C). The rRNA processing machinery is also influenced since genes coding for ribonuclease P (rnpA) and rimM (coding for a processing protein) are repressed in an RSH-dependent manner (Fig. 2C). In addition, genes like DNA helicases (dnaB, C, D), part of the so called “replicon”, are negatively influenced by (p)ppGpp, and genes of the recombination/repair system (ruvAB and dinP) were also repressed under amino acid starvation due to (p)ppGpp (Fig. 2B,C). In accordance with published data from E. coli and B. subtilis [24], [25], genes of other physiological processes that are typical for dividing cells (nucleotide, lipid and coenzyme biosynthesis and transport, inorganic ion transport and protein modification) are also part of the negative stringent response in S. aureus (Fig. 2B,C).
In contrast to B. subtilis [24], E. coli [25], or S. pneumoniae [26], genes involved in cell division and cell wall biosynthesis were not found to be significantly influenced by the stringent response in S. aureus. However, we can not exclude that such genes are affected at later time points.
We aimed to obtain a comprehensive overview of genes that are part of the predicted regulatory overlap between stringent response and CodY repression. Under stringent conditions most of the up-regulated genes were expressed only in the codY- positive background (Fig. 2A). An additional introduction of a codY mutation into this analysis (codY vs. rshSyn, codY) abrogated the up-regulation of these genes (Fig. 2C). Thus, under amino acid deprivation many genes are de-repressed in the wild-type bacteria (WT>rshSyn) through the relief of CodY repression, but stay constitutively repressed via CodY in the rshSyn mutant. These genes are part of the recently described CodY regulon [15], [16] mainly involved in amino acid metabolism and transport (Fig. 2B). Accordingly, most of them possess CodY binding motifs as described elsewhere [16] and as indicated in Fig. 2C (three asterisks). It was shown previously that CodY represses also several virulence genes in part via inhibition of the quorum sensing system agr. Of note, none of these codY influenced virulence genes were found to be significantly down-regulated in the rshSyn mutants.
Next, we investigated if there are genes activated by (p)ppGpp with no contribution of CodY. Microarray analysis revealed that only seven genes were expressed significantly lower in the rshSyn mutants in both, a codY-positive (codY+) and codY-negative background (codY−) (Fig. 2A, table 1). One gene coding for a putative ribosome-associated protein Y was described in E. coli to bind to the small ribosomal subunit and to stabilize ribosomes against dissociation when bacteria experience environmental stress [27]. Also traP was expressed significantly lower in the rshSyn mutants. The function of TraP remains unknown since a proposed involvement of this protein in the activation of the agr system was recently disproved [28], [29], [30]. Surprisingly, two genes, part of the staphylococcal defence mechanism, coding for β1 and β2 phenol-soluble modulins (psm β1/β2) were transcribed significantly lower in both rshSyn mutants. This result was of special interest because no other obvious virulence gene appeared to be part of the stringent response in S. aureus.
The cytolytic activity of β-type PSMs was described to be minor, and their role in virulence is less pronounced compared with α-type PSMs [31]. Since no probes for α-type PSMs are present on the microarray chip, we performed northern blot hybridisation to analyze the RSH-dependent expression of these important molecules (Fig. 3). To exclude strain specific effects, rshSyn mutants of the prototypic S. aureus strain Newman were also included in the analysis. After leu/val starvation, there is no or very little, α- and β-type psm transcription detectable in the rshSyn mutants of strain HG001 and strain Newman (Fig. 3A) in contrast to the wild-type strain, which shows strong induction of both psm classes (Fig. 3B). Introducing full-length rsh chromosomally into the rshSyn mutant strains could restore their deficiencies to induce psms expression (Fig. 3A). An involvement of CodY could be excluded since the rshSyn, codY double mutant shows the same low transcription of psms as the rshSyn single mutant.
Queck et al. [32] could show that psm expression is directly regulated via binding of the response regulator AgrA to the psm promoter region. Therefore, we analyzed whether agrA transcription is altered under the conditions tested. No significant transcriptional differences were detectable comparing wild-type strain and the corresponding mutants under leu/val deprivation (Fig. 3A/B). The response regulator AgrA also directly activates the transcription of the divergently transcribed regulatory RNAIII which in turn results in down-stream regulatory effects on several virulence genes. In the microarray analysis, no significant difference in RNAIII or other prototypic RNAIII target genes such as hla or spa were observed. To exclude possible AgrA defects, we screened the mutants for altered haemolytic activity [29]. No haemolytic deficiencies could be detected in the mutant strains (data not shown). Therefore, an AgrA defect in the rshSyn mutants can be excluded. Nonetheless agr mutants exhibit no psm transcriptions under leu/val deprivation (Fig. 3A), supporting that agr activity is still essential for psm expression under stringent conditions. However, RSH dependent induction of psmα1-4 and psmβ1,2 under stringent conditions is not mediated by agr activation.
Aside from the requirement of functional agr, little is known about psm regulation in S. aureus. The mechanism by which stringent response leads to increased transcription of certain genes is much under debate. In B. subtilis, (p)ppGpp does not appear to interact directly with RNA polymerase (RNAP) [5]. The nature of the iNTP during transcription was proposed to determine whether genes are under positive or negative stringent response. In this model, positive stringent controlled genes of B. subtilis are characterized by iATP, and these genes are presumably activated because of an increased ATP pool [4]. Previously, the transcriptional starting points of the α- and ß-psms were identified as an adenine ribonucleotide [32]. To analyze whether induction of phenol-soluble modulins in the wild-type strain correlates with an increased ATP pool, the intracellular ATP concentration was measured (Fig. 4). A significant increase of the intracellular ATP pool was detectable upon amino acid deprivation in all strains. However, no significant difference was found between the strains analyzed (Fig. 4). Thus, the ATP levels cannot explain the observed differences in psm expression between wild- type and rshSyn mutant.
Since α-type PSMs are strongly involved in the survival of S. aureus upon PMN treatment [31], [33], we analyzed whether there is an association between phagocytosis and the RSH-dependent stringent response. Comparison of published data revealed that the transcription pattern during neutrophil phagocytosis (Voyich et al., 2005) is similar to that observed after amino acid starvation and/or mupirocin treatment [13], [20], [21], namely down-regulation of the translational machinery and de-repression of CodY regulated amino acid transporters and biosynthesis genes. Hence, we hypothesized that phagocytosis may induce an RSH-dependent stringent response, similar to the depletion of amino acids. Under these conditions, the (p)ppGpp accumulation should result in increased PSMs synthesis. Thus, an rshSyn mutant, unable to transcribe psms, might be more sensitive to phagosomal killing. Indeed, the PMN bactericidal killing assay revealed that the rshSyn mutant is significantly less able to survive after phagocytosis compared to the wild-type strains (Fig. 5A). This effect could be reversed by the introduction of full-length rsh into the chromosome of the rshSyn mutants. A comparison of the two strains (HG001 and Newman) with their respective mutant derivatives revealed similar results. Also an assumed contribution of relP and relQ could be excluded, since a relP/Q double mutant showed no decreased survival phenotype compared to the wild-type strain (data not shown).
To further analyze whether reduced intracellular psm expression solely accounted for the rshSyn defect in survival, we constructed a psmα1-4, psmβ1,2 double mutant and an agr mutant, known to be defective in psm expression. The phenotype of the psmα, psmβ double mutant and the agr mutant closely resemble that of the rshSyn mutant concerning intracellular survival after phagocytosis (Fig. 5B). Of note, there were no significant differences between the psmα, psmβ double and the rshSyn, psmα, psmβ triple mutant, as well as between the agr single and the rshSyn, agr double mutant, indicating that there are no additive effects of these mutations. We speculate that survival of phagocytosis is mediated by intracellular psm expression mediated by RSH dependent (p)ppGpp synthesis.
To address this assumption, we used a plasmid, in which psmα1-4 and psmβ1,2 were cloned behind a tetracycline inducible promoter, to induce psmα or psmβ expression in the rshSyn mutant after phagosomal uptake. The intracellular induction of psmα and psmβ could significantly complement the reduced survival of the rshSyn mutant (Fig. 5C). Interestingly, the survival of the rshSyn, agr double mutant was not complementable, neither by psmα nor by psmβ expression.
We also performed PMN lysis assays by measuring the amount of released lactate dehydrogenase (LDH). Accordingly the mutants which are defective in psms expression were shown to be less toxic compared to the wild type strain (Fig. 5D).
These results indicate that the RSH-mediated stringent response in S. aureus plays a major role in survival PMN phagocytosis most probably through regulation of intracellular psm expression.
To analyze whether an RSH-dependent stringent response is induced during phagocytosis, we performed quantitative RT-PCRs from bacterial RNA obtained after 60 and 90 minutes of phagocytosis. At this stage, typically all of the bacteria are internalized [34]. We analyzed transcripts of typical genes of each of the three regulatory categories relative to the constitutive gyrB transcript (Fig. 6). For category I, the transcripts of rpsB, coding for ribosomal protein S2, and infB, coding for initiation factor 2 were chosen. As expected, the rpsB and infB transcription is significantly lower in the wild-type and complemented strain than in the rshSyn mutants (rshSyn single and rshSyn, codY double mutant) (Fig. 6). In contrast, ilvC transcription (category II gene) is significantly higher in the wild-type compared to the rshSyn mutant. Here, comparable to results of the microarray analysis, an additional codY mutation in the rshSyn mutant could restore ilvC transcription. For category III, psmα1-4 and psmβ1,2 transcripts were analyzed. The wild- type and the complemented rshSyn strain showed significantly higher transcription compared to rshSyn mutants after incubation for 60 or 90 min with phagocytes. These results are consistent with the assumption that an RSH-mediated stringent response is induced after phagocytosis in wild-type bacteria. Of note, low expression of psms within the PMNs was also observed in the rshSyn, codY double mutant, indicating that intracellular psm regulation occurs independently of CodY.
One may speculate that (p)ppGpp affect somehow the activity of the response regulator AgrA in the phagolysosomes thereby activating psm expression. Thus, we analyzed the intracellular expression of the prototypic AgrA target transcript RNAIII. However, no significant alteration of RNAIII transcription could be detected in the rshSyn mutants after 60 and 90 minutes of incubation (Fig. 6). In fact, the rshSyn, codY double mutant showed a slight increase of the RNAIII transcription compared to wild-type strain, which is in line with published data showing that agr transcription is slightly repressed by CodY [15], [35]. These results indicate that stringent response is induced during phagocytosis which is required for psm expression. The reduced transcription of psmα1-4 in the rshSyn mutant and rshSyn, codY double mutant after phagocytosis is clearly not caused by diminished AgrA activity.
To adapt to changing environmental conditions, bacteria rely on sensory and regulatory systems to modulate complex physiological processes. The stringent response is a highly conserved regulatory mechanism that is provoked by nutrient limitation. It is effective in most bacteria and is mediated by the rapid synthesis of the alarmones ppGpp or pppGpp. Previously, it was shown that, in S. aureus, the RSH enzyme alone is responsible for (p)ppGpp accumulation upon amino acid deprivation [13]. Here, we characterized the stringent response of S. aureus in a more comprehensive way. So far, no microarray studies, in which (p)ppGpp synthase mutants were compared with wild-type strains under stringent conditions, are available for S. aureus. For other bacteria only a few microarrays of this type were published [24], [25], [26], [36], [37]. We identified 161 genes that were significantly repressed via RSH. Most of these genes are hallmarks of the stringent response also appearing in other bacteria that have been analyzed. This conserved “core-regulon” mainly encompasses genes of the translational machinery, including initiation and elongation factors. Furthermore, genes coding enzymes of physiological processes used by dividing cells, like biosynthesis and nucleotide transport, are typically under negative stringent response. Thus, negative regulation by (p)ppGpp seems to be an evolutionarily conserved mechanism. However, the molecular mechanism by which (p)ppGpp leads to the inhibition of these genes remains mostly unclear but seems to be organism dependent [4], [5], [38], [39]. For firmicutes it was proposed that genes starting with an iGTP are repressed due to the lowering of the GTP pool [4]. We have shown previously that also in S. aureus stringent condition lead to lowering of the GTP level [13]. We have mapped the transcriptional start site of two rRNA operons of S. aureus (data not shown) and could confirm that the primary promoters initiate with iGTP which is in line with the mechanism proposed by Krasny et al., 2008 [4].
Induction of the stringent response also leads to the activation of genes proposed to be necessary for survival and maintenance. For E. coli, it was recently shown that most genes that are positively influenced during the stringent response are, depending on the (p)ppGpp concentration, either part of the Lrp (transcriptional regulator) or RpoS (alternative sigma factor) regulon [40]. So far, there is little indication that alternative sigma factors are involved in the stringent response in gram-positive bacteria. Instead in S. aureus, most of the genes that are less expressed in the rshSyn mutant compared with the wild-type strain are part of the previously described CodY regulon [15], [16]. For these genes, introduction of a codY mutation into the rshSyn mutant leads to an expression pattern that is similar to that of the wild-type strain. Thus, these genes are de-repressed upon the stringent response. An obvious link between the stringent response and CodY is the GTP pool. Amino acid deprivation leads to a lowering of the GTP pool [13], which in turn may inactivate the CodY repressor. A similar link between the stringent response and CodY was also shown for Listeria monocytogenes [10] and B. subtilis [24]. In contrast, in Streptococci, CodY and the stringent response seem to act independently of each other [23], [41]. This discrepancy is most likely due to species specific differences in GTP affinity of CodY [42]. In lactococci and streptococci branched-chain amino acids (e.g. isoleucine), act as the only known ligand that mediates the repressive function of CodY.
Notably, not all genes of the known CodY regulon seem to be affected under the induced stringent condition, analyzed here. Interestingly, virulence genes like the capsular gene cluster or the agr operon, which were found to be repressed via CodY [15], [16], did not appear to be affected during the stringent response. One explanation might be that the deactivation of the CodY repressor in the wild-type strain due to amino acid limitation is not as effective as knocking out the complete codY gene. Alternatively, the different CodY target genes may differ with regard to their sensitivity to CodY activity. One may assume that some of the CodY target genes are still repressed even under low GTP conditions as long as the primary CodY ligand isoleucine is present. This assumption is inline with recent findings that in B. subtilis CodY targets are differentially sensitive to alteration in the GTP pool: the ilv promoters but not the bcaP promoter was derepressed in a mutant not able to synthesize GTP [43].
Only seven genes were positively regulated under the stringent response independently of CodY, including those coding for the cytotoxic phenol-soluble modulins (PSMs). So far psm transcription was shown to be directly activated via the response regulator AgrA [32]. We could show that AgrA is still essential for the stringent controlled psmα1-4 and psmβ1,2 transcription, since no psms were detectable in an agr mutant strain. However, the induction appears not to be due to altered agrA activity, since agrA expression was not found to be affected during the stringent response. Also after phagocytosis only psm transcripts but not the Agr effector molecule RNAIII are diminished in the rshSyn mutants. How (p)ppGpp regulates the transcription of PSMs remains unknown. Additional regulatory elements upstream of the promoter region of psm genes [32] could possibly take part in the (p)ppGpp regulation, presumably mediated by unknown factor(s). In our microarray analysis, no apparent regulatory systems appeared to be affected, and thus, no candidate molecule that may indirectly be involved in (p)ppGpp dependent activation could be identified. Small molecules like (p)ppGpp or GTP may interact directly with proteins altering their enzymatic activity or binding affinities to DNA similar to the proposed GTP-CodY interaction. One may speculate that such an interaction may also alter the activity of other regulatory molecules such as the transcriptional regulator AgrA (Fig. 7).
Alternatively, it was postulated that under stringent conditions, the iNTP determines the transcription of the gene followed [4]. psm genes were shown to start with an iATP [32]. In B. subtilis, genes starting with iATP were predicted to be up-regulated after amino acid limitation through an increase of the intracellular ATP pool [4], [7]. However, this increase in ATP is obviously independent of the RSH-dependent (p)ppGpp synthesis, since we found a similar increase in the rshSyn mutant which is in line with results obtained in B. subtilis (Tojo et al., 2008). Thus, increasing the ATP pool is not sufficient to trigger induction of the PSMs since no enhanced psm transcription was observed in the rshSyn mutant despite a similarly elevated ATP pool than the wild-type and rshSyn complemented strain.
Our results indicate, by taking the example of a gene coding for a ribosomal protein S2 (rpsB) and initiation factor 2 (infB), that the down-regulation of translational proteins after phagocytosis is mediated via rsh activity (Fig. 6). Together, with the lower expression of ilvC, psmα1-4 and psmβ1,2 in the rshSyn mutants, these findings suggest that the stringent response is induced after phagocytosis (Fig. 7). This result is supported by the global transcriptional analyses of the S. aureus strains after uptake by professional [34] and non-professional [44] phagocytes. A transcription pattern very similar to those described, was found here especially down-regulation of typical hallmark genes of the stringent response, i.e., ribosomal proteins and up-regulation of amino acid biosynthesis genes.
However, the intracellular signal for an RSH mediated stringent response remains speculative. For S. aureus, so far, only amino acid limitation was shown to induce (p)ppGpp synthesis via RSH. We also analyzed other conditions such as glucose starvation as potential signals for RSH dependent (p)ppGpp synthesis but found no detectable (p)ppGpp accumulation (data not shown). Thus, these findings suggest that amino acid limitation triggers the response within phagocytes. Moreover, in L. monocytogenes, a gram-positive intracellular pathogen, the amino acid limitation sensing RSH was shown to be essential for survival and replication in macrophages [10]. However, this finding is in contradiction to results obtained from gram-negative intracellular pathogens. Here, the authors concluded that no amino acid limitation occurs in infected host cells. This conclusion is based on the finding that RelA, the enzyme solely responsible for accumulation of (p)ppGpp upon amino acid limitation in gram-negative bacteria, is not activated [17], [45], [46], [47], [48]. For these bacteria, the bifunctional (p)ppGpp synthase SpoT seems to be more important for adaption to the intracellular environment. It was presumed that balancing the basal levels of (p)ppGpp (due to the hydrolase and synthase activity of SpoT) is essential for intracellular adaptation [46]. There may be major differences between gram-negative and gram-positive bacteria concerning the mechanisms of signalling and responses to intracellular environments. Interestingly, treatment with azurophilic granule proteins [49] imposed transcriptional changes similar to those of the stringent response. It has to be elucidated whether parts of the reactive oxygen species (ROS) in phagolysosomes may be a trigger for the stringent response in S. aureus.
The predominant role of S. aureus PSMs in killing human neutrophils was previously shown [31]. The reduced survival of the rshSyn mutant after phagocytosis is probably due to a decreased expression of PSMs within the phagolysosome. No other candidate virulence gene which could potentially account for this observation appeared in our microarray analysis to be down-regulated in the rshSyn mutant. Comparing the survival of an rshSyn mutant to a psmα, psmβ double mutant or an agr mutant reveals a similar low survival rate which is in line with published data [33]. The distinct role of psms expression for survival could be demonstrated by complementation assays of the rshSyn mutant with tetracycline induced psmα1-4 or psmβ1,2. The fact that the reduced survival of an rshSyn, agr double mutant was not complementable by psm expression indicates that probably additional agr dependent factors are needed to survive after phagocytosis.
Little is known about the intracellular behaviour of S. aureus. Growing evidence suggests that S. aureus can survive after phagocytosis and persists in neutrophils or macrophages to hide from the immune system, as well as to travel to and infect distant sites in the host [50], [51], [52]. However, the importance of single virulence factors contributing to intracellular survival seems to be dependent on the type of host cell and bacterial strains analyzed. For instance, S. aureus can escape from the phagoendosomes of human epithelial and endothelial cells into the cytosol, which is mediated by δ-toxin, β-toxin and ß-PSMs [53]. PSMs may similarly contribute to escape from the more toxic phagolysosomes of neutrophils and thereby allow intracellular survival. Whether they also contribute to a lysis of phagocytes from the inside remains to be shown. This was recently proposed based on observations that PSMs are efficiently inhibited by human serum and therefore lyse neutrophils rather from the inside than from the outside [54]. These authors could also show that psm promoter activity is strongly induced after phagosomal uptake. Moreover there could be an indirect impact of PSMs on the lysis of phagocytes. Previously it could be shown that different clinical isolates of strain USA300 induced a programmed necrosis of PMNs [55]. One may assume that intracellular expressed PSMs thereby could play an important role for this induction.
The results we obtained are significantly important for the question how and where PSMs can act as potent cytolytic molecules The data of the current study support the conclusion that in phagolysosomes a stringent response is activated and that the activity of the (p)ppGpp synthase RSH is essential for the intracellular induction of psmα1-4 and psmβ1,2 expression. These cytolytic peptides in turn are responsible for the ability of S. aureus to survive after phagocytosis (Fig. 7). The signalling mechanisms leading to the activation of the (p)ppGpp synthases as well as the mechanism leading to gene activation through (p)ppGpp after phagocytosis is largely unknown and needs further examination.
Strains and plasmids are listed in Table S1. S. aureus strains were grown in CYPG (10 g/l casamino acids, 10 g/l yeast extract, 5 g/l NaCl, 0.5% glucose and 0.06 M phosphoglycerate) [56] or in a chemically defined medium (CDM) [15]. For strains carrying tetracycline, erythromycin or chloramphenicol resistance genes, antibiotics were used only in precultures at concentration of 5 µg/ml for tetracycline and 10 µg/ml for erythromycin and chloramphenicol, respectively. Bacteria from an overnight culture were diluted to an initial optical density at OD600 of 0.05 in 25 ml fresh medium using 100 ml baffled flasks and grown with shaking (220 rpm) at 37°C to the desired growth phase. For down-shift experiments, strains were grown in complete CDM including leucine/valine (leu/val) to an OD600 of 0.5. The cultures were filtered over a 0.22 µm filter applying vacuum, washed twice with sterile phosphate buffered saline (PBS) and bacteria were re-suspended in an equal volume of CDM medium with or without leu/val and grown for another 30 minutes. For serine starvation experiments, bacteria were re-suspended in an equal volume of CDM medium containing serine hydroxamate (SHX, 1.5 mg/ml) and incubated for 30 minutes.
The haemolysis test was performed as described previously [29]. Briefly, bacteria to be tested are streaked at a right angle to RN4220 and the plate was incubated overnight. ß-hemolysin of strain RN4220 and δ-hemolysin of strains to be tested form a zone of clear haemolysis (synergistic effect) on blood agar plates.
The rshSyn, codY double mutant was obtained by transducing the codY::tet(K) mutation into S. aureus strain Newman rshSyn [13] using Φ11 lysates of strains RN4220-21 [15]. Transductants were verified by PCR and PFGE. For complementation the full length rsh gene with a 960 bp upstream region was transduced into strain Newman rshSyn (Table S1) using Φ11 lysates of strain CYL316-199 [13].
The psmα and psmβ mutants were obtained by replacing psmα1-4 with a tetracycline resistance cassette and psmβ1-2 with an erythromycin resistance cassette using a newly developed temperature-sensitive shuttle vector pBASE6. This vector is based on the previously described pBT2 vector [57] with the additional advantage of counter-selection against the plasmid by inducible expression of S. aureus secY antisense RNA of the pKOR1 vector [58]. Therefore the HindIII-Bst1107I fragment of pBT2 was replaced by the 4.875 kb EcoRV-HindIII (partial digest) fragment of pKOR1, containing the tetR/secY regulatory unit. Since some of the restriction sites of the pBT2 multiple cloning site (MCS) are also present in the introduced pKOR1 part, the MCS was removed by EcoRI-HindIII digestion. Primers: mcsmod1 (ATTCCGGAGCTCGGTACCCGGGCTAGCGCGCAGATCTGTCGACGATATCA) and mcsmod2 (AGCTTGATATCGTCGACAGATCTGCGCGCTAGCCCGGGTACCGAGCTCCGG) were mixed in equimolar amounts, heated to 95°C and slowly cooled down to room temperature. This new MCS contains only unique restriction sites and was ligated into the EcoRI-HindIII digested vector, resulting in pBASE6.
For gene replacements, two fragments flanking the pmsα1-4, psmβ1-2 locus, the tetracycline and erythromycin resistant cassette were amplified and annealed by overlapping PCR to generate the pmsα1-4-tetM and psmβ1,2-ermC mutagenesis vectors pCG307 and pCG308, respectively. The amplicons were cloned into the BglII/SalI restriction sites of pBASE6. These plasmids were used to mutagenize strain RN4220 as described previously [13]. The obtained psm gene replacement mutant strains (RN4220-307 and RN4220-308) were verified by PCR. In the mutants the whole psmα1-4 operon respectively psmβ1,2 operon was replaced by the corresponding resistant cassette. The rshSyn, psmα, psmβ triple mutant was obtained by transducing the psmβ::erm(C) and psmα::tet(M) mutations into S. aureus strain HG001 rshSyn using lysates of strain RN4220-307 and RN4220-308.
RNA isolation for microarray analysis and northern blot analysis was performed as described previously [59]. Briefly, bacteria were lysed in 1 ml of Trizol reagent (Invitrogen Life Technologies, Karlsruhe, Germany) with 0.5 ml zirconia-silica beads (0.1 mm-diameter) in a high-speed homogenizer (Savant Instruments, Farmingdale, NY). RNA was isolated as described in the instructions provided by the manufacturer of Trizol.
RNA isolation after phagocytosis was performed as described in the instructions provided by the manufacturer of the RNA isolation kit (ExpressArt RNAready, AmpTec) with the modification adding an inhibitor removal buffer (high pure viral nucleic acid kit, Roche diagnostics) in the first step. Then DNA digestion was performed as instructed by the RNA isolation kit.
Northern blot analyses were performed as described previously [59]. Digoxigenin-labeled probes for the detection of specific transcripts were generated using a DIG-Labeling PCR Kit following the manufacturer's instructions (Roche Biochemicals). Oligonucleotides were used for probe generation as described previously [13], [15], [60] or are listed in Table S2.
Relative quantifications of α-type psms, β-type psms, infB, rpsB, RNAIII and gyrB transcripts were performed using LightCycler instrument (Roche). Briefly, RNA isolated from cultures after phagocytosis (60 and 90 min) was transcribed into complementary DNA using SuperScriptIII Reverse Transcriptase (Invitrogen) and 200 ng of random hexamer primers (Fermentas). Complementary DNA was diluted 1∶5 and quantitative real-time PCR was performed using the QuantiFast SYBR Green PCR Kit (Qiagen). A standard curve for each gene was generated using 5-fold serial dilutions of wild-type HG001 cDNA at timepoint 0 h (oligonucleotides see Table S2). Statistical analysis was performed with the Prism software package (version 5.0; GraphPad) using the Student t test two-tailed analysis (p<0.05).
S. aureus wild type HG001, isogenic mutants and the complemented strain were grown in CDM to OD600 = 0.5. Cells were shifted to CDM with and without leu/val as described above. Samples for intracellular ATP analysis were harvested 30 min after the shift by fast filtration over a 0.22 µm sterility filter applying vacuum. Cells were washed, quenched and nucleotides were extracted as described recently [61], [62]. The extracts were resuspended in 5 ml of 0.1 M Tris-acetate buffer (pH 7.75) and stored at −80°C. The detection of ATP was performed by the Enlighten ATP assay system using luciferase and luciferin (Promega). Therefore nucleotide extracts were diluted 1∶100 in Tris-acetate buffer and 10 µl mixed with 90 µl of the ATP assay reagents. Luminescence was measured with a luminescence detection reader (Infinite M200Pro, Tecan, Austria). The standard curve was generated by using known concentrations of ATP. The results of the intracellular ATP measurements represent the mean of 2 biological replicates measured in triplicates. Statistical analysis was performed with the Prism software package (version 5.0; GraphPad) using the Student t test two-tailed analysis (p<0.05).
The microarray was manufactured by in situ synthesis of 60-base-long oligonucleotide probes (Agilent, Palo Alto, CA), selected as previously described [63]. The array covers >98% of all open reading frames (ORFs) annotated in strains N315 and Mu50, MW2, COL, NCTC8325 and USA300, and MRSA252 and MSSA476, as well as Newman, including their respective plasmids.
Total RNA was purified from strain HG001 WT, rshSyn mutant, codY mutant and rshSyn, codY double mutant grown in CDM to an OD600 of 0.5. For each strain RNA of three independently grown cultures was analyzed. After additional DNase treatment, the absence of remaining DNA traces was confirmed by quantitative PCR (SDS 7700; Applied Biosystems, Framingham, MA) with assays specific for 16S rRNA [15]. Batches of 5 µg of total S. aureus RNA were labeled by Cy3-dCTP using SuperScript II (Invitrogen, Basel, Switzerland) following the manufacturer's instructions. Labeled products were then purified onto QiaQuick columns (Qiagen). Purified genomic DNA from the different sequenced strains used for the design of the microarray was extracted (DNeasy; Qiagen), labeled with Cy5 dCTP using the Klenow fragment of DNA polymerase I (BioPrime, Invitrogen, Carlsbad, CA), and used for the normalization process [64] Cy5-labeled DNA (500 ng) and a Cy3-labeled cDNA mixture were diluted in 50 µl of Agilent hybridization buffer and hybridized at a temperature of 60°C for 17 h in a dedicated hybridization oven (Robbins Scientific, Sunnyvale, CA). Slides were washed, dried under nitrogen flow, and scanned (Agilent, Palo Alto, CA) using 100% photon multiplier tube power for both wavelengths.
Fluorescence intensities were extracted using Feature Extraction software (version 9; Agilent). Local background-subtracted signals were corrected for unequal dye incorporation or unequal load of the labeled product. The algorithm consisted of a rank consistency filter and a curve fit using the default LOWESS (locally weighted linear regression) method. Data consisting of three independent biological experiments were expressed as log 10 ratios and analyzed using GeneSpring, version 8.0 (Silicon Genetics, Redwood City, CA). A filter was applied to select oligonucleotides mapping ORFs in the Newman genome, yielding approximately 95% coverage. Statistical significance of differentially expressed genes was calculated by analysis of variance [65] using GeneSpring, including the Benjamini and Hochberg false discovery rate correction of 5% (p value cutoff, 0.05) and an arbitrary cutoff of twofold for expression ratios.
The complete microarray data set has been posted on the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE99340 for the platform design and GPL7137 for the original data set.
Killing by human neutrophils was performed as described previously [66]. Briefly, bacteria of the logarithmic phase were washed and adjusted in potassium phosphate buffer (10 mM K2PO4). Neutrophils were isolated from peripheral blood of healthy volunteers by ficoll/histopaque gradient centrifugation as described previously [67] and resuspended in HBSS-HSA (hank's balanced salt solution, Sigma, containing 0.05% human serum albumin). Bacteria were opsonized by addition of pooled human serum (Sigma) to a final concentration of 10%. Opsonized bacteria (107/ml) and neutrophils (106/ml) were combined to a volume of 500 µl and samples were shaken at 37°C. For the psm complementation assays, neutrophils and bacteria were preincubated for 30 minutes without anhydrotetracycline (ATc) followed by addition of 0,1 µg/ml ACT for another 30 minutes. Aliquots were diluted in ice-cold water and vortexed vigorously to disrupt the neutrophils and halt bacterial killing. Appropriate dilutions were plated on tryptic soy agar plates and incubated at 37°C for the following day for enumeration of CFU. The percent of bacterial survival was calculated with the equation CFU+PMN at t60/CFU+PMN at t0. Statistical analysis was performed with the Prism software package (version 5.0; GraphPad) using the Student t test two-tailed analysis (p<0.05).
Following phagocytosis of S. aureus, lysis of human neutrophils was determined with a standard assay for release of lactate dehydrogenase (LDH) as described by the manufacturer (Cytotoxicity Detection kit; Roche Applied Sciences). Statistics were performed with the Prism software package (version 5.0; GraphPad) using the Student t test two-tailed analysis (p<0.05).
|
10.1371/journal.ppat.1004834 | Genetic Analysis Using an Isogenic Mating Pair of Aspergillus fumigatus Identifies Azole Resistance Genes and Lack of MAT Locus’s Role in Virulence | Invasive aspergillosis (IA) due to Aspergillus fumigatus is a major cause of mortality in immunocompromised patients. The discovery of highly fertile strains of A. fumigatus opened the possibility to merge classical and contemporary genetics to address key questions about this pathogen. The merger involves sexual recombination, selection of desired traits, and genomics to identify any associated loci. We constructed a highly fertile isogenic pair of A. fumigatus strains with opposite mating types and used them to investigate whether mating type is associated with virulence and to find the genetic loci involved in azole resistance. The pair was made isogenic by 9 successive backcross cycles of the foundational strain AFB62 (MAT1-1) with a highly fertile (MAT1-2) progeny. Genome sequencing showed that the F9 MAT1-2 progeny was essentially identical to the AFB62. The survival curves of animals infected with either strain in three different animal models showed no significant difference, suggesting that virulence in A. fumigatus was not associated with mating type. We then employed a relatively inexpensive, yet highly powerful strategy to identify genomic loci associated with azole resistance. We used traditional in vitro drug selection accompanied by classical sexual crosses of azole-sensitive with resistant isogenic strains. The offspring were plated under varying drug concentrations and pools of resulting colonies were analyzed by whole genome sequencing. We found that variants in 5 genes contributed to azole resistance, including mutations in erg11A (cyp51A), as well as multi-drug transporters, erg25, and in HMG-CoA reductase. The results demonstrated that with minimal investment into the sequencing of three pools from a cross of interest, the variation(s) that contribute any phenotype can be identified with nucleotide resolution. This approach can be applied to multiple areas of interest in A. fumigatus or other heterothallic pathogens, especially for virulence associated traits.
| Invasive aspergillosis (IA) caused by Aspergillus fumigatus is increasing due to medical interventions that suppress the ability of patients’ immune systems to control infections. These invasive lung infections are difficult to diagnose and consequently treatment is frequently not started promptly. Some controversy surrounds the role of mating type in virulence of A. fumigatus and the emergence of azole resistant strains has posed difficult challenges for clinical management of IA. We generated nearly identical A. fumigatus strains with opposite mating types that allowed us to test whether different mating types have different virulence profiles. We found no difference in virulence in three different animal models, which suggests that mating type does not influence virulence. We also took advantage of the essentially identical genomes of both strains to apply classical genetic approaches combined with genomics technologies to identify A. fumigatus genes that contribute to azole resistance. We performed genetic crosses of azole sensitive with azole resistant strains and analyzed the resistance status and genome composition of the offspring. Using this approach we cataloged several genes that were not previously associated with azole resistance. This information will be valuable for finding ways to manage azole resistance in IA patients.
| In spite of the significant progress made in diagnosis and treatment of patients with invasive aspergillosis (IA), the number of patients that succumb to the infection is still unacceptably high. Early diagnosis and treatment that can eradicate the fungus without further compromising the patients’ health remain as the most pressing challenges faced in the treatment of IA [1]. In the pathobiology of A. fumigatus, the predominant etiological species of IA, remarkable progress has been made since sequencing of its genome [2,3]. Moreover, the recently discovered heterothallic sexual cycle in the species [4] offers an unique opportunity to study genes relevant for virulence or drug resistance using genetic recombination [5].
Following the description of the sexual cycle in A. fumigatus [4], we described a pair of A. fumigatus supermaters that produce unusually large numbers of cleistothecia containing viable progeny in 4 weeks [5]. Even though the supermater pair can be used in an array of recombinational analysis, studies focusing on the association of genomic sequences and phenotype can be better addressed when the genomes of two mating strains are identical except for their mating type and sequences of interest. For instance, in the investigation of mating type and virulence in A. fumigatus it is critical to use isogenic strains to normalize the genetic background and target differences specific to the phenotype. Generation of isogenic strains and their use for genetic analysis enabled extensive progress in the molecular pathobiology of Cryptococcus neoformans including the role of mating type in fungal virulence [6–9]. In A. fumigatus, some studies suggested a predominance of MAT1-1 strains among cases of IA and a more severe virulence phenotype for MAT1-1 compared to MAT1-2 isolates in the larvae of Galleria mellonella [10,11]. Comparisons in these studies, however, have been made between MAT1-1 and MAT1-2 isolates of unknown genetic background from clinical and/or environmental sources. Because of the many factors associated with the pathobiology of A. fumigatus, it is difficult to specifically target the contribution of mating type in the virulence of such isolates with heterogeneous genetic background.
Increasingly important issues in clinical management of IA are the general limitation in treatment options—only three classes of drugs are available—and the ever-increasing problem of drug resistance [12]. Systemic aspergillosis infections are treated with amphotericin B, azoles [13], or caspofungin. The low efficacy of the available antifungal drugs requires a long duration of treatment that frequently leads to resistance for azoles and caspofungin [12,14]. Antifungals exploit differences between mammalian and fungal cells to kill the fungal organisms without adverse effects on the host. Drugs used to treat fungal infections target only two differences between human and fungal cells: the presence of ergosterol in fungal cell membranes and glucans in their cell walls [13]. Azole antifungals (e.g. itraconazole, posaconazole, voriconazole, fluconazole) inhibit an enzymatic step in the ergosterol biosynthetic pathway. Prevalent and well-studied resistance mechanisms in bacteria, fungi, and parasites include target modification and efflux from the cell. However, there remains an urgent need to understand the broad range of genes encoded in the genomes of fungal pathogens that participate in the resistance to the clinically therapeutic antifungals employed in treating infections.
Whole-genome sequencing of resistant strains was recently used to identify novel drug-resistance mechanisms in Escherichia coli [15], Candida glabrata [16], and Plasmodium falciparum [17], among others. A. fumigatus is a haploid fungus and colonies grown in vitro have average 5–7 nuclei in each hyphal cell [18] which makes the species very amenable to sequencing-based screen as it reduces the complications of diploid organism. Camps et al. used sexual crosses between clinically-isolated azole-resistant and sensitive strains of Aspergillus fumigatus to follow mutations that conferred resistance [19]. They identified mutations in the hapE gene, and demonstrated that sexual crossing is a powerful mechanism to confirm the role of specific mutations in resistance.
In this study we generated an isogenic pair of A. fumigatus with identical genomes except for the mating locus and the absence of a single 28 Kb region in one mating type. The isogenic pair was used in virulence and drug resistance studies. Virulence studies in animal models indicated that the mating type of A. fumigatus is not associated with virulence phenotypes. We then employed azole sensitive and resistant variants of the isogenic strains in classical genetic crosses to follow mutations that were linked with the resistance phenotype in order to more clearly elucidate which mutations actually confer resistance. The isogenic pair, AFB62 and AFB62F9, represents a powerful tool to study the pathobiology of A. fumigatus.
The A. fumigatus strains AFB62 (MAT1-1), a clinical isolate, and AFIR928 (MAT1-2), an environmental isolate, were chosen as starting parental strains because of their high fertility [5]. The MAT1-1 parent strain, AFB62, was chosen as the foundational strain for backcrossing. A highly fertile MAT1-2 progeny from a cross between AFB62 and AFIR928 was backcrossed with AFB62 and the resulting progeny of MAT1-2 mating type was backcrossed again with AFB62. After 4 generations of backcrossing, the genomes of an AFB62F4 (MAT1-2) strain and the foundational strain AFB62 were compared by comparative genome hybridizations (CGH). Of the genes in the AF293 microarray, the results showed that AFB62F4 and AFB62 were nearly isogenic with only a 7-gene difference between the strains (S1 Table). In contrast, the two starting parental strains had 86 absent or divergent genes [5]. In order to reach a more isogenic state, back-crossings were continued for seven additional generations (F11). A highly fertile ninth generation progeny, AFB62F9, was selected for use throughout the rest of this study since the fertility declined significantly after F9. The ninth generation isogenic pair AFB62F9 and AFB62 was able to produce viable ascospores in 4 weeks and the ascospore viability was on average 13%, which is comparable to 17% found for the supermaters AFI928 and AFB62.
To examine if the isogenic pair progeny exhibited Mendelian inheritance, we generated a mutant of AFB62 producing brown conidia by the deletion of the gene abr2 (Afu2g17530) [5] and a mutant of AFB62F9 producing white (albino) conidia by deletion of the alb1/pksP (Afu2g17600) gene. Mating of these two strains generated progeny that produced white, brown, and green conidia (S1 Fig). Since the abr2 and alb1 loci are both located on the chromosome 2, the presence of green conidia denotes reconstitution of the wild type genotype by crossing-over between the mutated loci. Independent assortment of the mating-type alleles located on chromosome 3 and the conidial color genes on chromosome 2 was confirmed by the observation of near 1:1 ratio of MAT1-1 to MAT1-2 in the progeny and 2.5–7% frequency of the wild type recombinants producing green colored conidia (S1 Fig). These findings showed that fertility and recombination of the isogenic pair were not significantly different from the supermater pair, AFB62 and AFIR928 (5–10% green conidial recombinants) [5], confirming the usefulness of the isogenic pair for genetic analysis.
The genomes of AFB62 and AFB62F9 were sequenced to determine the extent of their isogenicity. The AFB62 genome consisted of 27,882,634 bp and was assembled into 895 contigs ranging from 502 to 351,863 bp, with N50 of 78,619 bp. Comparisons between AFB62 and the genomes of A. fumigatus reference strains AF293 and A1163, indicated that AFB62 is more closely related to A1163 with many fewer SNPs (51,765 SNP and 30,263 SNP, respectively). The majority of the genomic differences between AFB62 and the two reference strains were concentrated in the sub-telomeric regions of the chromosomes or in genomic islands, which was similar to the previous observations between different A. fumigatus isolates [20]. Over 94% (8,789 genes) of AFB62 genes had homologs in AF293 and A1163. Most of the AFB62-unique genes (371 out of 493) had no assigned function (i.e. hypothetical), encoded for proteins with generic transmembrane domains, or belonged to one of 26 different GO terms with no significant enrichment in any one term (S2 Table). It is important to note, however, that the quality of the AFB62 genome assembly is significantly lower than either of the two reference genomes which can artificially inflate the total number of predicted open reading frames, and consequently, overestimate the number of unique genes.
The AFB62F9 sequencing reads were mapped to the AFB62 reference genome in order to identify differences and similarities between the two isogenic strains. Over 99.5% of the sequencing reads were mapped to the AFB62 reference for which almost all the residues (99.91%) were covered by AFB62F9 reads, showing that AFB62F9 is identical to AFB62 except for the MAT1-2 instead of MAT1-1 mating locus (Fig 1A) and the absence of a 28 Kb region located about 200 Kb away from the mating locus of AFB62 on chromosome 3 (S2 Fig). A close inspection of the AFB62F9 reads mapped to the chromosomal region encompassing the mating locus and the 28 Kb locus (~290 Kb) showed a high number of point mutations (e.g. SNP) in the AFB62F9 genome compared to the AFB62 genome. A total of 413 SNP were located in this region, which was almost half of all identified point mutations that differentiated the isogenic strains (849 total). The SNP density for this region was 14.2 SNP /10,000 bp, which was much larger than the rest of the genome with an average of 0.05 SNP/10,000 bp. This SNP density is similar to the SNP density obtained when different strains of A. fumigatus are compared to each other [21]. Taken together, the high SNP density and the absence of a 28 Kb fragment show that the MAT1-2 locus in AFB62F9 was inherited from the AFIR928 parent through recombination of this relatively large fragment (Figs 1B and S2). A non-related SNP-dense region ~400 Kb was also found on chromosome 6 (AF293 coordinates 645,096 to 1,045,729) containing 287 SNP (7.2 SNP/10,000 bp), suggesting that this region too was inherited as a single block from the AFIR928 parent. This region encoded a sister-chromatid segregation protein homolog (src1) an adenylate cyclase (acyA), and two transcription factors including TFIID. In AFB62F9, the promoter region of this gene had a 4 base pair deletion and a G→A substitution that could affect its transcriptional regulation. The acyA region in AFB62F9 contained 9 SNP including a non-synonymous substitution (P91H), which could potentially lead to different activity of this protein. Lastly, a frame shift mutation in TFIID complex could also have a significant impact in the transcriptional response of AFB62F9 compared to AFB62.
The two parental lines, AFB62 and AFIR928, are equally virulent in mouse models and have similar virulence to other isolates such as B-5233, a reference strain of A. fumigatus [5]. However, due to their other genomic differences, it was not possible to determine whether the mating locus played a specific role in pathobiology of A. fumigatus. In this study, the virulence between AFB62 and AFB62F9 strains was first compared using CGD mice that are susceptible to A. fumigatus due to an impaired production of reactive oxygen species by the NADPH oxidase complex [22,23]. No significant difference was found between the survival curves of mice inoculated with AFB62 and that of mice inoculated with AFB62F9 suggesting these two strains possess similar virulence in CDG mice (Fig 2A). Histological sections of lungs inoculated with AFB62 and AFB62F9 showed similar lesions with micronodular bronchopneumonia affecting approximately 50–75% of the parenchyma (Fig 3A and 3C). Branching septate hyphae admixed with moderate number of neutrophils was observed in the smaller bronchioles and adjacent alveoli and moderate necrosis was observed in alveolar septae. Larger airways and bronchial epithelium were relatively spared of inflammatory infiltrate and fungal hyphae. Additional virulence assays carried out with two MAT1-2 progeny of the 7th generation confirmed the findings with the 9th generation progeny.
Virulence of AFB62 and AFB62F9 was then tested in BALB/c mice immunosuppressed with hydrocortisone acetate. No significant difference was found between the survival curves of the mice inoculated with AFB62 and AFB62F9 (Fig 2B). Histology of lungs isolated 7 days post inoculation showed that both strains caused similar lesions: severe necrotizing bronchopneumonia with large numbers of septate branching hyphae with approximately 25–33% of the parenchyma being affected; prominent necrosis of the bronchial/bronchiolar epithelium; bronchial/bronchiolar lumina presented hyphae admixed with neutrophilic cell infiltrate, necrotic cells and fibrin (Fig 3E–3H).
The last model used to compare virulence between AFB62 and AFB62F9 was larvae of Galleria mellonella. This model has been used in several studies of virulence in A. fumigatus, including comparison between several random MAT1-1 and MAT1-2 strains of clinical and environmental origin [11,24–26]. Larvae inoculated with AFB62 died at the same rate as those inoculated with AFB62F9 (Fig 2C). Similar results were obtained when larvae were inoculated with five MAT1-1 and five MAT1-2 F10 progenies isolated from the cross between AFB62 and AFB62F9. The results of the virulence assays carried out in three different experimental models indicated MAT1-1 and MAT1-2 mating type strains in the AFB62 genomic background possess similar ability to grow invasively in lungs (Fig 3B and 3D), cause similar pathology of the disease and are equipotent in causing host death, suggesting that the MAT locus does not play a significant role in virulence of A. fumigatus.
We next undertook a study to identify mutations in the A. fumigatus genome that could result in resistance to itraconazole, posaconazole, and voriconazole. We subjected the isogenic pair, AFB62 and AFB62F9, to increasing concentrations of each of three azoles individually in order to select for such mutations (Fig 4A). Previous studies have shown that incubation in sub-lethal concentration of antimicrobials enhances the rate of spontaneous mutations that confer resistance [27]. Thus, we first determined the minimum inhibitory concentrations (MIC) of itraconazole, voriconazole, and posaconazole for both isogenic strains using E-strips, to allow us to ‘prime’ the strains with advantageous mutations prior to selection. Both strains were susceptible to the tested azoles (S3 Table) and fell within the observed MIC for other A. fumigatus strains [28]. Each isogenic strain was incubated in sub-inhibitory concentrations (0.5 – 1X MIC) of each antifungal on MEA plates for 72 h at which point three isolates from each strain selected under each azole were plated for single spore isolation. Each of the three isolates from media containing each azole was then transferred to a new plate containing 1 – 3X the inhibitory concentration (S3 Table) for 72 h and three new resistant colonies were selected for single spore isolation. Each of the three single spores was passaged one additional time under 10X MIC to select highly resistant mutants. Interestingly, incubation at 10X MIC for each lineage resulted in lawns, suggesting that the resistance mechanisms obtained under 1 – 3X MIC were sufficient for resistance at 10X. Thus, in order to isolate lineages with mutations that contributed to higher azole resistance, we increased the drug concentration empirically by 2X until incubation allowed less than 100 colonies per plate. The highest drug concentration was at minimum 12X for voriconazole and maximum of 78X MIC for posaconazole. Again, three isolates of each lineage were picked and single isolates were selected (Fig 4A). The resistance profiles of all single isolates were confirmed using E-test.
To identify mutations and other genomic differences between the resistant strains and wild-type parents, we sequenced and compared the genomic sequences from 15 intermediate (>1X) and 21 high concentration (>12X) isolates with its parental strain (Table 1). The isolates were selected to represent all the combinations of drug and parental lineages while maximizing the read depth obtained, thus not all 72 possible isolates were sequenced. Derivative strains selected under the highest drug concentrations contained between 1 and 13 SNPs (Tables 1 and S4). Seventy-one percent (15 of 21) of the drug resistant derivatives had at least one mutation in the erg11A gene [29] and 38% (8 of 21) in the 3-hydroxy-3-methylglutaryl coenzyme A (HMG CoA) reductase gene (hmg1) (Tables 1 and S4). No azole-resistant strains had the tandem duplication in the promoter of erg11A commonly associated with the L398H mutation [12]. No apparent preference for resistance mechanisms was observed in either isogenic parental strain, suggesting that the differences in genetic background or mating type did not have an effect on resistance mechanisms. A detailed description of mutations associated with each azole follows.
We took advantage of the mating ability of the isogenic pair to determine the association of particular mutations with drug resistance phenotype. To this end, we crossed the three highly azole-resistant isolates selected above from each intermediate resistance lineage from both parents (AFB62 and AFB62F9) in each drug with their drug sensitive mating partner (Fig 4B). The most fertile crosses, abundant cleistothecia with viable ascospores (>2000 fungal balls), were chosen for progeny sequencing (S5 Table). Ascospores from the selected crosses were grown on varying levels of the each azoles: no drug, intermediate (1 – 3X MIC), and high (10X) drug concentration (Fig 4B). The logic behind this approach is that mutations with no association to drug resistance will be equally present in the mating strains under no drug selection and any drug concentration, while those SNPs that confer resistance to the drug will be found more frequently under intermediate selection, and perhaps exclusively under high selection. Therefore, after selection, the genomes of the resulting pool of offspring were determined and the frequencies of each of the mutations from the resistant strains were tracked through each of the drug concentrations (no drug, 1X MIC, 10X MIC). As a control, we analyzed the frequency of MAT1-1 and MAT1-2 in the pool because these loci are not associated with resistance and would therefore be expected at a 1:1 frequency with or without drugs.
We found that all the erg11A gene mutations demonstrated the expected patterns in offspring pools (Table 2). In general, for each progeny under no selection, both the sensitive and resistant alleles could be detected in the population at similar frequencies (1: 1 ratio, or 0.5 / 0.5 allele frequency). In contrast, only the resistant allele was detected at either the intermediate or high drug concentrations in all posaconazole and voriconazole crosses (Table 2; allele frequency of 1). For voriconazole, the hmg1 alleles followed the same patterns as the erg11A resistance alleles, showing that these mutations were also under voriconazole selection, and must therefore contribute to resistance. In contrast, the mating loci were usually found at a 1:1 ratio in each pool. No other mutations in voriconazole or posaconazole isolates could be associated with the resistance phenotype.
Interestingly, the erg11A mutant alleles followed the same distribution when the itraconazole-resistant AFB62 parent was mated with the sensitive F9 derivative (Table 2). However, when the resistant F9 was crossed with the sensitive AFB62, the erg11A resistance allele could no longer be detected. Furthermore, two of the F9-itraconazole resistant strains did not harbor mutations in the erg11A or hmg1 genes, suggesting that these strains had alternate mechanisms of resistance. The mutation in ssc70 present in both of the high-resistant parents in these crosses also showed association with the drug resistance phenotype, suggesting that in these lineages, the ssc70 mutation was sufficient to confer itraconazole-resistance. There were no other mutations that followed the expected pattern in these pools, thus showing that in addition to the two genes described for posaconazole and voriconazole (erg11A and hmg1), ssc70 also played an important role in azole resistance. Thus, from the original SNPs identified in any given resistant strain (up to 13) we could narrow down the specific resistance-conferring mutation to only 1 or 2 in any strain, and to a total of 5 genes across all strains.
In order to confirm the role of our newly identified mutations in azole resistance, we attempted to make mutations in hmg1 and ssc70 genes to recapitulate the phenotype. We succeeded in generating mutant hmg1P1038T which had the same mutation present in strain AFB62V3 and its high-level voriconazole resistant derivatives AFB62V31 and AFB62V33. Strain hmg1P1038T was resistant to 4X the MIC of the wild-type strain for voriconazole. We tested whether hmg1P1038T was also resistant to the other azoles. Indeed, the strain was more resistant to itraconazole with 4X higher MIC, and 2X MIC posaconazole resistance (Table 3). The MIC for strain AFB62V31, that had the erg11 mutation in addition to the hmg1 mutation, was even higher (16X) suggesting that the combination of mutations conferred high-level resistance to all azoles. The data also showed that the spontaneous mutant AFB62F9I22 carrying an ssc70 mutation was more resistant to posaconazole and voriconazole in addition to itraconazole compared to the naïve AFB62F9 (Table 3). Despite multiple attempts to generate an ssc70 mutant, we were not successful, suggesting that this gene may be essential and/or not be amenable to genetic manipulation. In fact, the SSC1 gene in S. cerevisae, the ortholog of A. fumigatus SSC70, is known to be an essential gene [33].
Combined, these results demonstrate that implementing whole-genome sequencing coupled with classical genetic approaches can identify genetic loci responsible for a particular phenotype with unparalleled detail.
The discovery of the A. fumigatus sexual cycle [4] opened a road to genetic studies by recombinational analysis in this pathogen. However, the six-month incubation period required for the completion of sexual reproduction hampered its wide use as a genetic tool. Our subsequent identification of a supermater pair which can complete sexual cycle in 4 weeks [5] made the use of sexual cycle more tractable in studies of the pathobiology of A. fumigatus. To further enhance the use of this mating pair as genetic tool, we generated a MAT1-2 strain isogenic to AFB62 (MAT1-1) strain by successive backcrossing of MAT1-2 progeny obtained in each generation to the foundational strain AFB62. The genome of a ninth generation MAT1-2 progeny (AFB62F9) was found to be essentially identical to that of AFB62, except for the mating locus, the absence of a single 28Kb region, and 849SNP, thus providing a mating pair in which there is very little genetic diversity. The 28 Kb, which contains a reverse transcriptase and a bacteriodetes-like chitinase among others, appears to be unique to AFB62, as it is absent in 7 other sequenced strains of A. fumigatus [20]. We have backcrossed the strains until the 11th generation to improve the degree of isogenicity even further. However, generations beyond F9 showed lower and an inconsistent degree of fertility which would be unsuitable as a genetic tool for recombination analysis. Since fertility of the AFB62 x AFB62F9 mating was comparable to the supermater pair AFB62 x AFIR928 [5], and the 28 Kb fragment is unique to AFB62, this 28 Kb sequence may play a role in its high fertility. The molecular dissection of the 28 Kb sequence is warranted to understand its role in fertility.
The sexual cycle in aspergilli requires an array of genes to ensure successful reproduction. A genome wide comparison of three aspergilli, A. fumigatus (strain AF293), A. nidulans, and A. oryzae, identified 140 orthologs that were implicated in the mating process of Saccharomyces cereviseae [2,34]. Genomic comparison between A. fumigatus AF293 (MAT1-2) and A1163 (MAT1-1) strains showed that A1163 strain possess all the genes found in the AF293, except for the genes Afu3g06160 and Afu3g06170 within the mating locus. The gene Afu3g06170 is known to encode a transcriptional factor with a High Mobility Group (HMG) domain. The gene Afu3g06160 is not fully characterized and thus, its role in the mating process is unknown. MAT1-1 strains possess only one gene in the mating locus, a transcriptional factor (AFUB_04290) that has an alpha box domain. The proteins in the MAT loci are important not only for sexual identity but also they are likely to act as master regulators of the sexual cycle, since they are both transcriptional factors. In fact, Scewczyk and Krappmann [35] reported that the mating proteins of A. fumigatus regulate the expression of ppga, a pheromone-encoding gene, in a reciprocal manner: the MAT-1 protein is required for ppga expression while the MAT-2 protein appears to repress its expression.
The predicted function of some of the proteins encoded in the 400 Kb region suggests a possible involvement in reproduction, and consequently potentially being selected during mating. For instance, found in this region are genes encoding for a sister-chromatid segregation protein homolog (src1), an adenylate cyclase (acyA), and two transcription factors including TFIID. The SRC1 protein was shown to be important in chromatin organization during mitosis in the yeast Saccharomyce cerevisiae, thereby ensuring proper segregation of nucleic acid material during nuclear division [36,37] and ACYA has been implicated in sporulation and is central to intracellular signaling [38]. Our findings suggest the possibility that specific alleles that are important for mating regulation might not be equally represented/functional in the MAT1-1 and MAT1-2 mating type strains of A. fumigatus and that they likely reside outside of the mating type locus. Evolution of the MAT locus in aspergilli is somewhat controversial. One of the evolution models hypothesize that heterothallic species derived from a common homothallic ancestor whereas in another model homothallic species evolved from heterothallism [39,40]. The results of genome sequencing and comparison between AFB62 and AFB62F9 provide an opportunity to uncover the components required for normal genetic segregation.
Generation of isogenic strains and their use for genetic analysis enabled extensive progress in the study of molecular pathobiology of other fungi. For instance, the involvement of mating type in the virulence of Cryptococcus neoformans was assessed with using isogenic strains [6–9]. Since pathology of A. fumigatus has also been suggested to correlate with its mating type [10,11,41] the isogenic pair, AFB62 and AFB62F9, was used to test this correlation. In all of the models tested, hydrocortisone treated or CGD mice as well as G. mellonella larvae, the virulence of AFB62 (MAT1-1) and AFB62F9 (MAT1-2) was similar. Also, histological sections of lungs from mice infected with either AFB62 or AFB62F9 did not reveal substantial differences in the manner by which these strains caused the disease. In previous studies Alvarez-Perez and collaborators [10] described that among 28 strains isolated from IA cases, 22 were MAT1-1 and only six were MAT1-2 suggesting a correlation between MAT1-1 and invasive aspergillosis. Furthermore, survival studies on larvae of G. mellonella showed a slightly higher virulence of MAT1-1 compared to MAT1-2 in the tested clinical isolates [11]. Since the pathology of A. fumigatus involves numerous factors, assessing the specific contribution of the MAT locus in virulence of strains is not possible unless the strains are congenic except for the MAT locus. Although predominance of a single mating type in clinical cases has been shown in other fungi, for instance Histoplasma capsulatum [42–44] and Cryptococcus neoformans [6,9], it is not the case with A. fumigatus and our findings suggest that the presence of either MAT1-1 or MAT1-2 locus in the genome of A. fumigatus does not predispose the fungus to higher virulence or affect the development of IA in murine models.
Azoles are the most frequently used drugs in treating IA. As a consequence, the incidence of resistance to azoles is also on the rise. Currently the most commonly reported mechanism conferring azole resistance in A. fumigatus is mutation in the erg11A (cyp51A) gene, leading to alterations in the target protein. The most common resistance-conferring mutation in this gene is an alteration at codon 98 (L98H), alone or in combination with a tandem repeat in the promoter region. Other variations are frequently detected in this gene, many of which have been demonstrated to confer resistance to azoles [12]. In this study we not only identified single mutations with a dominant phenotype like those affecting erg11A, but also other associated mutations that form part of the complex phenotype of resistance. Particularly, we found mutations in the HMG CoA reductase (hmg1) gene associated with voriconazole resistance, but no other azole. Voriconazole has a significantly different chemical structure than itraconazole and posaconazole [13] which might explain the difference in distribution of mutations. It is likely that voriconazole targets both the 14-alpha sterol demethylase and HMG CoA reductase enzymes and consequently both mutations are necessary for full resistance. Indeed, we found that the strains containing mutations in both ERG11 and HMG CoA reductase (like AFB62V31) were much more resistant to voriconazole than the HMG CoA reductase mutant alone. All the mutations we identified in HMG CoA reductase occurred within the sterol sensing domain that is responsible for feedback regulation of sterol-regulated genes and sterol biosynthesis [32]. Mutations in this domain in other species can result in proteins that are highly stable and unable to respond to sterol degradation products [45]. If the same is true in A. fumigatus, the hmg1 mutants should also be resistant to other azoles. We found that the spontaneous and our engineered hmg1P1038T mutant were resistant to all three azoles at roughly the same levels (2X – 4X MIC than wild-type). The results suggest that these mutant strains are indeed less sensitive to toxic intermediates that result from azole inhibition of ergosterol biosynthesis in general. However, the underlying reasons for hmg1 mutations being so frequently identified under voriconazole selection and not detected at all under itraconazole or posaconazole selection are not clear.
We found mutations in the heat shock protein SSC70 under itraconazole pressure which followed the expected segregation pattern in sexual crosses. Heat shock proteins, like Hsp90, and the stress response in general, have been associated with azole and echinocandin resistance in Aspergillus fumigatus [46,47]. The ssc70 transcript has been shown to be differentially regulated in response to stress and antifungals [48,49], though specific mutations have not been shown as potential resistance mechanisms. The mutation we identified resulted in a substitution of an arginine to cysteine residue in the nucleotide exchange factor domain of SSC70. The specific mechanism that leads to resistance is not clear, but may be a result of increased turnover during nucleotide exchange, thus leading to a higher capability of dealing with misfolded proteins as a consequence of sterol toxic byproduct accumulation. We were unable to engineer an ssc70 mutant strain, which suggests that the balance between turnaround and toxic byproducts is essential for cell development in A. fumigatus.
Lastly, we also found a mutation in erg25 (C-4 sterol methyl oxidase) in a highly itraconazole-resistant isolate. The mutation results in an altered HXHH motif that affects the coordination of the di-iron cluster that functions in the reaction center of the oxidase. The enzyme ERG25 acts downstream of ERG11, and mutations in this gene have been commonly associated with suppression of erg11A mutations. For example, in a recent study to identify A. fumigatus mutations associated with azole resistance, Camps et al. [19] also found mutations in erg25, though they did not find that the mutations in this gene alone could result in azole resistance. In their sexual crossings, the erg25 mutation was present in both susceptible and resistant strains and their attempt of gene replacement in a susceptible background did not result in any itraconazole-resistant strains. It is noteworthy, however, that both studies identified mutations in the same gene. In our study, we detected a mutation that directly changed the active site of the enzyme, while Camps et al. found a truncating mutation [19]. The sexual crossings in our study between F9I11 (containing the erg25 mutation) and the susceptible AFB62 parent resulted in modest numbers of ascospores at all levels of itraconazole. We were not able to sequence the pools from this cross due to limited DNA yield from resulting pools. Thus, while it was not possible to observe the specific selection of erg25 mutations after sexual crossing and recombination, further clarification of its involvement is warranted. In Saccharomyces cerevisiae it was shown that the lanosterol and ergosterol biosynthetic pathways are intimately intertwined and that in the absence of ergosterol (either because of azoles or mutation), the cells are still viable, presumably because of the incorporation of lanosterol or other sterols in the membrane instead of ergosterol [50]. We also found a high frequency of mutations in ganA—that encodes an ABC multi-drug transporter, a hypothetical protein, and a non-coding SNP, which almost always appeared together in the absence of other mutations in erg11A or hmg1. In sexual crosses these mutations did not follow the expected differences in frequency had they been linked with resistance. However, this might have been a result of our study design, rather than a lack of significance of these mutations in azole resistance. We purposefully selected only the high-level resistant isolates for the sexual crossing experiments with the hypothesis that several mechanism of resistance would be at play at this high level, and thus growing the progeny under varying amounts of drug would elucidate those mechanisms. We found, however, that mutations in erg11A and hmg1 alone were sufficient for high-level resistance, and as a consequence all other mutant alleles could have followed random segregation in their presence. Application of sexual crossing to strains with non-dominant mutations in erg11A or hmg1 would help clarify the role of these mutations. Similarly, it was possible that new de novo mutations would occur in a subpopulation of the pools upon drug selection of the pool. In fact, based on our initial selection in which we plated 105 conidia, we would expect to see up to 102 de novo azole resistant strains. This mutant abundance translates to an allele frequency of 0.001, which is below the level of detection for our bulk sequencing analysis. Thus, our strategy focused on already identified mutations in an attempt to clearly demonstrate the role of specific mutations in resistance.
The mating efficiency and isogenicity of AFB62F9 and AFB62 make these strains useful for recombinational analysis in genetic studies of A. fumigatus pathobiology, particularly for traits for which selection of a phenotype, such as drug resistance, can be accomplished. In this study we demonstrate that classical genetic experiments coupled with whole genome sequencing is a highly effective way of identifying mutations associated with a specific phenotype in A. fumigatus.
The strains AFB62 (MAT1-1) and AFIR928 (MAT1-2) of A. fumigatus [5] were employed as the parental strains in the construction of an isogenic pair. The strains AFB62Δabr2 [5] bearing brown conidia and AFB62F9Δalb1 bearing white/albino conidia were used for genetic recombination assays.
AFB62F9Δalb1: deletion of the gene alb1/pksP gene (Afu2g17600) was obtained by transforming AFB62F9 with the vector pDHt/alb1::hph via ATMT [51]. Strain AFB62hmg1P1038T: a construct harboring the P1038T mutation in the hmg1 gene of AFB62V31 was created to replace the native hmg1 locus in AFB62. The construct harbored the hygromycin resistance cassette flanked by a 3,870 bp fragment consisting of the hmg1 (Afu2g03700) ORF plus 300 bp sequence adjacent to the termination code of the gene on one side and a 1,002 bp fragment located immediately after the termination code of the gene on the other side. Thus, the 1,002 bp fragment has the repeat of the 300 bp adjacent to the termination code cloned into the 3,870 bp fragment. Both fragments were amplified from AFB62V31. The construct was cloned into the pDHt/SK vector and electroporated into Agrobacterium tumefaciens to transform AFB62 conidia [51]. Southern analysis confirmed homologous recombination of a single copy of the construct in the mutant AFB62hmg1P1038T. Sequencing confirmed the P1038Thr mutation.
Mating, progeny isolation and viability were carried out as described elsewhere [5]. Generation of isogenic set: F1 progeny, which was produced by mating of AFB62 with AFIR928 [5], was backcrossed with AFB62. Ascospores from one of the most fertile crosses between F1 and AFB62 were isolated and backcrossed with AFB62 to generate F2 progeny. Back-crossings were successively repeated until F9 generation was obtained. Each backcrossing was carried out with 10–20 progenies isolated from the most fertile crosses from each generation. Mating plates were analyzed at 4–8 weeks. Heat-treated ascospores (70°C for 30 min) were plated on either malt extract agar (MEA) or Aspergillus minimal media (AMM) agar (with or without 10 mM of uracil and 10 mM of uridine), incubated at 37°C for 48h and stored for further use. Controls for conidial killing were carried out with parental conidia harvested from mating plates. Conidia were heat-treated at 70°C for 30 min, plated on MEA and incubated at 37°C. After 48 h no growth was observed confirming that parental conidia were killed when incubated at 70°C for 30 min. Mating type of progeny was assessed by PCR with primers specific for amplification of MAT1-1 and MAT1-2 mating types [5]. Mating of azole resistant isolates: the AFB62 resistant isolates were mated with the naïve AFB62F9 strain and the AFB62F9 resistant isolates were mated with the naïve AFB62. Crosses were analyzed at 20 to 25 weeks. Ascospores from fertile crosses were heat-treated, inoculated on MEB containing various drug concentrations and incubated for 48–72 h at 37°C and 220 RPM. Progeny of voriconazole resistant isolates were grown on 0 (no drug), 0.376 (1.5X MIC) and 3 μg/ml (12X MIC) of voriconazole. Progeny of itraconazole resistant isolates were grown on 0 (no drug), 10 (5X MIC) and 100 μg/ml (50X MIC) of itraconazole. Progeny of posaconazole resistant isolates were grown on 0 (no drug), 0.128 (1X MIC), 1.3 (AFB62F9 derivatives, 10X MIC) and 10 μg/ml (AFB62, 10X MIC) of the drug. Fungal ball-colonies derived from ascospores germination were washed with PBS, counted and lyophilized. Pools ranging from 100–2000 fungal-ball colonies derived from each cross were used for DNA isolation with DNeasy Plant Kit (Qiagen) according to manufacturer’s instruction.
1x107 conidia were inoculated into 10 ml of yeast nitrogen broth (BD, Sparks, MD) and incubated for 24 hours with shaking at 200 rpm. Mycelia were harvested, lyophilized and DNA isolated with DNeasy Plant Kit (Qiagen) according to manufacturer’s protocol. DNA quality was monitored with 0.8% agarose gel electrophoresis.
CGH was performed as described previously [2]. Hybridizations of the biological replicates were repeated in dye-swap sets. Hybridized slides were scanned and analyzed to obtain relative hybridization profiles. Normalized data were averaged over replications, and genes with differential hybridizations at the 95% confidence level were determined using intensity-dependent Z-scores (with Z = 1.96). The resulting data were analyzed to identify genes variably present in any of the probed genomes. Any genes with hybridization signals with greater than 1.4 fold differences (-0.75 ≤ log2 ≤ 0.75) between two different strains were considered absent or diverged.
Genomic DNA was sequenced using paired-end Illumina GA-II to 44-fold coverage. The assembly of AFB62 was done with CLC Genomics Workbench 4.1 using strict parameters: length fraction = 0.9, similarity = 0.9, allowing for random placing of non-specific reads. Sequence data for this project has been deposited to NCBI (BioProject ID: PRJNA237785).
Mice with Chronic Granulomatous Disease (CGD) and mice immunosuppressed with hydrocortisone were used for murine model studies. CGD mice, strain gp91phox-deficient mice (B6.129S6-Cybbtm1Din/J) (The Jackson Laboratory), were inoculated with 30 μl of 3.33 x 105 conidia/ml (5 mice per group). The immunosuppressed model was carried out with BALB/c mice (National Cancer Institute Division of Cancer Treatment, Bethesda, MD) treated with hydrocortisone acetate (Sigma, St Louis, MO) as described elsewhere [5]. These mice received 30 μl of 3.33 x 107 conidia/ml (10 mice per group). All mice were inoculated via pharyngeal aspiration [52]. Lung sections were stained with hematoxylin and eosin (H&E) and Gomori methenamine silver (GMS) for histopathology analysis. Survival data were analyzed using the log rank test. The animal experiments were performed under protocol approved by IACUC of the National Institute of Allergy and Infectious Diseases at the U. S. National Institutes of Health (NIH). Galleria mellonella larvae in the final larval stage (Vanderhorst Wholesale, Inc., St. Marys, OH) were inoculated with 5 μl of 2x107 conidia/ml as described elsewhere (15 larvae per group) [24,53]. Larvae were then incubated at 37°C and mortality monitored daily. Mortality was assigned based on change of color from light beige (healthy larvae) to dark brown and unresponsiveness to tactile stimulus.
The Institutional Animal Care and Use Committee of the National Institute of Allergy and Infectious Diseases approved all animal studies (#A4149-01). Studies were performed in accordance with recommendations of the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health.
MIC assays were carried out following the method for broth dilution [54] or with Etest. Etest was performed by following recommended protocols (BioMerieux). Briefly, 106 conidia were plated on MEA plates, the Etest strip was placed on the surface of the plate and the plates were incubated at 37° for 6 days. The MIC was determined as the concentration in the Etest strip where the growth was inhibited. Unless otherwise specified, MICs were determined by Etest.
1x 105 conidia of AFB62 or AFB62F9 were plated on MEA containing sub-MIC concentrations of itraconazole (2.0 μg /ml), posaconazole (0.032 μg/ml), or voriconazole (0.094 μg/ml) and incubated at 37°C for 1 week. Resulting conidia were resuspended in 0.01% TritonX-100 and 1x 105 conidia were plated on 1X – 3X MIC (S3 Table) in order to obtain fewer than 100 colonies per plate. Three resistant colonies from each azole for each parent (AFB62 or AFB62F9) were picked, grown individually in MEA plates containing the same amount of azole as was used for selection, and incubated until conidia were visible, these represent the intermediate resistance isolates. A total of 18 isolates were collected at this stage (2 parents x 3 drugs x 3 isolates each). 1x 105 conidia from the resistant strains were plated onto MEA containing 10X – 50X MIC in order to obtain fewer than 100 colonies on each plate. Again, three colonies from each plate were picked and grown individually on MEA containing the high MIC concentration of the corresponding azole (S3 Table). A total of 48 highly resistant isolates were collected (2 parents x 3 drugs x 3 lineages from intermediate selection plates x 3 isolates) and stored as conidial stocks.
For DNA isolation of intermediate or highly resistant isolates, 1x 107 conidia were inoculated into 10 ml of malt extract broth (MEB) containing drug in various concentrations. Drug resistant isolates of AFB62 and AFB62F9 were grown on voriconazole at concentrations of either 0.376 or 2 μg/ml or itraconazole at either 2 or 10–50 μg/ml. AFB62 posaconazole isolates were grown on either 0.32 or 10 μg/ml and AFB62F9 isolates were grown on either 0.128 or 1.3 μg/ml of posaconazole. Mycelia were harvested after 24–48 h incubation at 37°C in orbital shaker (220 rpm). DNA was isolated using DNeasy Plant Kit (Qiagen) according to manufacturer’s instruction. DNA quality was monitored by 0.8% agarose gel electrophoresis.
For intermediate and high-resistance isolates obtained during in vitro selection, the sequencing reads from each isolate were first trimmed on quality (q> = 30) and then mapped to the AFB62 assembly using mpileup using default parameters [55]. Identified single nucleotide polymorphisms (SNPs) were filtered to identify high quality SNPs using the following criteria: AC = 1, AF = 2, DP > = 10, MQ> = 40 and FQ< = -35, and covered by at least 4 reads in each direction. For pools, the proportion of reads with the wild-type and mutant nucleotides were queried from the unfiltered mpileup results of mapping the reads from the pools to the AFB62 genome.
|
10.1371/journal.pntd.0003327 | Pregnancy Outcome in Relation to Treatment of Murine Typhus and Scrub Typhus Infection: A Fever Cohort and a Case Series Analysis | There is a paucity of published reports on pregnancy outcome following scrub and murine typhus despite these infections being leading causes of undifferentiated fever in Asia. This study aimed to relate pregnancy outcome with treatment of typhus.
Data were analyzed from: i) pregnant women with a diagnosis of scrub and/or murine typhus from a fever cohort studies; ii) case series of published studies in PubMed using the search terms “scrub typhus” (ST), “murine typhus” (MT), “Orientia tsutsugamushi”, “Rickettsia tsutsugamushi”, “Rickettsia typhi”, “rickettsiae”, “typhus”, or “rickettsiosis”; and “pregnancy”, until February 2014 and iii) an unpublished case series. Fever clearance time (FCT) and pregnancy outcome (miscarriage and delivery) were compared to treatment. Poor neonatal outcome was a composite measure for pregnancies sustained to 28 weeks or more of gestation ending in stillbirth, preterm birth, or delivery of a growth restricted or low birth weight newborn.
There were 26 women in the fever cohort. MT and ST were clinically indistinguishable apart from two ST patients with eschars. FCTs (median [range] hours) were 25 [16–42] for azithromycin (n = 5), 34 [20–53] for antimalarials (n = 5) and 92 [6–260] for other antibiotics/supportive therapy (n = 16). There were 36.4% (8/22) with a poor neonatal outcome.
In 18 years, 97 pregnancies were collated, 82 with known outcomes, including two maternal deaths. Proportions of miscarriage 17.3% (14/81) and poor neonatal outcomes 41.8% (28/67) were high, increasing with longer FCTs (p = 0.050, linear trend). Use of azithromycin was not significantly associated with improved neonatal outcomes (p = 0.610)
The published ST and MT world literature amounts to less than 100 pregnancies due to under recognition and under diagnosis. Evidence supporting the most commonly used treatment, azithromycin, is weak. Collaborative, prospective clinical trials in pregnant women are urgently required to reduce the burden of adverse maternal and newborn outcomes and to determine the safety and efficacy of antimicrobial treatment.
| Typhus is an under-recognised and under-studied public health problem in Asia. In rural areas of Southeast Asia murine and scrub typhus are probably the most common treatable cause of fever. The estimated number of scrub typhus cases in Southeast Asia, more than 1 million yearly, results in approximately 50–80,000 deaths per year. Treatment delays due to lack of appropriate diagnostics and lack of awareness lead to a substantial health and economic impact in the one of the world's most densely populated regions. Only 97 cases in pregnancy are available from the published world literature over the past 18 years. Only 82 of these had known outcomes, including two maternal deaths. The proportion of poor neonatal outcome including stillbirth, prematurity and low birth weight was high occurring in more than 40% of pregnancies, and higher when the fever clearance time was longer. While poor neonatal outcomes were observed with all antibiotics prescribed, azithromycin appeared to be associated with shorter fever clearance times but this was not statistically significant. Evidence to support the use of azithromycin is weak. The correct antimicrobial or combination for undifferentiated fever in pregnant women in Southeast Asia is unknown.
| The true burden of scrub typhus (ST) and murine typhus (MT) in South-east Asia remains largely unknown as diagnostic tests are rarely used or available [1]. Still less is known about typhus in pregnancy with no (large scale) epidemiological studies addressing the subject in this group [2]. No data suggests that typhus occurs more or less commonly in pregnancy. Studies showing that the malaria vector An. gambiae mosquitoe was more attracted to pregnant women than non-pregnant women [3], have not been performed with the vectors of ST (mites) and MT (fleas). There is a scatter of small case series and reports on ST and MT [4]–[16] which tend to suggest serious repercussions for the mother and fetus, and congenital transmission has also been reported [10], [11]. There is a definitive lack of new studies or data in the published literature with a four year gap between the two latest publications [16], [17]. This a likely reflection of the difficulty of confirming the diagnosis and following a woman treated in out- or in-patients through to pregnancy outcome in the obstetric department.
The tetracycline-class of antibiotics, principally doxycycline, treat rickettsial diseases [18]. Congenital malformation has not been associated with doxycycline use [19], [20] but the class effect ascribed to tetracyclines including possible effects on the musculoskeletal development of the fetus [21] and staining of teeth in young children [22] precludes its use. While short courses of doxycycline are probably safe there is insufficient evidence to support their use as first line therapy [5], [23]. Macrolide antibiotics such as azithromycin, are safe in pregnancy and reported to be equally effective to doxycycline for the treatment of scrub typhus [24]. The aim of this study was to undertake a detailed analysis of ST and MT cases from a previously published fever cohort and of the published literature to provide further detail on pregnancy outcomes in relation to treatment.
For women participating in the fever cohort the patient information sheet and consent form were available in Karen and Burmese languages. Willing participants signed (literate) or provided a thumb-print (illiterate) on the consent form. Approval for the study was granted by the Oxford Tropical Research Ethics Committee, UK (#013-03) [16].
Details of the methods can be found in this previously published study on the epidemiology of fever in refugee and migrant pregnant women on the Thai-Myanmar border where scrub and murine typhus accounted for 12.3% (26/211) of fever diagnoses in 203 women [16]. Briefly, febrile (aural temperature >37.5°C) pregnant women following the antenatal clinics of Shoklo Malaria Research Unit (SMRU) were offered a fever screen. Relevant to this study is the diagnosis typhus and malaria. Typhus was confirmed by PCR [25]–[28] and/or in vitro isolation of Rickettsia spp. [29] and/or positive reference serology measured by four-fold rise in paired sera IFA [30], [31] to define acute infection as MT (Rickettsia typhi) or ST (Orientia tsutsugamushi). Malaria was confirmed by microscopy of thick and thin malaria smears, stained with Giemsa and examined under oil immersion. Smears were only declared negative after 200 fields were read.
None of the ST and MT results were available for the patient as these were initially stored and later analyzed off-site. Women were admitted and treatment was initiated according to the clinical examination and available laboratory test results. This implies patients may have been treated for other infections such as malaria because the result was immediately available and consistent with the clinical picture. ST or MT may not have treated as it was not suspected and the laboratory result was unavailable. Patients were monitored for fever clearance time (FCT) by 6-hourly aural (Genius) temperature measurement. Women followed regular antenatal care after discharge and were encouraged to deliver with skilled birth attendants where they are weighed directly after birth. Infants born at home and not weighed on the day of birth had their weight adjusted as described for this population [32]. Poor neonatal outcome was a composite measure for women whose pregnancy was sustained to 28 weeks or more of gestation and included stillbirth, prematurity and fetal growth restriction defined as a birth weight for gestational age of <10th centile for the population [33].
Since Dec 2012 pregnant women at SMRU with fever have routinely been tested with the SD Bioline Tsutsugamushi from Standard Diagnostics, Korea, a rapid test, in addition to other standard fever screening tests. This rapid diagnostic test is non-specific testing IgG, IgM and IgA Orientia tsutsugamushi antibodies. Acute and convalescent serum was collected. Remaining serum was stored at −20°C. These samples were collected over a one year period and processed in the same laboratory as per the aforementioned fever cohort women. ST was confirmed by PCR [25]–[28] and/or positive reference serology measured by four-fold rise in paired sera IFA [30], [31]. Women were followed up in the routine antenatal service and encouraged to birth with skilled birth attendants at Shoklo Malaria Research Unit. Of 25 women with a positive rapid test there were three with a confirmed ST diagnosis and a known pregnancy outcome included for analysis.
The pooling of data into a single data set includes the aforementioned fever cohort women, additional unpublished cases, and data reported from case reports or case series in 16 different publications [5]–[7], [9], [10], [12], [15]–[17], [34] (Supporting Information S1). As the population birth weight percentiles were unknown for these publications, low birth weight (<2500 g) replaced fetal growth restriction in the definition of poor neonatal outcome. Individual patient data including age, gestation at infection, days of fever on admission, FCT, treatment, maternal death, birth outcome (including miscarriage (delivery before viability set at <28 weeks for resource limited settings), delivery and unknown), birth weight and gestational age at delivery, live birth or stillbirth, birth weight and congenital normality were extracted from published studies into a data spreadsheet (Supporting Information S2).
Data were described using the statistical program SPSS Statistics version 20.0 for Windows (SPSS Benelux inc., Gorinchem, Netherlands). Continuous normally distributed data were described by the mean (standard deviation, range) and non-normally distributed data by the median [min-max] and comparisons used the Student's t-test or Mann-Whitney U test, respectively. The number and percentage were given for categorical data and compared using the X2-test with Yates' correction, or the Fisher's exact test when applicable.
Poor neonatal outcome was compared by univariate analysis to the treatment group. Given the small sample size and to avoid erroneous conclusions no modeling of poor neonatal outcome was attempted.
From 2004 to 2006, 26 febrile pregnant women were identified as having evidence of rickettsial infection with confirmation by serology in 9 women, serology and PCR in 17 women (MT n = 7, ST n = 9, and both n = 1). In 4 of these cases O. tsutsugamushi was isolated from blood culture and all of these women were PCR and serology positive. Culture positive cases included the single patient with mixed ST and MT infection.
Note that in 5 cases there was concomitant infection with malaria. Malaria cases were all microscopic: three MT (with two concomitant P. falciparum and one P. vivax infections) and two ST with concomitant P. falciparum. These five patients were grouped respectively into MT and ST [16]. The patient who had dual infection with ST and MT (dynamic serology for MT, and PCR positivity for both) documented at 37+2 weeks+days gestation could not be placed into either group (ST or MT) for analysis. Briefly she was a 33 year old with 4 previous births, a FCT of 119 hours with IV ceftriaxone and a normal birth outcome (singleton, estimated gestation 41+3 weeks+days and birth weight of 3600 g).
Overall there were 13 women with ST, 12 with MT and one with ST and MT. The baseline demographic and obstetric characteristics have been detailed for MT and ST diagnostic groups (Table 1). There were no significant differences observed in baseline characteristics. Of note is the higher proportion (non-significant): of flea-borne MT was higher in the refugee camp than in migrants; and of smokers in the ST group.
From Dec 2012 until Feb 2014 there were 22 febrile pregnant women with a positive SD Bioline Tsutsugamushi RDT and serum samples available for further diagnostic testing. Of these women only three had a 4-fold rising ST IgM titre (two 1∶400 to 1∶1600 and one 1∶1600 to 1∶6400) and none had a positive ST PCR result. All 3 women delivered term, live born, normal infants one of which was of low birth weight (treated with paracetamol and fever had cleared by the time the lab result was available). Treatment in the other two women was azithromycin in one case based on protocol, and multiple antibiotics for the other case: initially ciprofloxacin for suspected pyelonephritis, two days later erythromycin for suspected exacerbation of asthma and one day later azithromycin based on the positive RDT result with rapid fever clearance, <24 hours, once this drug was commenced.
Over an 18 year period, from 1992 to 2014 there were 97 women with a diagnosis of MT or ST (or both) in pregnancy including the 26 reported from the SMRU fever cohort [16] described above, the 3 unpublished cases, and 85 in peer reviewed publications [5]–[7], [9], [10], [12], [15]–[17], [34] (Supporting Information S1). The number of cases (per publication) included 42 (n = 1), 26 (n = 1), 9 (n = 1), 5 (n = 2), 2 (n = 2) and 1 (n = 3), from five different countries including 52 from India, 30 from the Thai-Myanmar border, 11 from Korea, and 1 from Taiwan. The three unpublished cases were also from the Thai-Myanmar border. Reported diagnostic methods were predominantly serology-based, which included Weil-Felix and IFA testing, and few were paired titrations and/or PCR (Supporting Information S1).
Definitive conclusions about the effects of rickettsiosis in pregnancy are not possible given the dearth of available evidence in the global literature. The low number of published cases from the past 18 years, amounting to 87 with a known pregnancy outcome and only 67 with a gestation of at least 28 weeks gestation, strongly suggests that scrub typhus and murine typhus is severely under recognized. In view of the 2.4% maternal mortality and the poor pregnancy outcome in survivors reported here, there is an urgent need to improve access to diagnosis and treatment. The performance of the currently available tests is not satisfactory and represents a major barrier. The one year field experience with scrub typhus RDT in pregnancy at SMRU has led to the abandonment of the test and also highlights the risk of over diagnosis of scrub typhus in endemic settings.
If rickettsial infection does not result in miscarriage the outcomes for the neonate are not encouraging. In the SMRU fever cohort where population data collected during the same period is available, typhus appeared to result in a higher proportion of preterm birth (14.3%) and birth weight below the 10th centile (22.2%) than reported for malaria during pregnancy (7.3% (58/794) and 17.0% (88/519) respectively) [33].
This was also highlighted by the south India case series of fever before delivery where pregnancy loss with scrub typhus was significantly higher than observed from routine obstetric data at the same hospital: 33% vs 2.8%; P<0.001 [17].
The pathogenic mechanisms associated with adverse pregnancy outcomes in rickettsial infection are unknown. In malaria the predominant pathology is monocyte infiltration in the placental intervillous space, but whether this is similar in tyhus is unknown. The rickettsia and orientia-associated vasculopathies result from either direct endothelial infection and/or an inflammatory monocyte-dominant perivascular infiltration [36]–[38]. The consequences of endothelial damage in murine typhus and an inflammation-associated pro-coagulant state in scrub typhus are suggestive of a predominantly circulatory impediment, possible due to thrombotic occlusion and or coagulopathy [39]. This contrasts with the disease observed in the SMRU fever cohort where mild, self-limiting, flu-like illness, without signs of organ dysfunction such as jaundice and renal insufficiency, was observed [40]. Indeed three women all with MT made a spontaneous recovery without antibiotics. There were no convincing clinical or complete blood count features that helped differentiate ST from MT [41], [42] or suggested that MT was a milder illness than ST in pregnancy as described previously in children (Table 2) [42]. Eschar was barely mentioned in any of the reports although it can be a valuable diagnostic clue for ST, if recognized by clinicians [1].
The range of treatments given to pregnant women was wide and consistent with the difficulty of making a clinical diagnosis of ST and MT with a non-specific early phase of the disease [43]. It reflects the confusion that undifferentiated illness can cause in remote and rural areas where access to fever diagnostics is very limited. In response to this problem the guidelines at SMRU now include azithromycin as an empirical first line therapy for pregnant women in undifferentiated fever cases but since azithromycin was not associated with improved neonatal outcome this might not be the right choice? FCT has classically been used as one of the markers of an efficacious treatment of ST and MT [44] and has served to advocate for azithromycin as treatment in pregnant women with scrub typhus in Thailand [8] and Korea [5], as well as its safety profile in first trimester [8]. The antimicrobial armamentarium is limited for this disease in pregnancy. Doxycycline is inexpensive and the drug of choice in non-pregnant patients for rickettsial illness and the shortest described treatment course is at least 3 days as well as having short fever clearance times [45]. Doxycycline can be used in pregnancy, if no alternative is available and there is no other contraindication [46], which is a real case scenario in much of the rural tropics, in addition to it being a cheaper alternative than azithromycin. Most treatment efficacy studies of typhus have been limited to a short follow-up of 28 days but like malaria [47] a longer duration of follow-up is likely to be required to monitor for relapse in pregnant women and further pharmacokinetic studies are required [48]. While rifampicin monotherapy with doses of 900 mg and 600 mg (mean fever clearance times 22.5 and 27.5 hours, respectively) led to shorter fever clearance times than doxycycline monotherapy (mean fever clearance time 52 hours) in northern Thailand [49] prolonged fever clearance times were observed with the combination of rifampicin and doxycycline. A study on rifampicin for brucellosis which co-administered doxycycline observed reduced drug concentrations of doxycycline and was associated with treatment failure or relapse [50]. Rifampicin and azithromycin can decrease the level or effect of azithromycin probably by an interaction with P-glycoprotein [MDR1] transported which effects how azithromycin is eliminated from the body [45].
The usefulness of antimalarials to treat obligate intracellular bacteria is biologically plausible and appreciated only because the SMRU fever cohort explicitly aimed to investigate comprehensively the cause of fever by offering a battery of tests, identifying truly coinfected women [16]. If antimalarials are effective in the treatment of rickettsia, it may in part explain why there remains a paucity of diagnosis of rickettsia in areas with malaria. Antimalarials should be tested in-vitro for their effectiveness against Orientia and Rickettsia species. Thailand has been highly geared to early diagnosis and treatment of malaria over the past 4 decades and the problem of rickettsial illness may become more overt as malaria continues to decrease [51]. It needs to be noted that even if scrub typhus was diagnosed and treated as effectively as malaria, the incidence would likely not be affected since humans are dead-end hosts and immunity is short lived [2].
Many infections that predominate in low-income countries such as malaria, tuberculosis, hepatitis and rickettsioses, fail to be included as causes of maternal and neonatal mortality in part due to the difficulty of making the diagnosis and weakness in data collecting systems [52]. Indeed one of the maternal deaths in the series presented here, without PCR diagnostics, would have been classified as an indirect maternal death from sepsis as she died from multi-organ failure [53]. While maternal mortality from sepsis is reported to be on the decrease with recommendations for improved sanitation, death from ‘sepsis’ in tropical countries may be more complex than it first appears due to undiagnosed tropical neglected diseases [52]. Likewise neonatal death from preterm birth and stillbirth requires a significant investment to elucidate the true cause in tropical countries because a first trimester infection may impact many months down the line.
The main limitations of this study are the use of retrospective data subject to reporting bias and the weak serology-based rickettsial diagnostics of the case series [54]. Despite this the proportion of poor neonatal outcome was similar in the fever cohorts with more reliable confirmation including PCR (Table 5). Only real-time diagnostic capacity, classification of disease severity and standardized approaches to treatment in prospectively followed pregnancies will allow clarification of the findings [55]. Given that typhus represents the leading cause of treatable undifferentiated fever in Southeast Asia [56], [57], and that its regions of high endemicity are among the most populous areas in the world, the need for prospective fever-in-pregnancy studies to estimate the burden and best treatment of disease in this group is obvious. Asymptomatic infection is also a high possibility and the implications for pregnancy unknown. Theoretically azithromycin is the drug of choice in pregnancy but the worldwide data collated here only provides support for level-4 evidence for the use of this drug and this must improve.
The data presented here highlights the potential implications and severity of this easily treatable infection and demonstrates the inexorably slow pace of improvement in our understanding of rickettsial illness in pregnancy: less than 100 women with a known pregnancy outcome in 18 years. Progress towards the 2015 countdown to Millenium Development Goals to improve maternal, newborn and child survival will not be met in relation to scrub and murine typhus. Evidence to support the use of azithromycin is weak. The correct antimicrobial or combination for undifferentiated fever in pregnant women in South-east Asia is unknown.
|
10.1371/journal.pntd.0004033 | Rheumatic Heart Disease-Attributable Mortality at Ages 5–69 Years in Fiji: A Five-Year, National, Population-Based Record-Linkage Cohort Study | Rheumatic heart disease (RHD) is considered a major public health problem in developing countries, although scarce data are available to substantiate this. Here we quantify mortality from RHD in Fiji during 2008–2012 in people aged 5–69 years.
Using 1,773,999 records derived from multiple sources of routine clinical and administrative data, we used probabilistic record-linkage to define a cohort of 2,619 persons diagnosed with RHD, observed for all-cause mortality over 11,538 person-years. Using relative survival methods, we estimated there were 378 RHD-attributable deaths, almost half of which occurred before age 40 years. Using census data as the denominator, we calculated there were 9.9 deaths (95% CI 9.8–10.0) and 331 years of life-lost (YLL, 95% CI 330.4–331.5) due to RHD per 100,000 person-years, standardised to the portion of the WHO World Standard Population aged 0–69 years. Valuing life using Fiji’s per-capita gross domestic product, we estimated these deaths cost United States Dollar $6,077,431 annually. Compared to vital registration data for 2011–2012, we calculated there were 1.6-times more RHD-attributable deaths than the number reported, and found our estimate of RHD mortality exceeded all but the five leading reported causes of premature death, based on collapsed underlying cause-of-death diagnoses.
Rheumatic heart disease is a leading cause of premature death as well as an important economic burden in this setting. Age-standardised death rates are more than twice those reported in current global estimates. Linkage of routine data provides an efficient tool to better define the epidemiology of neglected diseases.
| Rheumatic heart disease is the result of an abnormal immune response to the bacteria Streptococcus pyogenes. The disease causes permanent scarring of the heart values, which results in heart failure, stroke and early death. It primarily affects the world’s poorest and most disadvantaged populations and despite the availability of cheap and effective prevention strategies receives little attention from policy-makers and funders. One of the major difficulties has been measuring how many people die prematurely from this disease. Simply counting up deaths is highly inaccurate and so an alternate strategy was needed. Focusing on Fiji in the Western Pacific, we pulled together information from several different health databases using a process called record-linkage. We then worked out how much more frequently rheumatic heart disease patients die than you might expect when comparing them to persons of similar age, gender and ethnicity in the general population. From these data we estimate about twice as many patients were dying from the disease than had been previously suggested. Most of these deaths occurred earlier than was thought with substantial knock-on effects for the economy. On balance we think this strategy for measuring mortality is useful and robust, and it will be increasingly possible to employ it elsewhere.
| Rheumatic heart disease (RHD) is the chronic consequence of an aberrant immune response to infection by the bacterial pathogen Streptococcus pyogenes that results in permanent scarring of the heart valves. [1] This process, which may manifest clinically as heart failure, stroke and early death, [2] remains a major public health problem in developing countries. [3–5] Despite this, efforts to measure the disease burden and institute control strategies are impeded by the lack of up-to-date epidemiologic data from endemic areas. [4, 6, 7]
Although current global estimates assert there are approximately 275,000 deaths due to RHD each year, [8] deriving such figures has been problematic. [4, 7] In 2005, a WHO report found mortality estimates based on either vital registration data or verbal autopsy techniques to be unreliable, largely because of the difficulty distinguishing RHD-attributable death from other causes of cardiac death. [9] To remedy this, the authors extrapolated from estimates of prevalence and studies of natural history. However, with few current data [10], the only available data were those from urban populations living in the UK, USA and Japan in the early to mid-twentieth century and socially disadvantaged indigenous populations living in Australia and New Zealand today. [9] We therefore sought to measure RHD-attributable mortality in Fiji, a developing nation in the Western Pacific, where a high prevalence of RHD has consistently been reported. [11, 12]
We established a new national and historical cohort of RHD patients in Fiji by probabilistic record-linkage, using diagnostic information and outcome events ascertained from routine clinical and administrative records for the period 2008–2012 (Fig 1). We used relative survival methods to estimate and examine RHD-attributable deaths in persons aged 5–69 years. [13] We then used census data to calculate crude, age-specific and age-standardised RHD-attributable death rates for the wider population, as well as years of life lost (YLL) from which we estimate the cost to the economy. Finally, we compared our RHD mortality estimate with vital registration data for 2011–2012.
Fiji is an independent nation located in the Western Pacific with an estimated population of 837,271 at the most recent census in 2007. [14] The population consists of two major ethnic groups, Indigenous iTaukei Fijians (57%) and Fijians of Indian Descent (38%). Fiji is ranked 96th of 186 nations for the composite human development index. [15]
The routine data used in this study was obtained from four sources: an electronic patient information system, a database of death certificates, a disease control register, and echocardiography clinic registers.
Information that referred to the same person was amalgamated from the four datasets using identifier fields. [19] We designed and calibrated a probabilistic record-linkage procedure using 1,406 known duplications in the patient information system from which we calculated the sensitivity and specificity (S1 Box). In its final configuration, our procedure identified the known duplications with sensitivity of 91.4% and specificity of 99.9% with record pairs considered a match if they achieved a posterior probability of over 50%. Stata® code for the procedure is available for download at: http://users.ox.ac.uk/~clme1250/data_linkage/linkage.html.
Our first step was to find at least one match in the patient information system for each record available from the death certificate database, the control programme register and the echocardiography clinic registers. In the absence of a universal identifier, we used a variety of identifier fields in the records including names, dates and demographics to detect pairs of records that referred to the same person (S2 Table). After cleaning and standardisation of names, dates and other identifier fields, we performed an initial shortlisting step (termed ‘blocking’) by finding groups of individuals of similar age with similar names. Next we compared identifiers within each pair or shortlisted records and classified them as being in agreement, partial agreement, disagreement or missing. [20] These classifications allowed for out-of-order names and dates as well as abbreviations and spelling discrepancies of names. [21] We then undertook a further blocking step using combinations of the identifier classifications to define smaller blocks that were expected to contain 5–75% true matches based on the number of pairs per search record. Next, for each block, we estimated the likelihood that each combination or pattern of classifications indicated a true match under the Fellegi-Sunder model of record-linkage; [22] match and nonmatch probabilities were estimated using an expectation maximisation algorithm as previously described. [23] Finally, we obtained a posterior probability of a match by multiplying the raw likelihood by an estimate of prior probability obtained from the product of: 1) The probability that a random pair represented a match, which equated to the reciprocal of the size of the final block in which that pair was found [20]; 2) An estimate of the probability that a given individual had actually been registered in the patient information system [20], based on their locality of residence, age, gender and ethnicity; and 3) For pairings with death certificates only, the probability an individual had died that year based on their age, gender and ethnicity. [24] Record pairs achieving a posterior probability of 50% or more in at least one block were consider a match.
Once links had been identified, we repeated the procedure to confirm or refute the merger of two or more records in the patient information system. Because it was impractical to search for duplicates across the entire patient information system, we limited this search for duplicates to a shortlist of records that were potentially relevant because of a possible match or because they contained useful clinical information such as a relevant admission diagnosis. Finally we pulled clinical information from each of the sources into a single linked record and deleted identifiers. If there was a discrepancy between the records, the patient information system was assumed to be correct unless that field was missing. Where there were discrepancies amongst two or more patient information system records these fields were set to missing.
We focused on 2008–2012 because the most complete data were available for this period. The study was restricted to people aged 5–69 years because RHD is expected to cause very few deaths before five years of age and cause-of-death information can be unreliable in old age. Based on diagnostic information in control programme records, echocardiographic data, hospital discharges and death certificates, any individual with follow-up at ages 5–69 years who had a least one record of a diagnosis of either RHD or its precursor acute rheumatic fever (ARF) was eligible for inclusion (S2 Box). We assumed the onset of disease was in childhood irrespective of the date the individual became known to clinical services, circumventing potential bias due to late presentation. The primary outcome was the time to all-cause death defined by a date of death in either the patient information system or the death certificate database. The secondary outcomes were cause-specific death defined: 1) Narrowly, where the underlying cause-of-death was ascribed to an International Statistical Classification of Diseases and Related Health Problems 10th Revision (ICD10) code pertaining to RHD or ARF; and 2) Broadly, where the death certificate listed an ICD10 code pertaining to either RHD, an alternative description of valvular heart disease, or a cardiac complication thereof (i.e. heart failure, stroke, infective endocarditis, arrhythmia) as an immediate or underlying cause-of-death, without listing ischaemic heart disease. We surmised there was no loss to follow-up because the linked patient information system and the death certificate database would include most deaths in the country during 2008–2012. Deaths in the cohort that went undetected due to, for example, emigration, would have led us to under-estimate RHD-attributable mortality.
Data were inspected for missing and outlying points, categorical variables were tabulated and continuous variables were summarised in histograms. The cohort data were expanded into thirteen five-year age categories and standardised mortality ratios (SMR) and relative survival rates were calculated by applying the background age category, gender and ethnicity specific death rates to the cohort. [13] Background death rates were calculated using the count of deaths in the patient information system and death certificates after duplicates had been removed divided by estimates of population using data from the Fiji Bureau of Statistics. [14] Before application these rates were checked for adherence with expectations by performing of standard quality control checks such as log rate versus age category. We assumed time-to-death from RHD was independent of time-to-death from other causes, consistent with the previous application of relative survival to RHD mortality, [25] but as an alternative performed competing risk analyses based on broad and narrow definitions for cause-specific death. Using Poisson regression, we modelled the relative risk conferred by decade of age, gender and ethnicity, adjusting for calendar year of exit to account for artifactual differences in the background death rates during the five year study.
We used excess mortality in the cohort to investigate RHD-attributable mortality before age 70 years in the wider population under the assumptions that all RHD patients in the country had been detected and no RHD deaths occurred before age 5 years. We calculated crude, age-specific and age-standardised RHD-attributable death rates for the general population, deriving 95% confidence intervals (CI) using Poisson, and using estimates of population from the Fiji Bureau of Statistics as the denominator. We tested the robustness of these results by changing the stringency of the record-linkage procedure through adjusting the posterior probability at which record pairs were considered to match. We also calculated YLL in each of the thirteen age categories using WHO Life Tables [26] and calculated rates for the wider population. We used the WHO World Standard population for direct standardisation. [27] Finally, the availability of ICD10 coded underlying cause-of-death classifications during 2011–2012 permitted comparison of our estimates with the number of reported deaths due to RHD itself and other conditions. To make the comparison, we collapsed ICD10 codes into the diagnostic categories used in the Global Burden of Disease (GBD) project, [28] grouping ill-defined codes separately. [29] To maximise parity we recalculated the number of RHD-attributable deaths based only on deaths associated with a death certificate.
We used a human capital approach [30] to define the cost of premature mortality to Fiji. We estimate wider cost of illness due to RHD/ARF in Fiji in a separate paper, providing further details and justification for the methods. [31] We estimated the cost of a death in each of the five-year age categories by multiplying life-expectancy in years using Fiji’s per-capita gross domestic product for the year in which the death occurred as estimated by the World Bank [32] discounting 3% each year. For each category and year of the study, we then multiplied this estimate by the excess deaths and summed the results to obtain the total cost of RHD-attributable deaths over the five-year period.
We searched for population-based studies of RHD mortality published in a thirty year period, 1985–2014, in Ovid Medline, Embase and Global Health (S4 Fig). We used the search terms “rheumatic heart disease” and “mortality” with a previously described filter to detect studies from developing countries. [33] We excluded case reports, case series, studies focused on valve surgery and studies not specific to RHD.
Permission for the study was granted by the Fiji National Research Ethics Review Committee (2013–89) in addition to the Oxford Tropical Research Ethics Review Committee (1055–13). Once the record-linkage procedures were complete, all data analysed were anonymised.
In total, 1,773,999 records were available, including 34,773 records that pertained to a death. Links were identified in the patient information system for 87.1% of control programme records, 85.3% of echocardiography clinic records and 66.0% of death certificates (S3 Table). After selecting eligible individuals with an RHD or ARF diagnosis, a cohort of 2,619 individuals remained for analysis (Fig 2). Of these, 1,038 (39.6%) were present in more than one database (S1 Fig). Characteristics are summarised in Table 1; the person-time observed totalled 11,537.5 person-years.
During follow-up, 430 of the 2,619 (16.4%) RHD/ARF patients were linked to a death in the patient information system, the death certificate database or both. This equated to 2.1% of the 20,796 deaths in the general population in the same age bracket during this time. The all-cause unadjusted death rate amongst RHD/ARF patients was 3.7% per year (95% CI 3.4–4.1%). Death rates based on cause-of-death information are summarised in S6 Table. From late childhood onwards, the death rate observed in the cohort exceeded that in the wider population (SMR 8.3, 95% CI 7.5–9.0, S2 Fig). The relative survival was 96.9% (95% CI 96.1–97.5%) at one year and 81.2% (95% CI 79.2–83.0%) at five years (S3 Fig). The risk of death among RHD/ARF patients increased with age over and above background rates; there was also increased risk for both male and iTaukei patients (S4 Table).
Based on the 378 excess deaths, of which 177 (46.8%) occurred before age 40 years, we estimated there were 9.1 RHD-attributable deaths (95% CI 8.2–10.1) per 100,000 person-years in those aged 0–69 years. This estimate remained stable to adjustments in the record-linkage threshold and between the 2008–2010 and 2011–2012 components of the dataset (S7 Table). Age-specific rates of RHD-attributable death increased throughout life (Fig 3A). Standardised to the portion of the WHO World Standard Population aged 0–69 years, our primary estimate equates to 9.9 deaths (95% CI 9.8–10.0) per 100,000 person-years (S5 Table). Additionally, we estimated 323.3 YLL (95% CI 317.9–328.6) per 100,000 person-years, equating to a WHO standardised rate of 331.0 YLL (95% CI 330.4–331.5) per 100,000 person-years (S5 Table). Age-specific RHD-attributable YLL rates were elevated from late childhood onwards (Fig 3B). The cost of these deaths for the five-year period was current Fiji Dollar $58,810,903, which at mid-market rates equates to United States Dollar $30,387,153.
Finally, we estimated there were 132 RHD-attributable deaths during 2011–2012 based only on deaths associated with a death certificate, compared to 81 RHD deaths reported in vital registration data (Fig 4). Only five other conditions caused more than 132 deaths in the wider population while ten caused more than 81 deaths (S8 Table). Moreover, only drowning caused more than the 40 deaths attributable to RHD at ages 5–29 years, which was greater than the number attributed to suicide and road injury, both well-recognised causes of death in young people.
These are the first national population-based age-standardised estimates of mortality due to RHD in a developing country (S4 Fig), and confirm that RHD is an important cause of premature death in Fiji leading to a substantial loss of life and economic productivity. The study was made possible by applying record-linkage techniques to the routine clinical and administrative data that are increasingly available in an electronic format in many developing countries. The results are robust to changes in record-linkage thresholds and remain broadly similar throughout the five years the study covers, despite changes in local death reporting practices during this time.
Using the background death rates in the general population, we were able to estimate RHD-attributable mortality by measuring excess mortality. These methods, which are widely used for population-based cancer survival analyses [13], are highly applicable to RHD, a disease for which cause-of-death information is often absent or unreliable. [6, 25] For example, if our results are compared to underlying cause-of-death classifications in vital registration data for 2011–2012, we find 1.6-fold more RHD-attributable deaths at ages 5–69 years, a discrepancy that peaked at 3.8-fold in the 30–49 years age group (Fig 4). This finding is consistent with a recent study of mortality amongst RHD patients in Western Australia which, by reviewing death certificates and other clinical data, concluded a third of RHD-attributable deaths were ascribed to other underlying cause-of-death diagnoses. [34]
Few existing data are available for comparison. One recently published study presents RHD-attributable death rates based on vital statistics for South Africa for the period 1997–2012 during which time the crude all-age death rate declined from 1.2 to 0.7 per 100,000 person-years. [35] As the authors acknowledge, however, the reliability of estimates based on death certification is questionable [35], particularly given that in 1999 both the UK and Japan reported RHD death rates over two-fold higher than this. [9] Interestingly, our age-standardised and age-specific rates are more comparable to those made for the Coloured population in South Africa for 1978–1982, the authors of that report estimating an age-standardised rate of 3.5 and 4.2 per 100,000 for men and women respectively. [36] Moreover, our age-standardised estimate is similar to that reported for Indigenous populations in New Zealand [37] and Alaska [38] during the 1970–1980s although it exceeds death rates reported from New Zealand [39] and Australia [25] more recently. Alternatively, we can compare our results to the estimates made by the GBD project (Fig 3). [28] Directly standardised to the population aged 0–69 years, we report higher death and YLL rates than GBD (S6 Table), the latter amounting to a 2.6-fold difference in the death rate and 2.4 in the YLL rate compared to the GBD’s developing countries estimate. Thus our data not only have important implications for the Pacific region but also, if generalisable to other developing countries, for global summary estimates.
Although the results appear reasonable, they have some limitations. First, there are potential shortcomings to using background death rates to estimate exposure-attributable mortality. If we were wrong to assume time-to-death from RHD was independent of time-to-death from other causes, we may have over-estimated RHD mortality; however, the impact would be small and alternatives such as cause-specific survival remain unsatisfactory. [13] Second, the cohort was heterogeneous with respect to the chronicity and severity of the illness, and there was no means to distinguish new onset from relapses of chronic disease. This led us to make the conservative assumption that participants were at risk from childhood onwards, which would lead to an under-estimate of mortality if the true onset was later. Third, the study was retrospective and relied on routine clinical and administrative data, which are likely to have contained errors. In particular, while death certificate submission in Fiji is relatively complete [17, 18], data pertaining to underlying cause-of-death should be interpreted with some caution; [18] our comparison with other causes of death may slightly exaggerate the disease’s importance. Fourth, these data provide neither sufficient detail nor follow-up to answer important outstanding questions about why such a burden of disease exists in this setting. For example, the discrepancy between the two largest ethnic groups remains unexplained, although a number of cultural, socioeconomic, geographic and potentially biological factors may contribute. Finally, we are unable to report on RHD-attributable deaths beyond age 69 years although we note GBD estimated a fifth of global RHD deaths occurred in this age group.
By illustrating the high burden of premature death due to RHD in Fiji, these data help substantiate the assertion that RHD remains, on a global scale, [6] an important cause of mortality. By using record-linkage techniques, we have demonstrated that routine clinical and administrative data can be used to quantify the impact of RHD in developing countries, a finding which has important implications for both research and disease control.
|
10.1371/journal.pntd.0005049 | Modelling Vaccination Strategies against Rift Valley Fever in Livestock in Kenya | The impacts of vaccination on the transmission of Rift Valley fever virus (RVFV) have not been evaluated. We have developed a RVFV transmission model comprising two hosts—cattle as a separate host and sheep and goats as one combined host (herein after referred to as sheep)—and two vectors—Aedes species (spp) and Culex spp—and used it to predict the impacts of: (1) reactive vaccination implemented at various levels of coverage at pre-determined time points, (2) targeted vaccination involving either of the two host species, and (3) a periodic vaccination implemented biannually or annually before an outbreak.
The model comprises coupled vector and host modules where the dynamics of vectors and hosts are described using a system of difference equations. Vector populations are structured into egg, larva, pupa and adult stages and the latter stage is further categorized into three infection categories: susceptible, exposed and infectious mosquitoes. The survival rates of the immature stages (egg, larva and pupa) are dependent on rainfall densities extracted from the Tropical Rainfall Measuring Mission (TRMM) for a Rift Valley fever (RVF) endemic site in Kenya over a period of 1827 days. The host populations are structured into four age classes comprising young, weaners, yearlings and adults and four infection categories including susceptible, exposed, infectious, and immune categories. The model reproduces the 2006/2007 RVF outbreak reported in empirical surveys in the target area and other seasonal transmission events that are perceived to occur during the wet seasons. Mass reactive vaccination strategies greatly reduce the potential for a major outbreak. The results also suggest that the effectiveness of vaccination can be enhanced by increasing the vaccination coverage, targeting vaccination on cattle given that this species plays a major role in the transmission of the virus, and using both periodic and reactive vaccination strategies.
Reactive vaccination can be effective in mitigating the impacts of RVF outbreaks but practically, it is not always possible to have this measure implemented satisfactorily due to the rapid onset and evolution of RVF epidemics. This analysis demonstrates that both periodic and reactive vaccination ought to be used strategically to effectively control the disease.
| Evaluation of the relative impacts of RVF vaccination has not been previously carried out. We present a model that simulates RVFV transmission between two livestock hosts (cattle as a separate host and sheep referring to both sheep and goats) and two mosquito species (Aedes and Culex species). We then apply the model to evaluate policy-relevant impacts of vaccinating (1) different proportions of animals at different times to the simulated outbreak, (2) either of the host species, and (3) different proportions of animals in a periodic biannual or annual vaccination preventative strategy. Vector population growth is dependent on rainfall extracted from the Tropical Rainfall Measuring Mission (TRMM) for an RVF endemic site in Kenya over a period of 1827 days. The model reproduces the 2006/2007 RVF outbreak reported in empirical surveys in the target area and other seasonal transmission events that occur during the wet seasons. Consistent with anecdotal evidence, mass livestock vaccination can greatly reduce the potential for a major outbreak. The model predicts that the effectiveness can be improved by increasing the proportion of vaccinated animals, targeting vaccination against cattle and strategically augmenting periodic preventative strategies with reactive strategies once a RVF outbreak is predicted.
| Rift valley fever (RVF) is a mosquito-borne viral zoonosis that causes periodic outbreaks accompanied by low-level virus activity during inter-outbreak periods mainly in sub-Saharan Africa [1]. The disease mainly affects sheep and goats, cattle and camels [2]. Humans can be exposed following a bite from an infected mosquito and or through direct contact with tissues from infected animals [3]. The disease was initially reported in restricted regions in Africa but has progressively spread to almost the whole continent, the island of Madagascar and the Arabian Peninsula [1][4]. The disease outbreaks often occur when favourable environmental drivers such as elevated and widespread rainfall and flat topography that promotes flooding [5] develop in areas where there are susceptible hosts [2] and competent mosquito vectors [6] and predisposing socio-economic practices such as herd replacement patterns [7].
Climatic factors seem to play a more dominant role as almost all historical outbreaks have been associated with cyclical patterns of the El Niño/Southern Oscillation (ENSO) phenomenon, which results in elevated and widespread rainfall over the Greater Horn of Africa (GHA) [5]. The outbreaks are often associated with adverse public health and economic impacts [8][9][10][11] as well as social impacts. Specifically, on a macroeconomic scale, Rich and Wanyoike [11] estimated that the 2006/2007 RVF outbreak in Kenya generated losses of over Ksh 2.1 billion (US$32 million then) on the Kenyan economy. The continued occurrence and geographical spread of RVF outbreaks points toward the need to understand the dynamics of the outbreaks as well as explore the approaches to their control.
Following the 2006/2007 RVF outbreak in Kenya, a retrospective analyses of the implemented responses revealed systematic delays due to the failure of the relevant institutions to recognize risk factors, act on early warnings messages (until the initial human cases were confirmed approximately two months after cases were observed in livestock) [12], and identify appropriate interventions. Consequently, stakeholders and decision-makers from the GHA region developed a risk-based Decision Support Framework (DSF) [12] that could be used to guide responses to similar emergencies in the future [12]. Livestock vaccination is one of the measures that were identified in the framework given that it has a good potential to reduce the impacts of the disease in livestock, contamination of the environment and subsequent exposure to humans [13].
There are many challenges that affect successful utilization of vaccines in the management of RVF outbreaks. First, the inter-outbreak period of the disease (approximated at 3–7 years [14]) is much longer than the shelf life of the currently available vaccine (Smithburn vaccine; 4 years) [12]. This discourages vaccine manufacturers from maintaining large stocks of these products given the risk of losing a large proportion of them through expiry. Most of these vaccines are often manufactured on order, for example, when the risk of an outbreak heightens. Secondly, the heavy rains and flooding that characterizes the high risk periods limit access and hence the delivery of vaccines to the rural areas. Thirdly, livestock species that are highly susceptible to the disease and hence would benefit from vaccination (such as goats and sheep) have a high population turn-over rates, limiting the maintenance of herd immunity especially in the pastoral areas. These challenges indicate an urgent need for policies that can guide utilization of RVF vaccines.
Mathematical models for simulating RVF epidemics have been developed [15] [16] [17] [18]. However, most of them are not suitable for evaluating vaccination strategies because they do not incorporate (i) climate variability (mainly precipitation changes) which greatly influences the timing of vaccination and other reactive interventions, and (ii) livestock population dynamics which influence the duration of herd immunity. We develop a model comprising two hosts—cattle as a separate host and sheep and goats as one combined host—and two vectors—Aedes species (spp) and Culex spp. Consequently, the model incorporates these components and use it to address policy-relevant questions on the effectiveness of reactive and periodic vaccination strategies including: (1) How can various vaccination coverages (VCs) implemented at different times before an outbreak affect the size of an outbreak in livestock? (2) To what extent is it possible to reduce outbreak size in both livestock species by focusing vaccination on one species? (3) How can periodic vaccination be used together with reactive vaccination particularly in the high risk areas? We incorporate two hosts with the recognition that pathogens such as RVFV that can infect multiple host species have different dynamics than single-host pathogens. Faced with scarcity of host-specific transmission parameters, this study sets the stage for the understanding of pathogen transmission dynamics and cost-effective control of RVF in multihost disease systems.
In developing the model, we make the following assumptions:
Two probability distributions generated using the fuzzy and logistic regression models based on TRMM rainfall values are successfully used to drive Aedes and Culex mosquito populations, respectively. Fig 2 shows the temporal relationship between these probability distributions and the respective vector:host ratios. In general, peaks in vector:host ratios lag those of fuzzy and logistic probability distributions by approximately 8–17 days and 30 days, respectively. Between days 9256 and 9450 when there was heavy/persistent rainfall, the fuzzy and logistic regression models generated high probability values which led to an upsurge in the mosquito populations, hence high vector:host ratios (Fig 2). The other wet seasons before this had short-lived precipitation events that were not adequate to support an upsurge of the Culex mosquito population though that of Aedes mosquitoes responded positively.
In the simulated outbreak, Aedes adults that emerge from infected eggs, last for a total of 148 days and peak at day 80. Susceptible Aedes mosquitoes also develop at the same time peaking on day 87. Culex mosquito population appears 36 days after the emergence of Aedes population. Culex mosquitoes gain RVFV infection from viraemic hosts from day 69 after initial transmissions by Aedes spp. The maximum FoI exerted to Aedes spp from cattle and sheep are 0.016 and 0.006 respectively. The maximum FoI exerted to Culex spp from cattle and sheep are 0.015 and 0.0057 respectively.
Predicted RVFV incidence in hosts is shown in Fig 3. These predictions show five transient RVFV transmissions associated with seasonal rains and one main outbreak associated with heavy and persistent precipitation. In general, seasonal transmission events fail to result in full-blown outbreaks given that no amplification of populations of Culex spp occurs (Fig 2). The outbreak curve has a characteristic shape–RVFV activity begins slowly until Culex spp population surges, resulting in the amplification of the virus. The predicted peak outbreak incidence of RVFV in cattle is 12% on day 112 of the outbreak while that for sheep is 8% on day 123. The predicted duration of the outbreak is 184 days. The maximum force of infection exerted to cattle and sheep are 0.24 and 0.06, respectively.
A simulated RVF outbreak in this study was defined by noting the predicted peak endemic incidence in hosts. The peak endemic incidence was used as the threshold for definition of an outbreak. By comparing endemic verses epidemic patterns predicted in the model, it appears that the number of cases predicted during the outbreak captured is 80% more than those predicted for the endemic periods. We use the 80% threshold for evaluating impacts of the various vaccination scenarios being studied.
Sensitivity analyses showed that the infectious period and infectivity in both hosts and vectors (particularly Culex spp) were sensitive to the cumulative incidence of RVF. Others included survival and mosquito biting behavior of Culex spp (Table 2).
We present a deterministic model that combines precipitation patterns, mosquito population dynamics and host demographics to simulate RVFV transmission. The model predicts elevated RVFV activity during the wet seasons as well as a full-blown RVF outbreak following periods with excessive and persistent precipitation. Elevated and persistent rainfall is a risk factor for RVF outbreaks—all the 11 reported RVF outbreaks in Kenya occurred in years when the average annual rainfall increased by more than 50% in the affected districts [14].
The novelty of the model is in the bridging of separate probability distributions that uses satellite-derived daily precipitation for the study area that ensure temporal succession of separate vector species population growths. Since we are not interested in the importance of trans-ovarial transmission and its implications on the generation of the outbreak [17], we exclude these detailed dynamics in Aedes mosquitoes. Adult Aedes mosquito emergence events are dependent on water (rainfall) that inundates breeding habitats [21]. We, therefore, base the dynamic distribution of Aedes species on accumulated rainfall amounts using a fuzzy distribution model similar to that employed by Emert et al. (2011) [20]. The fuzzy distribution model computes dynamic suitability conditions of hatching of Aedes eggs that mimic the reported strong relationship between Aedes mosquito emergence and weather (rainfall) variability [21]. The assumptions driving the fuzzy distribution model, as described in the Methodology section, seem rational and might denote a qualitatively plausible relationship of Aedes egg hatching process than a simple linear function of rainfall.
Culex mosquito population dynamics are driven using an approach of obtaining parameters from a statistical analyses of reports of livestock cases and a particular pattern of rainfall during the 2006/2007 outbreak. We used this function based on empirical studies that reported that the mosquito breeding sites were colonized by massive swarms of Culex (and other species) if they remained flooded for at least 28–42 days [21]. Additionally, livestock keepers in the study area reported a mean average of 23 days between the start of heavy rains and the appearance of mosquito swarms during the 2006/2007 RVF outbreak [8], though most likely these included both primary and the secondary species. Our model accurately captures this temporal relationship between cumulative rainfall and secondary mosquito species emergence.
A different approach of growing seasonal vector populations in modelling RVFV transmission in West Africa was implemented by Soti et al. [37] using a hydrology model. Their hydrological model uses daily rainfall as input to simulate variations of water pool surface areas. We have not used this approach as the epidemiology of RVF occurrence in West Africa and GHA is different. Whereas in GHA RVF outbreaks are known to be closely associated with ENSO phenomenon [5], periods of RVF outbreaks in West Africa do not necessarily coincide with years of highest total rainfall [38]. Indeed, RVF epidemiological landscape in West Africa is influenced by the generation of temporary ponds and a particular rainfall temporal distribution (populations of Aedes and Culex spp depend on the alternation of rainy and dry periods) [38]. Although rainfall, just as in GHA, is the main driver of hydrologic dynamics of water pools in West Africa, the mechanistic vector productivity of specific habitats and RVFV transmission and the consequent epidemiological inference in the two ecologies can be substantially different. Empirical studies are needed in the two distinct ecologies to accurately quantify the amount and distribution of rainfall regimes (and how they interact with soil infiltration rates) required for hatching of primary vectors.
We implement the legendary assumption which considers primary and secondary vectors playing a synergistic role in generation of RVF outbreaks. Innovative ways of empirically examining these assumptions are needed to answer questions such as whether primary vectors alone [17] or whether secondary vectors alone (for example, if augmented with movements of animals) [7] can drive RVF full-blown outbreaks. In addition, this model hypothesizes that water availability may play a more dominant role in driving the vectors population dynamics. Future model refinements should incorporate not only the effects of temperature and humidity, vegetation and nutrient competition on vector population dynamics but also on the extrinsic incubation periods of RVFV in vectors [39].
For RVF control to be evaluated and optimum control strategies devised, an increased understanding of the transmission dynamics among hosts and vectors is paramount. In this way, we apply the model to identify the key factors driving the number of potentially averted RVF cases in a simulated outbreak. The analyses show that vaccination, as a sole intervention, can be effective in mitigating the impacts of RVF outbreaks. The success of RVF vaccination is predicted to be defined by the targeted vaccination coverage and the time to the outbreak. The proportion of cases averted is related to the targeted vaccination coverage, particularly for low levels. The policy implication of this prediction is that resources and planning required to achieve a given VC corresponds to the number of cases expected to be averted. For a given VC, higher herd immunity at the outbreak onset is predictably highly beneficial. Vaccinating early reduces herd immunity, over time, through removal of immune animals via expected mortality and offtake and birth of susceptible animals. The model predicts that 3–6% more cases can be averted if, for the simulated VCs, vaccination is implemented close to the outbreak. Averting 3–9% more cases can lead to large numbers of deaths being averted particularly in the more RVF-induced mortality susceptible species such as sheep. For greater effectiveness, this prediction implies that a careful balance between a given VC and optimal timing is critical. These predictions concur with recent modelling study predictions that a higher rate of vaccination may help to reduce the epidemic size and a maximal attempt of vaccination just before an outbreak is highly beneficial [18]. In sub-Saharan Africa, vaccination against RVFV has been used for many years either to prevent disease occurrence [40] or to mitigate disease impacts [41]. Our model predictions clearly demonstrate the usefulness of effective implementation of this intervention. Ideally, however, all members of a population need not be vaccinated because as the number of susceptible hosts in the population is reduced, the efficiency with which a pathogen is transmitted is greatly reduced (the concept of herd immunity) [42]. The model predicts that this indirect protection is accelerated as vaccination coverage is increased and, moreover, it is experienced more in sheep relative to cattle.
Early and optimal timing, in turn, depends upon a sensitive and functioning RVF surveillance and prediction system and a rapid response capacity by the national veterinary authorities [40]. One such surveillance system integrates ENSO related climate anomalies including elevated sea-surface temperatures and satellite-derived normalized difference vegetation index data (NDVI) [5]. During the 2006/2007RVF outbreak, this system retrospectively provided a 2 to 4 month period of warning in the GHA region [5]. However, the RVF DSF estimates the lead-time to order, produce, deliver sufficient vaccine to the field and attain herd immunity in livestock to be approximately 5 months [12]. This implies that vaccine orders need to be placed prior to the first RVF early warning. Currently, this is impractical unless the lead time for prospective predictions of RVF outbreaks is lengthened. Still, even if the latter were achieved (to, e.g. 5 months), mobilizing adequate resources to procure the vaccines within the short period is a difficult task in resource-scarce countries in the GHA. Moreover, by this time, the co-occurrence of heavy rains and flooding in the rural areas coupled with the absence of all-weather roads can present huge logistical challenges in vaccine delivery. Innovative strategies are clearly needed as part of outbreak preparedness plan.
To overcome some of these challenges, the RVF DSF proposes a strategic regional vaccine shared bank which could be rapidly deployed in times of need [12]. To supplement this proposition, we modeled a periodic vaccination strategy implemented under different vaccination coverage biannually or annually for 2 years in advance of an outbreak. The objective was to assess the impacts of these strategies in not only reducing the outbreak size but also the possibility of complementing them with a reactive strategy close to the outbreak onset. Complementing very low VCs biannually for two years and low reactive VCs is highly effective, e.g. a VC of 10% is predicted to completely avert an outbreak when integrated with a reactive VC of 35%. Annual vaccination is equally effective though at a lower scale. In a large livestock population, averting an outbreak could mean avoiding morbidity and mortality of thousands of animals, reducing vulnerability of local livestock-dependent livelihoods and national economies and, more importantly, reducing chances of virus exposure to humans. Rift Valley fever vectored vaccines are currently being developed [43] and evaluated [44] and this might change (i) the way these vaccines are administered in the field, i.e., some could be given at biannual intervals and or others annual. These combinations can influence the efficacy of the RVFV component of the vaccine. As earlier highlighted in this paper, the shelf-life of current vaccines [12] is shorter than the average inter-outbreak period [14] which presents an economic disincentive to vaccine manufacturers in situations where reactive vaccination campaigns are planned. Similarly, resource-constrained governments are not keen on funding periodic vaccination campaigns partly due to unpredictability of occurrence of the outbreaks. Periodic vaccination campaigns are also a disincentive in situations where livestock population-turn over due to offtakes and expected mortality temporally leads to lower herd immunity. Our analysis is therefore well placed to give policy directions on how vaccination can be used to meet these challenges. Further evaluation of the response impact of integrating periodic and reactive vaccination strategies in preventing the occurrence of a RVF outbreak is an important area for future research and policy development.
Multihost pathogens are more likely to have ecologically different dynamics than pathogens that infect only a single host species. In a host population, multiple host species can be viewed as a form of heterogeneity that partitions the total host population into subpopulations between which the FoI experienced by each host species and the FoI exerted by each host species varies [45]. Based on the assumptions we make in the model, the FoI experienced by cattle is larger due to their higher vector: host ratio arising from their lower number in the population and their assumed larger surface area relative to sheep. The assumptions are qualitatively realistic given the differences in host species’ exposed surface area which is obviously higher in cattle relative to sheep, all other parameters (e.g. blood meal preference) being constant. Similarly, the FoI experienced by vector species from cattle are higher than that from sheep. Consequently, based on our assumptions, the model predicts that cattle dominate the bi-directional RVFV transmission process between hosts and vectors.
With such a pathogen ecological framework where a host species may dominate the virus transmission, we examined the possibility of directing control against either of the host species. Targeting cattle alone provided major protection to cattle and sheep. This benefit arises from the reduction of transmission of RVFV. However, targeting sheep alone provided protection to sheep alone. This prediction, if empirically validated has important policy implications for the implementation of both periodic and reactive vaccination strategies for two reasons: (1) cattle have longer lifespan and lower population turnover relative to sheep and, therefore, would be able to sustain herd immunity for longer, (2) in our case study area (and indeed in all pastoral areas), cattle are fewer relative to sheep (and goats) and (3) in the pastoral communities, cattle are likely to be moved long distances translating to potential spatial spread of RVF compared with sheep and goats. These reasons can greatly influence the cost-effectiveness of a strategy that focuses control against cattle in the population.
Our model, by necessity, includes a number of simplified assumptions about reality in a number of ways that have a bearing on the predictions. We have assumed transmission-related parameters in Table 1 to be similar in both host species. This implies that our outputs were based on two main parameters (i) the use of temporally varying FoI arising from seasonal growth of vectors, and (ii) the different numbers of host species in the population. However, model sensitivity analyses found that RVF cumulative incidence may be influenced most by infectious periods and infectivity in both hosts and vectors (particularly Culex spp). Other sensitive parameters include survival and mosquito biting behavior of Culex species. The same parameters have been reported to be sensitive to similar outcomes in RVF modelling, e.g. Chitnis et. al [17] reported that an outbreak size was sensitive to vector-to-host ratio, mosquito biting rate and the infectivity of hosts. Other models [15][16][35] reported adequate contact rates between vectors and hosts and the rate of recovery livestock as sensitive to the basic reproduction number. However, their definition of adequate contact rates between vectors and hosts considered a composite term whereas in our model, we disaggregated the term into its individual components including the vector biting rate, host infectivity, blood meal index and vector host ratio. These findings suggest that apart from RVF vaccination, reducing the the probability of transmission from the vector to the host can be effective in RVF outbreak control. In addition, given the importance of understanding RVFV transmission processes, the lack of knowledge about the processes make gathering of relevant field and experimental data on these biological processes an urgent research priority.
Further simplifying assumptions that we make in the analyses of vaccination impacts ignore the individual components that constitute the actual proportion of susceptible hosts vaccinated (herein referred to as an ideal VC (iVC)). This can be obtained as a product of the proportion of vaccinations properly applied (efficiency of vaccination) and the probability that the vaccine would provide protection from infection (the efficacy of the vaccine) [46], both of which limited data are available for RVF. Naturally, these two proportions are each less than 100% in most cases. Multiplicatively, the further the values are from 100%, the less the iVC. A recent study evaluated the effectiveness of RVF Clone 13 vaccine and reported that 67% of vaccinated cattle and between 91% and 97% of vaccinated sheep and goats develop protective antibodies to the vaccine [44]. Applying an efficiency of vaccination of 80% based on the performance of mass vaccination teams as assessed by the Pan African Rinderpest Campaign in pastoral areas in GHA [47], vaccinating an entire population in our study would result to an iVC of approximately 54% in cattle and 76% in sheep. To achieve our simulated VCs, therefore, call for high levels of both the efficiency of vaccination and high efficacy of RVF vaccines such as that reported in sheep and goats [44]. Further explorations required include cost-effectiveness analyses taking account of integrating VCs and time to outbreak, integrating periodic and reactive strategies and directing interventions to one host species under different scenarios of efficiency of vaccination and efficacy of vaccines.
In conclusion, our results suggest that targeted vaccination can be effective in mitigating the impacts of RVF outbreaks. However, it is not always possible to have this measure implemented satisfactorily due to the rapid onset and evolution of RVF epidemics. The analyses further demonstrates that both periodic and reactive vaccination ought to be used strategically to effectively control the disease. In addition, challenges associated with prediction of the outbreak, availability and delivery of vaccines need to be addressed. Factors driving the number of potentially averted cases include the targeted VC and timing of vaccination in relation to the time to the outbreak. Based on our assumptions, cattle appear to dominate RVF transmission between hosts and vectors. Predictably, directing vaccination against cattle, whether in a periodic and/or a reactive vaccination startegy, may be more effective as it confers herd immunity to both species. The work presented here advances our understanding of impacts of different vaccination strategies. We consider that these predictions provide a first step of information needed by policy makers to plan effective periodic and reactive strategies for mitigating the effects of RVF outbreaks. However, detailed cost-benefit analysis should be integrated with these findings to support decision-making and prioritize these strategies.
|
10.1371/journal.pgen.1000287 | Genetic Analysis of Human Traits In Vitro: Drug Response and Gene Expression in Lymphoblastoid Cell Lines | Lymphoblastoid cell lines (LCLs), originally collected as renewable sources of DNA, are now being used as a model system to study genotype–phenotype relationships in human cells, including searches for QTLs influencing levels of individual mRNAs and responses to drugs and radiation. In the course of attempting to map genes for drug response using 269 LCLs from the International HapMap Project, we evaluated the extent to which biological noise and non-genetic confounders contribute to trait variability in LCLs. While drug responses could be technically well measured on a given day, we observed significant day-to-day variability and substantial correlation to non-genetic confounders, such as baseline growth rates and metabolic state in culture. After correcting for these confounders, we were unable to detect any QTLs with genome-wide significance for drug response. A much higher proportion of variance in mRNA levels may be attributed to non-genetic factors (intra-individual variance—i.e., biological noise, levels of the EBV virus used to transform the cells, ATP levels) than to detectable eQTLs. Finally, in an attempt to improve power, we focused analysis on those genes that had both detectable eQTLs and correlation to drug response; we were unable to detect evidence that eQTL SNPs are convincingly associated with drug response in the model. While LCLs are a promising model for pharmacogenetic experiments, biological noise and in vitro artifacts may reduce power and have the potential to create spurious association due to confounding.
| The use of lymphoblastoid cell lines (LCLs) has evolved from a renewable source of DNA to an in vitro model system to study the genetics of gene expression, drug response, and other traits in a controlled laboratory setting. While convincing relationships between SNPs and mRNA levels (eQTLs) have been described, the degree to which non-genetic variables also influence phenotypes in LCLs is less well characterized. In the course of attempting to map genes for drug responses in vitro, we evaluated the reproducibility of in vitro traits across replicates, the impact of the EBV virus used to transform B cells into cell lines, and the effect of in vitro culture conditions. We found that responses to at least some drugs and levels of many mRNAs can be technically well measured, but vary both across experiments and with non-genetic confounders such as growth rates, EBV levels, and ATP levels. The influence of such non-genetic factors can both decrease power to detect true relationships between DNA variation and traits and create the potential for non-genetic confounding and spurious associations between DNA variants and traits.
| Genetic mapping offers an unbiased approach to discover genes and pathways influencing disease traits and responses to drugs and environmental exposures [1]. Unlike model organisms that can be exhaustively phenotyped and readily exposed to drugs and toxins in the laboratory, there are substantial limits to the phenotypes that can be safely elicited or measured in human subjects. Thus, there would be great value in a human in vitro model that faithfully reflects both in vivo genetics and physiology while allowing for systematic perturbation and characterization in high throughput. Such a model would be particularly useful to study the function of sequence variants mapped by whole genome association studies of common human diseases that do not fall in obvious coding sequences [2]–[6], many of which are presumed to influence disease traits through subtle effects on gene regulation.
One such model system has been proposed and extensively studied: EBV-transformed lymphoblastoid cell lines (LCLs) derived from human B-lymphocytes [7]–[13]. Lymphoblastoid cell lines have long been produced as renewable sources of DNA as part of normal and diseased cohorts. Initially, LCLs derived from genotyped CEPH pedigrees [14] and HapMap participants [15] were used to identify genomic regions linked to and associated with inter-individual variation in mRNA transcript levels (these “expression” QTLs are referred in the text below as “eQTLs”) [16]–[19]. A small number of such eQTLs have been found to also be associated with human disease [20]–[22]. LCLs have also been used to search for genetic variants that predict for response to radiation and drugs in vitro [23]–[26]. Some investigators have performed joint analysis of eQTLs and drug response QTLs, seeking non-random relationships between genotypes at single nucleotide polymorphisms (SNPs), baseline mRNA levels, and response to chemotherapeutic agents [27],[28]. One recent study reported identification of eQTLs that explain up to 45% of the variation seen between individuals in cell sensitivity to chemotherapy [28].
The utility of genetic mapping in LCLs is a function both of how well LCLs reflect the in vivo biology of the people from whom they were collected, and the ability to eliminate potential sources of confounding that could reduce power and cause spurious associations between cell lines (and the DNA variants they carry) and traits. While the DNA sequence of an LCL is typically a stable representation of the human donor [29], relatively less is known about the stability of cellular traits studied in vitro, and how they are influenced by non-genetic factors. Certainly, there are many opportunities for non-genetic factors to be introduced in the path from the human donor to the study of an LCL in vitro (Figure 1): the random choice of which subpopulation of B-cells are selected in the process of immortalization, the amount of and individual response to the EBV virus, the history of passage in cell culture and culture conditions, the laboratory protocols and reagents with which assays are performed, and the measurements used to assess drug response and mRNA phenotypes.
Encouraged by previous studies and the emerging HapMap resource, we set out to use LCLs to map genetic contributors to drug response in LCLs. In the course of this work we examined the relative contributions of DNA sequence variation, biological (day-to-day) variability, and confounders such as growth rate, levels of the EBV virus, ATP levels, and cell surface markers [30]. We investigated these factors in relation to two classes of phenotypes – drug response and mRNA expression levels. We find that inter-individual rank order based on both drug responses and mRNA expression levels is only modestly reproducible across independent experiments. Measurable confounders (in vitro growth rate, EBV copy number, and cellular ATP content) correlate more strongly and to a larger fraction of traits than do DNA variants. Even after correcting for confounders, and after integrating both eQTLs and mRNA correlations to drug response into a single model, we were unable to find convincing evidence for QTLs associated with drug response. Our observations suggest that, in addition to larger sample sizes, careful attention to influences of potential confounders will be valuable in the attempt to perform genetic mapping of drug responses in LCLs in vitro.
We studied 269 cell lines densely genotyped by the International HapMap Project [31]. Cell lines were cultured under a structured protocol and characterized at baseline for a variety of cellular phenotypes including growth rate, ATP levels, mitochondrial DNA copy number, EBV copy number, and measures of B-cell relevant cell surface receptors and cytokine levels. Each cell line was exposed in 384-well plates to a range of doses for each of seven drugs selected based on their divergent mechanisms of action and importance in clinical use for treatment of B-cell diseases, focusing on anti-cancer agents: 5-fluorouracil (5FU), methotrexate (MTX), simvastatin, SAHA, 6-mercaptopurine (6MP), rapamycin, and bortezomib. Drug response was measured using Celltiter Glo, an ATP-activated intracellular luminescent marker that, when compared to mock-treated control wells, can represent relative levels of cellular viability and metabolic activity. Data can be downloaded from the Broad Institute web site: http://www.broad.mit.edu/mpg/pubs/hapmap_cell_lines/.
Total RNA was collected at baseline and mRNA transcript levels (hereafter referred to as “RNA”) were measured genome-wide on the Affymetrix platform. Expression data is available on GEO Accession # GSE11582. For QC and normalization details, see Materials and Methods.
Baseline characterization and plating for drug response experiments was performed in batches of 90 cell lines from each HapMap analysis panel (CEU, JPT/CHB, and YRI) on each of three experiment days. The order of cell lines within each panel was randomized to avoid inducing artificial intra-familial correlation. Each drug was tested at a range of doses around the expected IC50 as reported for the drug by the NCI DTP; each dose of drug was tested in two wells per plate and on two separate plates. These replicate measurements for each cell line allowed assessment of intra-experimental variation.
To evaluate day-to-day (i.e. inter-experimental) variation in all traits, a subset of 90 cell lines (30 from each of the three HapMap panels) was grown from freshly thawed aliquots and the entire experiment was repeated. To evaluate the effect of technical error on measured RNA levels, a set of 22 RNAs previously expression profiled (using Illumina HumanChip) at Wellcome Trust Sanger Institute (WTSI) was included in expression profiling at the Broad on Affymetrix arrays.
Gene mapping of drug response (or any cellular phenotype) in LCLs requires that the phenotype be: (1) technically well measured, (2) biologically reproducible across independent experiments, and (3) remain relatively free from confounding factors. We assessed each of these characteristics in turn before performing genome-wide association scans.
To evaluate variability in drug response across replicate plates assayed on a given experiment day (technical reproducibility), we calculated the “relative” response of a cell line to each drug by measuring the (signed) distance of that cell line's dose-response curve for the drug on a given plate to the dose-response curve for the drug averaged across all cell lines assayed that day, in that replicate plate set. (The two replicate plates for each cell line performed on an experiment day were arbitrarily placed into set A or B.) This non-parametric approach allowed all drugs to be treated uniformly (see Methods) and generated two data points per cell line, per drug, per day. We ranked the cell lines based on their relative response in plate set A and separately based on values from plate set B. The rank-correlation (Spearman's rho) for relative response across sets A and B was high (rho = 0.86 to rho = 0.99, Table S1), indicating that drug response on a given day is both highly reproducible and technically well measured in this experimental design.
To evaluate variability across independent experiments on separate days (biological reproducibility), we repeated the assay on a subset of ∼90 cell lines (30 from each of the three HapMap analysis panels). (At this point, we noted that our assays for rapamycin and bortezomib suffered from weak responses and strong dependence on drug batch, respectively, and removed these drugs from future analysis; see Methods for details). For the remaining five drugs, cell lines were ranked based on relative response on day 1 and again on day 2 as above, and the rank-correlation (Spearman's rho) was calculated. In comparison to the high technical reproducibility on a given experimental day, inter-cell line variability in drug response was much less reproducible across independent experiments (rho = 0.39–0.82, Table S2).
We noted that the rank order of cell lines based on relative drug response was strikingly similar between three drugs (5FU, 6MP, and MTX). In fact, the rankings of cell lines based on these three drugs were as similar to one another as to rankings based on biological replicates of the same drug on different days (Figure 2A and Table S3). Wondering if this observation was limited to our dataset, we examined the publicly available data of Watters et al.[25] (Figure 2B). We found a very similar correlation of relative response to a distinct pair of drugs, 5FU and docetaxel, in their experiments. (This correlation likely explains why these investigators found linkage for both drugs to the same genomic locus.) Such a correlation in relative response to multiple drugs could, in theory, indicate a shared genetic mechanism common to many drugs, but it could also suggest the influence of an experimental confounder that more strongly influences drug response than does genetic variation.
We searched for and identified one such confounder: the baseline growth rate of the individual cell lines was highly correlated to the relative responses to these drugs (Figure 2C; Table S3). Growth-rate was modestly reproducible across days (rho = 0.37), with very limited evidence for heritability (h2 = 0.35; pval = 0.08). (We note that our study is not well-powered to detect h2<0.5 (Figure S1).) The dependence of drug response on growth rate in LCLs, though not previously reported, is unsurprising: all three agents depend upon cell division. Using a differential equation model of drug response accounting for the kinetics of exponential growth under exposure to drug (see Methods), we estimated a growth rate adjusted EC50 for each cell line for each of the three affected drugs. This approach removed the bulk of the correlation between drug responses and between drug response and growth rate (Table S4), though some correlation of responses persisted. Standard EC50s were fit for Simvastatin and SAHA.
Given the residual correlation across drugs, we searched for other non-genetic confounders. Baseline ATP concentrations (estimated based on the average of Celltiter glo values for all mock-treated wells, see Methods) were correlated to the growth rate adjusted EC50s for MTX and 5FU (Figure 2D). Like growth rate, ATP levels were reproducible across biological replicates (rho = 0.6) without statistically significantly evidence for heritability (h2 = 0.19, pval = 0.12). After further adjusting the growth rate adjusted EC50s for MTX and 5FU for ATP levels using linear regression, the correlation across drugs was nearly abrogated (Table S5).
Having adjusted for confounding due to growth rate and ATP levels, and largely eliminating correlations across drugs that were attributable to in vitro rather than inherited influences, we performed genome-wide association studies. Specifically, we examined the relationship between the EC50s for each drug and SNPs from HapMap Phase 2 with Minor Allele Frequency (MAF) >10% [32]. We did not observe any associations that surpassed genome wide significance (p-val<5e-8). The study was well powered to detect only strong QTLs, those that explain >15% of the variance in drug response (Figure S2). Nonetheless, the distributions of statistical association between SNPs and EC50s did not significantly exceed expectation under the null hypothesis. Our lack of evidence for association between SNPs and drug responses is consistent with prior publications [24]–[28], none of which identified specific SNPs that exceeded genome wide significance.
Previous studies observed baseline levels of RNA expression correlated to response to cisplatin and etoposide [24],[27],[28]. A correlation does not imply a causal contribution to drug response, as a third factor could simultaneously affect both phenotypes. Nonetheless, in the effort to identify a subset of genes whose regulation may truly influences drug response, it may be valuable to integrate information on SNP associations with RNA levels (eQTLs) and RNA correlations to drug responses. We therefore turned our attention to RNA measurements in LCLs.
As with drug response, genetic mapping of variants that influence RNA expression requires that interindividual variation in RNA levels is (a) reproducible on a given day, (b) reproducible across experiments performed on different days, and (c) influenced by genetic variation to a greater extent and independent of confounding by experimental artifacts.
One common metric for evaluating reproducibility in expression data is to rank the level of expression of all genes in a given sample, and to compare these ranks of genes (relative to one another) to those obtained in a separate hybridization of another aliquot of the same RNA (technical replicates) or in RNA from the same cell line on a different day (biological replicates). When we assessed the reproducibility of ranked RNA levels using this metric, we observed a high correlation across biological replicates: (Figure 3A – black curve). Moreover, we observed a similar correlation between profiles from any pair of unrelated individuals (Figure 3A – red curve), and across human cell lines in comparison to those from chimpanzee (Figure 3A – blue curve). What this reflects is the simple fact that the dynamic range in expression levels across genes is stable across primates, and much larger in magnitude than the inter-individual variation in the level of any given gene.
A more relevant metric for gene mapping is the reproducibility in rank order of different individuals based on the level of expression of a given gene. If the level of a single RNA transcript in one individual is reproducibly higher than the same RNA transcript in another individual, then it may be possible to identify genetic variants contributing to inter-individual variation of this RNA transcript (i.e. an eQTL). In contrast, if variation in the level of an RNA transcript across individuals is low relative to the technical and biological noise in a single individual, then there will be limited power to map genetic influences that alter expression of the gene.
We examined inter-individual variation in RNA levels for each of 3,538 genes measured to be expressed in the cell lines (using standard criteria for expression arrays). The analysis included LCLs from 49 unrelated individuals that were independently thawed, cultured and profiled on two different days (Figure 3B). In contrast to the results in Figure 3A, which showed excellent technical reproducibility, we see that the rank-correlation of individuals on different days (based on measured levels of individual genes) is typically modest (rho = 0.25–0.3). That is, in our experiment, only a fraction of the 3,538 RNA transcripts examined in LCLs vary reproducibly between individuals relative to technical and biological noise.
To parse the contributions of technical and biological noise, we examined the reproducibility of rank orders of cell lines when aliquots from the same RNA sample were profiled on two different array platforms. Specifically, RNAs for 14 unrelated individuals (from YRI HapMap subset) were profiled using the Illumina system at WTSI, and these same RNA samples were profiled on Affymetrix microarrays at Broad. To evaluate the contribution of technical measurement error, we calculated reproducibility in the rank order of individuals based on these technical replicates. We observed a median rank-correlation of rho = 0.55 (Figure 3C – gold curve), much higher than the biological reproducibility observed when two RNA samples for the same 14 individuals were independently prepared in a single lab and expression profiled on the same platform (rho∼0.3, Figure 3C – green curve). Thus, biological variation in RNA expression is greater than measurement error, even across different technologies.
To further minimize the impact of technical measurement error, we henceforth restricted analysis to one thousand genes that displayed the greatest technical reproducibility in rank ordering individuals (rho>∼0.7, median rho∼0.85). Genes excluded by this threshold include both those that are technically well measured but invariant across individuals, and those for which inter-individual variation is obscured by technical noise. (As the WTSI performed four technical replicates while Broad performed only a single technical replicate, WTSI data had lower overall variance.) Genes excluded by this filter typically varied less across individuals, particularly in the better-measured WTSI dataset. (median standard deviation of 1000 best-measured genes = 0.27 vs 0.17 for the other ∼2500 expressed genes; p-val<1e-15).
When analysis was limited to these one thousand genes, the correlation across biological replicates improved but was still modest (rho = 0.55, Figure 3D – cyan). That is, despite excellent technical reproducibility overall (Figure 3A) even relative to inter-individual variation (Figure 3C), the rank order of individuals based on most genes was only partially reproducible.
We reasoned that some of the biological noise might be due to other measured factors, as had been the case for drug response. Using a threshold of 5% variance explained, growth rate was correlated to levels of expression of only relatively few genes (<5%). In contrast, ∼15% of genes showed correlation to EBV copy number (Figure 4A), some of which encode genes known to participate in transduction pathways downstream of EBV signaling [33],[34],[35]. Moreover, the level of expression of >25% of genes was correlated to ATP levels (Figure 4B). In total, over 40% of genes have at least 5% of their variation in RNA levels correlated to one of three confounders above (Figure 4E).
The correlation of RNA levels to such factors could, in principle, represent intrinsic characteristics of each LCL (which could potentially be due to inherited DNA sequence variation, acting indirectly through susceptibility to EBV infection or inducing a metabolic state). Alternatively, growth rate, EBV infection, and metabolic state could represent experimental artifacts that obscures genetic contributions to gene expression variation. Interestingly, measurements of EBV copy number, ATP level, and growth rate at Broad correlate to levels of RNA expression generated independently at WTSI [18],[19] (Figure 4F), albeit more weakly than for the expression profiles generated on the same samples at the Broad. Thus, these confounders display a component intrinsic to each cell line, as well as a substantial component that is not a reproducible attribute of the cell line.
To examine how much of the variability in gene expression might be demonstrably attributed to inherited DNA variation, we searched for cis-eQTLs associated with RNA expression levels in our experiment. Using HapMap Phase 2 SNPs with MAF>10% that lie within a 0.15 Mb window around each gene, we performed standard linear regression between expression values of that gene and SNP genotypes coded 0,1,2 (representing the number of minor alleles carried by the individual). In our dataset, ∼9% of genes harbored a cis-eQTL that explained 5% or more of the gene's variance in expression levels (Figure 4C, reporting the excess of genes compared to permuted datasets). Even more eQTLs were evident in the WTSI expression data (which, due to the use of four technical replicates, has lower technical noise): >20% of genes were associated with a SNP that explains 5% or more of the variance (Figure 4D).
Consistent with previous analyses [16],[17],[18], in both data sets only a small fraction of genes displayed a cis eQTL that explained a large proportion of variance in RNA levels. Moreover, the fraction of genes that showed correlation to growth rate, EBV, and ATP substantially exceeded the fraction associated with a cis-eQTL of the same strength (compare figure 4E to 4C).
To parse the association of SNPs and other measures with variation in gene expression, we decomposed the total variance in expression of each gene into inter-individual and intra-individual (experimental) variation. As expected, eQTLs contribute only to inter-individual variation (Figure 5A), while EBV and ATP are correlated to either inter-individual or intra-individual variation, depending on the gene (Figure 5B and 5C).
Taken together, these observations have a number of implications: First, RNA levels for more genes are correlated to the measured non-genetic cellular factors than are associated with individual cis-eQTLs. Second, these non-genetic factors may influence gene expression not only by varying across cell lines in a reproducible manner (like SNPs), but also by varying across experiments for the same cell line. Third, for some genes, a given non-genetic factor is correlated to inter-individual variation (genes arrayed along the x-axis in Figure 5), and yet for other genes that same factor is correlated only to intra-individual variation (genes arrayed along the y-axis). Factors correlated to inter-individual variation could, in principle, represent processes related to the action of a genetic variant, whereas those that only vary across experiments represent noise with respect to genotype-phenotype association.
We observed a large number of genes whose level of RNA expression at baseline was correlated to drug response. Levels of RNA transcripts for 20% of genes in the Broad Institute dataset and 18% in the WTSI dataset were correlated (at a rho2>0.05) to EC50 for at least one of the drugs assayed (after growth-rate and ATP adjustment). EC50s for SAHA and 5FU appeared to have the strongest relationship to RNA levels, correlating to 8.7% and to 7.7% of genes measured at the Broad and WTSI, respectively.
Applying the variance components analysis to see how inter- and intra- individual variation in growth-rate and ATP adjusted EC50s are potentially influenced by RNA levels (and “assigning” to a given gene its strongest correlated drug), we observed that RNA levels are predominantly correlated to inter-individual differences in EC50s (Figure 5D). Much less of the correlation between RNA expression and EC50s reflects intra-individual variation. This observation supports the hypothesis that interindividual variation in RNA levels due to eQTLs may contribute to variation in drug response.
Having evaluated SNP associations with RNA levels (eQTLs), and the correlation of RNA levels to drug response, we asked whether the two relationships might point to eQTL SNPs associated with drug response. First, we asked whether there was an enrichment of genes both correlated to drug response and associated with an eQTL. Second, for the subset of genes with both an eQTL and correlation of RNA levels to drug response, we asked whether the eQTL SNPs were associated with drug response. Finally, we evaluated whether the strength of SNP association with RNA levels (eQTL) is correlated to the strength of SNP association with drug response. None of these analyses strongly supported an influence of eQTL SNPs on drug response.
We first examined the fraction of genes whose expression is associated with an eQTL and correlated to drug response. As seen in Figure 4, ∼14% and 4.5% of genes have cis-eQTLs (r2>0.08, FDR<10%) in the WTSI and Broad Institute datasets respectively. In the same data, levels of RNA of 18% (WTSI) and 20% (Broad) of genes are correlated to drug response (rho2>0.05, FDR<10%). When we consider the intersection of eQTL-bearing genes and drug-response correlated genes in each dataset independently, however, we see that only 1.4% (WTSI) and 0.9% (Broad) of genes are both correlated to drug response and bear a cis-eQTL. Neither intersection contains more genes than would be expected by chance alone and, at most, only a small fraction of genes are involved.
Among the 1000 “best-measured” genes in each RNA dataset, we identified a total of 23 genes that happened to contain both an eQTL and showed correlation of RNA levels to drug response. We asked whether these 23 eQTL SNPs showed a non-random distribution of association with drug response. When we first regressed the drug EC50 against genotype for each of the 23 SNPs above, we saw an excess of association over that expected under the null distribution (Figure 6). Moreover, the associations of SNPs with drug response appear to be in the direction predicted by the pair-wise SNP-RNA and RNA-Drug response relationships (Figure S3). A simulated dataset with the same SNP/RNA/Drug variances and independent SNP-RNA/RNA-Drug pairwise covariances (i.e. the eQTLB/RNAB scenario in Figure 6A) as those observed fails to demonstrate the excess association between SNPs and drug-response (Figure 6B – gray lines). Though no highly significant examples were documented, these observations are consistent with the existence of eQTLs associated with drug response (i.e. the eQTLA/RNAA scenario in Figure 6A).
While Figure 6B might suggest that many eQTLs are associated with drug response, we recognized a potential bias that might inflate this association in absence of (or in addition to) real signal: the “winner's curse” [36] overestimate of effect size incurred during discovery of eQTLs (Figure S4). To examine this possible source of spurious association, we replaced all simulated eQTL effects in (Figure 6B) with an eQTL whose true effect is r2 = 0.05, but whose observed effect in simulated datasets is r2>0.08. In this (more realistic) simulation, we recreate an inflation of p-values similar to that observed. (Figure 6C). This analysis suggests that winner's curse may contribute to the apparent excess of association in Figure 6B.
Finally, if these eQTLs were truly influencing drug response, one might expect that stronger eQTLs would have stronger associations with drug response. We plotted the strength of each eQTL against the strength of association between the eQTL SNP and drug response. Counter to expectation, the strongest associations between SNPs and drug response are observed for SNPs that are weak eQTLs, while most of the stronger eQTLs have no association with drug response (Figure 6D). We do observe three SNPs with relatively strong drug response and RNA levels association (Figure 6D blue arrow): rs1384804-C8orf70 (Ensembl:ENSG00000104427)-MTX, rs3733041-GLT8D1 (Ensembl:ENSG00000016864)-5FU, and rs2279195-SH3TC1 (Ensembl:ENSG00000125089)-Simvastatin with SNP-Drug p-values of 0.03, 0.05, and 0.02 respectively. While these may be interesting candidates for follow-up and replication, statistical significance is extremely weak, and thus much larger sample sizes are required to achieve genome-wide significance.
Recent studies have shown that a subset of genes contain cis-eQTLs that explain a modest fraction of inter-individual variation in RNA levels. Other studies used LCLs to perform linkage and association scans for drug response [26],[27],[28]. However, few reports characterize the biological reproducibility of these phenotypes, and none to our knowledge have characterized their correlation to in vitro measures such as growth rates, EBV copy number, and metabolic activity. We document that most traits we studied, whether drug responses or RNA transcript levels, are only partially reproducible across experiments, and that more genes are correlated to cellular growth rate, ATP levels, and EBV copy numbers than to genetic variants (at comparable fractions of variance explained). Thus, in addition to issues of statistical power relative to genetic size of effect (Figure S2), day to day variability in a trait and confounding factors are major influences on gene mapping experiments in LCLs.
Consistent with prior reports, our genome-wide association studies of drug response did not reveal any SNPs associated with drug response with genome-wide significance. The inability to detect such SNPs is likely due to lack of power to detect weak effects with limited sample size (Figure S2) and in the presence of significant confounding and noise.
Several studies attempted to improve power to discover SNPs associated with drug response [15],[16] by integrating eQTLs and RNA correlations to drug response [18],[19]. Whether these eQTLs are incidental or actually contributing to drug response depends on whether the cognate RNAs influence drug response or are merely correlated to drug response by a non-genetic factor that simultaneously affects both phenotypes. Our results fail to show convincing association of eQTL SNPs with drug response (EC50s adjusted for growth rate and ATP levels). Moreover, some apparent association can be attributed to “winner's curse” (a bias possibly avoidable in the future with the creation of large cohorts for eQTL discovery). We do observe three potential associations that may merit future study: rs1384804 near C8orf70 to MTX, rs3733041 near GLT8D1 to 5FU, and rs2279195 near SH3TC1 to Simvastatin.
The hallmark of genetic mapping is causal inference: the interpretation that genetic variants at a particular genomic locus are influencing a trait of interest. This interpretation requires confidence that the association between genetic variation and phenotype is not due to confounding, but rather represents a causal relationship. In an experimental cross, causal inference is supported by meiotic randomization and the shared parents of all offspring. In a genome-wide association study, causal inference can be supported if the genomic background of study participants is observed to be null distributed and potential confounders are eliminated. Our data suggest that GWAS of LCLs need to carefully consider the major impact of non-genetic confounding in relation to the documented effects of eQTLs. In addition to reducing power, confounding by non-genetic factors can cause spurious associations between cell lines and phenotypes, violating the conditions under which causal inferences can be made.
A major limitation of our study is the relatively small sample size of the HapMap samples for performing genome-wide association studies. As much larger collections of LCLs (such as those proposed to study cell lines from eight thousand and one-hundred thousand individuals by the Framingham Heart Study [37] and the National Children's Study [38], respectively) are currently being collected, we are optimistic that larger studies have potential to map pharmacogenetic loci in LCLs. By highlighting these aspects of the LCL model, as well as pointing to how some of them may be addressed, we hope to build a stronger foundation on which these important experiments can be planned and carried out.
EBV-transformed lymphoblastoid cell lines were acquired from the NHGRI Sample Repository for Human Genetic Research in frozen aliquots. Cells were thawed in 5 mL culture medium (RPMI medium 1640 (Invitrogen) supplemented with 10% FetalPlex (Gemini), 2 mM L-Glutamine (Invitrogen), and 1× penicillin/streptomycin (Invitrogen)). Cell lines were counted daily using Z2 Coulter Counter (Beckman Coulter) and passaged as needed to maintain a concentration of 2–5×1e5 cells/ml at 37 C in a 95% humidified 5% CO2 atmosphere.
Initially, cells were grown until 5×1e5 cells/ml were reached in 50 mL total volume. Then, ten identical aliquots were frozen in 1 mL freezing media containing 50% FetalPlex, 40% RPMI 1640 medium, and 10% DMSO (Sigma) at −80 C for 24 hrs and transferred to liquid nitrogen. These aliquots were used to provide biologic replicates for the experiments described below.
Aliquots were thawed on experiment day #1 as described above. Cell lines were counted daily and passaged as need to maintain a concentration of 4–8×1e5 cells/ml in 10 mL culture medium. On experiment day #7, cells were counted and distributed for use in the various experiments described below. One cc of culture was used for immediate immunophenotyping via FACS and Luminex beads. One cc of culture was used for RNA and DNA extraction using Trizol (Invitrogen) following the manufacturer's protocol. The remaining eight cc of culture were used for drug response assays described below.
The drugs that we studied are bortezomib (courtesy of T. Hideshima), rapamycin (Biomol), 5-fluorouracil (Sigma), methotrexate (Sigma), 6-mercaptopurine (MP Biomedicals), SAHA (Biovision), and simvastatin (Calbiochem). These drugs were arrayed in a source plate in the concentrations according to supplemental figure. The source plate was pinned into each cell line in duplicate, resulting in each drug concentration being assayed in each cell lines 4 times.
For drug response assays, LCLs for each cell line were diluted to 1×1e5 cells/ml, and 25 uL of cell culture were plated into each well of two white solid flat bottom 384 well plates (Corning cat# 3704) using a microplate dispenser (Multidrop Combi, Thermo Scientific). Next, 100 nL was pin-transferred from the source plates into the plates containing cells using an automated 384 channel simultaneous pippettor (CyBi-Well, CyBio). Plates were incubated at 37 C in a 95% humidified 5% CO2 incubator.
After 48 hrs, plates were removed from the incubator to room temperature for 10 minutes prior to being vortexed for 30 seconds. 25 uL of Celltiter Glo (Promega Cat No. G7573) diluted 1∶3 in PBS was added to each well with the Multidrop microplate dispenser and shaken for two minutes. Luciferase luminescence was then immediately measured for each well using a multiplate illuminometer (Envision, Perkin Elmer). Raw luminescence data is available online: http://chembank.broad.harvard.edu/assays/view-project.htm?id=1000477.
The experiment was monitored for cell-culture handling, plating, pinning, and assay errors and failed cell lines/plates/drug-rows were excluded from down-stream analysis. (Most cell lines were successfully assayed on two plates for all drugs, however; specific counts are below.) Luminescence values in drug-exposed wells were divided by the median control-well luminescence in the same plate row (after excluding plate edge wells) to obtain 4 viability fractions per cell line, per drug, per dose, in each experiment. For evaluation of technical reproducibility, the median of the 2 fractions on each plate was taken as the cell line's response to that dose on that plate. For evaluation of biological reproducibility and all other analyses, the median of the 4 fractions was taken as that cell line's response to that dose in the experiment. Drug responses were examined, and it was noted that the experiment failed to achieve meaningful cytotoxic response to rapamycin, with most cell lines reaching a maximum fractional viability of only ∼0.6–0.7, even at highest concentration of drug assayed. It was concluded that the viability assay was not a relevant read-out for rapamycin response, and the drug was not considered in further analyses.
Overall cell line response to a given drug was then calculated by taking the average response to a dose across all cell lines in the experimental batch (cell lines were assayed in batches of ∼90), subtracting the average from the value for each cell line, and then averaging the result for each cell line across all doses. (The 4–5 low-concentration doses where all cell lines had a fractional viability of ∼1 were excluded from the calculation.) In this way, the (single value) relative response of a given cell line to a drug was calculated, representing the non-parametric distance of that cell line's dose-response curve to the average dose-response curve for that drug in the experiment. (For the analysis of technical reproducibility, the calculation was done using only replicate plate A for all cell lines, and then using only replicate plate B, and the two values were compared). Quality control then proceeded by examining the dependence of response on the compound stock plate from which the drugs were pinned. (Compound stock plates were prepared with enough drug to run ∼20 cell lines and drug response should be independent of the drug stock.) Indeed, it was noted that for 5FU, 6MP, Simvastatin, SAHA, and MTX, dependence on drug stock was weak, while for bortezomib, the dependence was profound, with large differences in response between different plates, significantly in excess of the differences between cell lines on a given plate. Thus, bortezomib was excluded from further analysis. Though dependence on compound plate for the other 5 drugs was weak, average response for each compound stock plate was subtracted from each cell line using that plate (for each drug independently) and this normalized response was carried forward.
In summary, after the processing steps above in the main batch of experiments, 254 cell lines were successfully assayed for response to 6MP, 256 for MTX, 260 for Saha, 262 for Simva, and 259 for 5FU. 84 cell lines were then again successfully measured for all 5 drugs as biological replicates. (For ease of comparison, technical reproducibility is also reported using only the two plates from these biological replicate samples.) These values are available as “relative responses” in the online supplement. Analyses in Figure 2 use this data for the ∼200 successfully measured unrelated individuals, after again centering within each HapMap panel. Also, the median (non-boundary) control well luminescence over the two plates for each cell line was taken as the “ATP content” of the cell line. The value was divided by 100,000 and centered within each HapMap panel.
To account for the effect of growth-rate on response to MTX, 5FU, and 6MP, we reasoned as follows: Assume a simple ODE model of cell line population growth: , where P(t) is the # of cells in the population at a given time, and r is the (unobserved in the specific drug-exposure experiment) growth rate parameter. This ODE has the solution: P(t) = P0ert. When the cell line is exposed to drug, its growth-rate is impaired in a concentration-dependent manner. Taking inspiration from first-order Michaelis-Menten kinetics, we can model this as: , which is solved by . As our observed luminescences are ratios between drug wells and control wells at given concentrations, we can write , which can simplified as .
There are two identifiable parameters in this model: the concentration necessary for half-maximal reduction in growth-rate (EC50) which is independent of growth rate r itself, and r * maximal reduction of r, a product term dependent on growth rate whose components cannot be independently estimated. The model was fit for each cell line, for each drug independently, using median measurements at all doses. QC was performed by excluding all models with RSS>0.08. The –r*MaxRed term was discarded, and the EC50 was carried into further analysis after centering the values within each HapMap panel. (257 cell lines were successfully fit for 5FU, 251 for 6MP, and 255 for MTX.) Models were also successfully fit to all 84 biological replicates of 6MP and 5FU, and 82 replicates of MTX. ATP correction for 5FU and MTX was then carried out by taking the residuals of the linear regression DRUG∼ATP.
SAHA and Simvastatin were modeled by a standard sigmoid[39], with response (fractional viability) at a given . Notably, max inhibition and EC50 are not the same as above, here representing a minimal viability and the concentration at which that minimal viability is achieved, respectively. Maximum inhibition (aka minimum viability) were <0.05 for most cell lines for simvastatin and varied between ∼0.1–0.3 for SAHA. The EC50 was carried into further analysis after centering the values within each HapMap panel. Again, QC was performed by excluding all models with RSS>0.08. (257 cell lines were thus successfully fit for Saha and 261 for Simvastatin.) Models were also successfully fit to all 84 biological replicates of Saha and 5FU, and 83 replicates of Simvastatin. The GWAS for drug response was performed with all successfully measured individuals, while analyses presented in Figures 2,5,6 were performed with unrelated individuals only.
Each cell line was seeded at a concentration of 2×1e5 cell/mL in 2 mL. LCLs were counted daily for five consecutive days with an automated particle counter (Z2 Coulter Counter, Beckman Coulter). A regression of the form log(conc day i) = r*i+log(conc day 0) was fit for each cell line to obtain the estimate of growth rate r. QC was performed by evaluating the 95% confidence interval of the r estimate and rejecting estimates whose interval width exceeded 1.1. Thus, estimates of growth-rate for 237 cell lines were obtained. These values were normalized within each population for all analyses. An abbreviated second replicate of the experiment was repeated on a subset (155) of the cell lines with only the 3rd day counts collected to evaluate growth rate reproducibility.
From each LCL, ∼25,000 cells were incubated with R-Phycoerythrin–conjugated mouse anti-human antibody to cell surface markers (CD19, CD20, CD21, CD40, CD58, CD80, CD86, CD95, CD227, IgD, IgG, IgM, HLA-DQ, HLA-DR, and IL6R) at 4°C for 30 min. Cells were washed once with PBS and 1% fetal bovine serum and were fixed with 1% paraformaldehyde. Data on cell-surface expression in each cell line were acquired using a fluorescence-activated cell sorter (BD Biosciences FACSCalibur system). To quantify expression for each LCL, we used flow cytometry, requiring at least 500 cells per LCL for it to be included in our analysis. Fluorescence intensity was measured for the anti-cell surface protein antibody and a control isotype antibody for each LCL. A marker (and, separately, a control) histogram was created by placing individual cell measurements into 1,024 equally spaced intensity bins. Counts in the control histogram were subtracted from the marker histogram to obtained a “normalized” histogram of cell-counts in each of the 1,024 intensity bins. The average intensity was then calculated from this normalized histogram and the log of this value was carried forward into QC as the average normalized marker expression for that LCL.
QC then proceeded by regressing this marker expression on the total cell count obtained for that marker within a given experimental batch of LCLs. (samples were batched by HapMap panel) We reasoned that if the experiment was successful, there should be no dependence of cell-surface marker expression on the quantity of viable cells obtained in the experiment; if there was such a dependence, the marker expression was likely reading out handling differences between LCLs, not true, intrinsic differences in expression. Indeed, by this metric, we found that during the first batch of experiments that was attempted (for the CEU panel), only 4 markers were successfully measured, while subsequent batches (YRI+CHB/JPT samples) succeeded for 14 and 9 markers respectively. In most markers that passed this filter, it was further noted that a few cell lines showed very low expression, far from the overall distribution of the values for each batch. While it is conceivable that these represent true differences, we interpreted these values as individual LCL measurement failures, and further truncated the lowest 5% of values within each marker in each batch. Thus, the final dataset contains measurements of: 85 cell lines for CD19 and CD20, 169 for CD21, 166 for CD227, 248 for CD40, 164 for CD58, 166 for CD80 and CD86, 248 for CD95, 80 for HLADQ, 85 for HLADR and IgM, and 165 for IgD, IgG, and IL6R. These values were centered within each panel and carried into further analysis.
30 HapMap cell lines were screened with a multiplex antibody bead kit from Biosource (Cytokine 25-Plex for Luminex (Catalog #LHC0009)). Of the 25 cytokines originally selected for this assay, 8 were reliably detectable (lower concentration: IL8, IL10, IL12p40, TNFa, IP10; moderate concentration: MIP1a, MIP1b, RANTES). Of these, it was found that measurements for MIP1a and MIP1b were strongly correlated; thus we decided to include only MIP1b in further experiments. These 7 cytokines were assayed in the remainder of the cell lines according to the following protocol:
One cc for each LCL was placed into a single well of 96-deep well plate. The samples were centrifuged at 500 rpm for 5 minutes at room temperature. The supernatant was placed into a new 96-well plate, and placed dry ice to be stored at −80 degrees All assays were performed on a single thaw.
The cytokines were measured following the manufacturer's protocol. In order to ensure that the measured cytokine concentration fell in the linear part of the standard curve, the lower concentration cytokines were multiplexed together (final dilution 1∶2); and MIP1b and RANTES were multiplexed together (final dilution 1∶6).
The concentration of each cytokine was calculated based on the standard curve generated by the same plate, after subtracting out the “blank” background. A 3-parameter model was used to convert median fluorescent intensity (MFI) to protein concentration (ng/ml). A subsequent correction was applied to account for the dilution factor at the time of the assay. All final concentrations are expressed as pg/ml and log-transformed. 262 cell lines were successfully measured for IL10, IL12, IL8, IP10, and TNFa, and 266 measurements were obtained for MIP1b and RANTES. (79 and 87 biological replicate measurements were also obtained for the above two sets of cytokines respectively.)
All LCLs were cultured in the fashion described above. Prior to the plating of cells for the Drug Response Assay, 5×105 cells were set aside for RNA extraction. Cells were immediately lysed with Trizol Reagent (Invitrogen). RNA was collected according to the manufacturer instructions. 1.25 ug total RNA (OD>1.8) was diluted to a total volume of 10 uL. RNA was processed and hybridized onto Affymetrix Human U133A whole genome RNA expression genechip arrays according to the manufacturer's protocol. Gene expression summary values for the whole dataset were computed by RMA[40],[41] and log-transformed. Measurements were successfully obtained for 257 HapMap cell lines in the main experiment, for 64 biological replicates, for 24 cell lines originally thawed at the WTSI, as well as multiple replicates of 5 cell lines derived from chimpanzees. (Expression data is available on GEO Accession # GSE11582).
For analysis, the dataset was further processed as follows: 1) The ∼22 K total probe sets on the Affymetrics U133A were restricted to the 9084 judged expressed (p-value<0.06) by the Affymetrix software in at least 2/3 of 50 randomly selected scans. 2) These 9084 expressed probes were matched by Genbank transcript accession number (NM_#) to the 13,300 targets judged expressed by the same criterion in the WTSI Illumina HapMap experiments (using the probability of detection p-value output by the Illumina software.) This yielded a reduced set of 3600 Affymetrix probes (3592 Illumina targets) whose transcripts were reliably detectable in both experiments. 3) To obtain a comparable dataset from the WTSI Illumina data, we took the median over their 4 technical replicates for each target and quantile normalized across all samples. 4) We averaged within each gene symbol, in each dataset, for each sample, to get the set of 3538 genes expressed in both experiments and measured on both platforms. 5) To prevent family structure from introducing bias, the dataset was restricted to unrelated individuals only for the analyses in Figures 3–6: 198 each in the main Broad and WTSI experiments, 49 biological replicates at the Broad, and 16 samples for whom RNA was extracted at the WTSI and measured in both locations. Both centered (for each gene within each panel) and uncentered data is available in http://www.broad.mit.edu/mpg/pubs/hapmap_cell_lines/ and were each used as appropriate.
All previously collected DNA was diluted to PCR concentration of 2.5 ng/uL and arrayed in 384 well storage plates (AbGene Cat No. AB-0564). Custom TaqMan assays were designed using Primer 3 (http://frodo.wi.mit.edu/) and ordered from Applied Biosystems. The EBV copy number assay interrogated a 66_bp fragment at the DNA polymerase locus (EBV forward primer 5′GACGA TCTTGGCAATCTCT3′, EBV reverse primer 5′TGGTCATGGATCTGCTAAACC3′, EBV probe 5′6FAM-CCACCTCCACGTGGATCACGA-MGBNFQ3′). The mtDNA copy number assay examined a 72 bp fragment at the ND2 locus (mtDNA forward primer TGTTGGTTATACCCTTCCCGTACTA, mtDNA reverse primer CCTGCAAAGATGGTAGAGTAGATGA, mtDNA probe sequence 5′6FAM-CCCTGGCCCAACCC-MGBNFQ3′).
As an internal reference, a 90 bp assay from the NRF1 locus on chromosome 7 was multiplexed with EBV or mtDNA (NRF1 forward primer 5′CTCGGTGTAAGTAGCCACAT 3′, NRF1 reverse primer 5′GAGTGACCCAAACCGAACAT 3′, NRF1 probe 5′VIC-CACTGCATGTGCTTCTATGGTAGCCA-MGBNFQ 3′). Equal efficiency of amplification was observed for each assay in the multiplex reaction. Final Concentrations for EBV primers, mtDNA primers, EBV probe, mtDNA probe, NRF1 primers and NRF1 probe were .25 uM, .25 uM, 10 uM, 10 uM, 1 uM and 10 uM respectively. 5 ng of DNA template was used for each TaqMan reaction performed according to the manufacturer's protocol. Relative EBV and mtDNA copy number was determined by the difference of CT method[42]. Log-transformed. EBV measurements were obtained when cell lines were first received from Coriell (257), during the main batch of experiments (257), and for the biological replicate set (86). Mitochondrial DNA measurements were obtained only for 252 cell lines in the main experiments.
We are interested in the fraction of gene-trait (or gene-eQTL) relationships that are real (i.e. would reach statistical significance given enough samples) and above a given r2 thresh-hold in the current sample. So, we want P(real, r2 > = c) in joint distribution notation, i.e. a relationship can be real (non-null) or spurious (null) and can exceed a certain threshold or not. By regressing a trait on multiple genes, we observe: P(r2 > = c). It is the fraction of relationships exceeding any given threshold, the green curve. By permutation, we also have: P(r2 > = c|not_real), the blue (average of black) curve. So, we write, by conditioning on whether a relationship is real or not:Or, rewriting, we have:Everything on the right hand side is known, except P(real), the true proportion of gene-trait relationships in the data. This can theoretically be estimated ala Storey et al. 2003 [43] but the estimate can be unreliable in the setting of dependencies, as is the case in our data since genes are largely in clusters. So, we take the worst case scenario, setting P(real) = 0. Thus, we have:So, P(r2 > = c)−P(r2 > = c|not_real) is then a lower bound for P(real, r2> = c), the black curve. It is important to note that the interpretation of this lower bound is limited to the sample size used in the analysis. Given more samples, the estimate will change to even more genes being affected by traits or eQTLs, albeit at lower r2s.
To estimate the amount of inter- and intra- individual variation present for each gene in the ∼50 unrelated individuals thawed and measured twice at the Broad Institute, we fit a random effects model of the form yij = μ+αi+εij, where i indexes the individuals, and j is 1 or 2 for the biological replicate being considered. The estimated variance component is then the inter-individual variation in gene expression for the gene, while the residual variance is the intra-individual variation. To evaluate the effect of a cis-eQTL or cellular phenotype on an RNA, a fixed effect x corresponding to trait was then added to the model to get: yij = μ+βxij+αi+εij. The resultant change in variance components can then be interpreted as the “effect” of that trait or snp on RNA expression. The directionality of the effect is clearly only known for SNPs, but the nature of relationship (inter-, intra-, or both) can be examined for any trait. It's worth pausing to reflect on what these “effects” mean: If including a QTL SNP genotype in the model reduces inter-individual variance (as the overwhelming majority of SNPs do, Figure 5A), it implies that fixed differences in genotypes (QTLs) between individuals correlate to fixed differences in expression between individuals in the corresponding gene. (as one would expect) If, on the other hand, the intra-individual variance component is reduced when accounting for a given trait, the implication is that day-to-day variations in the trait correspond to day to day variations in the RNA. As would be expected, some genes also show a combination of the two effects. Finally, these estimates are quite noisy, suffering from random fluctuations in RNA levels, measurement error, and the relatively small sample size available for the analysis; estimation is likely even less reliable for weaker effects. Nevertheless, the analysis is instructive for the stronger signals and overall patterns and would improve given more samples and technical replicates.
1,045,141 autosomal SNPs with MAF>10% in each of the 3 (CEU, YRI, CHB/JPT) HapMap panels were selected from the Phase 2 HapMap build 21 for association testing to drug response phenotypes. The between/within family model of association was tested for each SNP against each drug, in each panel independently, using PLINK[32] v1.02 with options “–qfam-total –geno 1 –aperm 100 100 000 000 0.00000005 0.0001 5 0.001”. For each drug, p-values for each SNP were then combined across panels using Fisher's method. 25,735 X-chromosome SNPs were tested analogously, but using an additive model on unrelated individuals only with PLINK command line “–assoc –geno 1”; none exceeded 5e-8. QQ plots for the autosomal SNPs for each drug are available at: http://www.broad.mit.edu/mpg/pubs/hapmap_cell_lines/snps_vs_drug_response_pvalues/.
R – Aside from GWAS scans performed using PLINK, all other analyses were performed using R version 2.5.0[44].
|
10.1371/journal.ppat.1001199 | Interferon-Inducible CXC Chemokines Directly Contribute to Host Defense against Inhalational Anthrax in a Murine Model of Infection | Chemokines have been found to exert direct, defensin-like antimicrobial activity in vitro, suggesting that, in addition to orchestrating cellular accumulation and activation, chemokines may contribute directly to the innate host response against infection. No observations have been made, however, demonstrating direct chemokine-mediated promotion of host defense in vivo. Here, we show that the murine interferon-inducible CXC chemokines CXCL9, CXCL10, and CXCL11 each exert direct antimicrobial effects in vitro against Bacillus anthracis Sterne strain spores and bacilli including disruptions in spore germination and marked reductions in spore and bacilli viability as assessed using CFU determination and a fluorometric assay of metabolic activity. Similar chemokine-mediated antimicrobial activity was also observed against fully virulent Ames strain spores and encapsulated bacilli. Moreover, antibody-mediated neutralization of these CXC chemokines in vivo was found to significantly increase host susceptibility to pulmonary B. anthracis infection in a murine model of inhalational anthrax with disease progression characterized by systemic bacterial dissemination, toxemia, and host death. Neutralization of the shared chemokine receptor CXCR3, responsible for mediating cellular recruitment in response to CXCL9, CXCL10, and CXCL11, was not found to increase host susceptibility to inhalational anthrax. Taken together, our data demonstrate a novel, receptor-independent antimicrobial role for the interferon-inducible CXC chemokines in pulmonary innate immunity in vivo. These data also support an immunomodulatory approach for effectively treating and/or preventing pulmonary B. anthracis infection, as well as infections caused by pathogenic and potentially, multi-drug resistant bacteria including other spore-forming organisms.
| Innate immunity is critical to host defense and plays a central role in protecting the lungs from respiratory pathogens. Among the mediators important in the innate host response to pulmonary infection are chemokines, proteins originally described for their ability to regulate immune cell trafficking during an inflammatory response. More recently, chemokines have been found to exert direct antimicrobial activity against a broad range of bacteria and fungi in vitro. While these observations suggest chemokines may contribute to host defense by killing microorganisms at local sites of infection through activities not associated with cellular chemokine receptors, the biological relevance of direct chemokine-mediated antimicrobial activity in vivo has not been established. Here we show that the murine chemokines CXCL9, CXCL10, and CXCL11 exert direct antimicrobial effects against B. anthracis in vitro and that neutralization of these CXC chemokines, but not their shared receptor CXCR3, increases host susceptibility to pulmonary B. anthracis infection in vivo. These data provide unique insight into the host mediators important in host-pathogen interaction and pathogenesis of disease and support the emerging concept that host chemokines mediate efficient, pleiotropic roles that include receptor-independent promotion of host defense in vivo.
| The pulmonary airways represent a major site of interaction between the mammalian host and microbial pathogens. Infection resulting from the exposure of the respiratory tract to a variety of microorganisms is opposed by pulmonary innate immunity, a complex host response that protects against infection by directly mediating initial host defense in the airspace while helping to shape the activation of adaptive immunity [1], [2]. Among the primary components of innate immunity are secreted mediators including chemokines, small proteins produced mainly by epithelial and phagocytic cells in response to pattern-recognition receptor engagement and pro-inflammatory cytokines [3]. Chemokines were originally recognized for their ability to induce directed migration of leukocytes and facilitate controlled cellular accumulation and activation during an inflammatory response through receptor-dependent interactions between chemokines and their specific G-protein-coupled receptor(s) expressed by responsive cells [4].
In addition to their role in cellular recruitment, a number of chemokines have been found to mediate direct antimicrobial effects against a broad range of Gram-positive and Gram-negative bacteria in vitro [5]–[8]. While the mechanistic details of these effects remain undefined, antimicrobial activity is thought to result from interactions between positively-charged regions present at the chemokine' C-terminus and negatively-charged moieties at the microbial cell surface, resulting in cell lysis [8]. Although chemokines have been shown to be central components of the host response to pulmonary infection [9], these molecules have primarily been viewed in the context of receptor/ligand interactions, without consideration for direct ligand-mediated antimicrobial activity. As such, the biological relevance of receptor-independent, chemokine-mediated antimicrobial activity in host defense in vivo remains to be established.
The disease anthrax is caused by the Gram-positive, spore forming bacterium Bacillus anthracis. The infectious B. anthracis spore consists of distinct, concentric layers that encase the spore's genomic material and provide protection against multiple stresses including high temperature and lytic digestion [10], [11]. Depending on the spore's route of entry, B. anthracis causes three distinct types of disease: inhalational, gastrointestinal, and cutaneous anthrax. Inhalational anthrax results as a consequence of spore deposition within the host airspace. Here, spores encounter effectors of host innate immunity and are taken up by phagocytes including macrophages [12] and dendritic cells [13]. It is thought that spore germination, the resumption of metabolic activity and outgrowth as a vegetative cell, begins following phagocytosis at these localized sites of infection [14], [15] and that the vast majority of germinating organisms are killed [16]. During transit by phagocytic cells to the regional lymph nodes, however, a small subset of surviving bacilli are believed to mediate membrane disruptive events allowing escape from phagocytic vesicles and, subsequently, the phagocytic cell [17]. Extracellular bacilli evade host immune responses through the production of two principle, plasmid-encoded virulence factors: a tripartite toxin encoded by pXO1 and responsible for broadly suppressing the host immune response [18], and a poly-D-glutamic acid capsule encoded by pXO2, that protects against phagocytic killing [19]. These and other bacterial factors allow B. anthracis to multiply rapidly, resulting in systemic dissemination, toxemia, and death of the infected host [20].
CXCL9, CXCL10, and CXCL11 are homologous, interferon-inducible members of the CXC chemokine family that lack the tripeptide structure/function motif Glu-Leu-Arg (ELR) important in neutrophil chemoattraction [9]. As such, these interferon-inducible ELR- CXC chemokines signal through a common receptor, CXCR3, to facilitate selective recruitment of mononuclear leukocytes, natural killer cells, and plasmacytoid dendritic cells to sites of inflammation [9], [21]. We [22] and others [5], [8] have previously reported the ability of human CXCL9, CXCL10, and CXCL11 to exert direct antimicrobial activity against B. anthracis, as well as Escherichia coli, Listeria monocytogenes, and Staphylococcus aureus. Furthermore, we have observed that CXC chemokine induction in the lungs of C57BL/6 mice challenged intranasally with B. anthracis Sterne strain spores is associated with significant reductions in spore germination and subsequent disease progression [22]. Based on these observations, we hypothesized that murine CXCL9, CXCL10, and CXCL11 exert direct antimicrobial effects against B. anthracis and thereby mediate a receptor-independent contribution to host defense against pulmonary B. anthracis infection.
In the present study, we demonstrate that the murine interferon-inducible ELR- CXC chemokines CXCL9, CXCL10, and/or CXCL11 exert direct antimicrobial effects against toxigenic, unencapsulated B. anthracis Sterne strain (pXO1+ pXO2−) as well as toxigenic, capsule-forming Ames strain (pXO1+ pXO2+) spores and bacilli. Furthermore, we show that neutralization of these CXC ligands, but not their shared cellular receptor CXCR3, in C57BL/6 mice challenged with B. anthracis Sterne strain spores significantly increases host susceptibility to inhalational anthrax. These observations support that the CXC chemokines directly contribute to host defense against pulmonary B. anthracis infection in vivo, providing unique insight into the effector mechanisms of the innate host response to bacterial infection. These data also support the consideration of antimicrobial chemokines in the development of novel, therapeutic strategies for countering multidrug resistant pathogens.
As the induction of the interferon-inducible ELR- CXC chemokines within the lungs of spore-challenged mice is associated with resistance to inhalational anthrax [22], we sought to determine whether murine CXCL9, CXCL10, and CXCL11 exert antimicrobial activity against B. anthracis Sterne strain. Included in these studies were two control murine CC family chemokines, CCL2 and CCL5, whose molecular weights and basic isoelectric points are similar to the CXC chemokines examined [5]. Disruptions in spore germination and bacterial cell viability were assessed using colony-forming unit (CFU) determination, performed in the presence or absence of heat treatment to differentiate between heat-resistant spores and heat-sensitive bacilli.
By 6 h post-treatment, untreated, CCL2- and CCL5-treated spores underwent considerable germination and vegetative outgrowth as evidenced by a loss of heat-resistant CFU (germination) and an increase in heat-sensitive CFU (vegetative growth) as compared to the initial inoculum (Figure 1A). Treatment of B. anthracis spores with CXCL9 resulted in an approximate 1,000-fold reduction in viable organisms as compared to the untreated control. Retention of spore dormancy was observed to be significantly greater in the presence of CXCL9, yet viable spores represented less than 10% of the initial inoculum; interestingly, this reduction in viable spores did not coincide with an appearance of heat-sensitive, germinated organisms. These effects were concentration dependent (EC50 = 5.00±1.10 µg/ml; Figure S1A,B) and suggest that CXCL9 inhibits spore germination and disrupts the maintenance of spore viability. In support of this notion, CXCL9-treated spores demonstrated a lack of primary outgrowth as determined by microscopic visualization (Figure 1B), and the resumption of metabolism (a hallmark of spore germination) was absent in CXCL9-treated spore samples 6 h post-treatment as measured by an Alamar Blue based assay of metabolism (Figure 1C). Treatment of B. anthracis spores with CXCL10 or CXCL11 resulted in significantly decreased levels of heat-resistant organisms and an approximate 10-fold reduction in vegetative outgrowth as compared to the untreated control (Figure 1A). While possibly exerting a sporicidal effect, neither CXCL10 or CXCL11 was found to block spore germination (Figure 1A,C). Chemokine-mediated antimicrobial activity against B. anthracis Sterne strain bacilli was also observed, with all three interferon-inducible ELR- CXC chemokines capable of mediating significant decreases in vegetative cell viability as determined by both CFU analysis (Figure 1D) and Alamar Blue reduction (Figure 1E). CXCL9 demonstrated considerable bactericidal activity, mediating the complete killing of the initial bacilli inoculum in a concentration-dependent manner (EC50 = 3.96±0.75 µg/ml; Figure S1C,D). Of note, the antimicrobial hierarchy of the murine CXC chemokines presented here (CXCL9 >> CXCL10 ≈ CXCL11) is distinct from the hierarchy previously observed for the human CXC chemokines (CXCL10≥ CXCL9 >> CXCL11) [22].
In contrast to B. anthracis Sterne strain organisms, vegetative Ames strain bacilli carry the capsule biosynthetic operon encoded by pXO2 and are capable of generating a protective poly-D-glutamic acid capsule. While the increased virulence of encapsulated organisms has primarily been attributed to enhanced bacterial evasion of cell-mediated host responses [19], the capsule may act as a barrier against soluble immune mediators. Therefore, the antimicrobial potential of murine CXCL9 against fully virulent B. anthracis Ames strain organisms was examined using CFU determination. Also, as the ability of human ELR- interferon-inducible CXC chemokines to mediate antimicrobial effects against B. anthracis Ames strain is unknown, we sought to determine the capacity of human CXCL10 to directly target fully virulent spores and encapsulated bacilli relevant to human disease; human CXCL10 has previously been shown to exert antimicrobial effects against B. anthracis Sterne strain organisms similar to those reported here for murine CXCL9 [22].
Treatment of Ames strain spores with murine CXCL9 or human CXCL10 was found to result in significantly reduced levels of spore germination and primary outgrowth as compared to the untreated control (Figure 2A), supporting a lack of a role for pXO2-encoded components in spore susceptibility to these chemokines. We next examined the ability of murine CXCL9 and human CXCL10 to exert a direct bactericidal effect against toxigenic, encapsulated bacilli. Treatment of encapsulated Ames strain bacilli with murine CXCL9 resulted in an approximate 100-fold reduction in bacterial viability as compared to the untreated control 6 h post-treatment (Figure 2B). Similarly, human CXCL10 was found to display antimicrobial activity against encapsulated bacteria, mediating a five-log reduction in viable vegetative cells. Ames strain bacilli were visualized in India ink preparations, confirming that the initial bacilli inoculum consisted of encapsulated bacterial cells, and that the capsule was not lost under experimental conditions (Figure 2C). That murine CXCL9 and human CXCL10 exert direct antimicrobial effects against Ames strain spores and encapsulated bacilli, and that these effects are similar to those observed for Sterne strain organisms indicate B. anthracis Sterne strain is an appropriate model organism for studying chemokine-mediated antimicrobial activity against this pathogen.
In order to determine the biological relevance of direct chemokine-mediated antimicrobial activity during pulmonary B. anthracis infection, we used a murine model of inhalational anthrax in which endogenous CXCL9, CXCL10, and/or CXCL11, or their shared cellular receptor CXCR3 were selectively neutralized. Antibody-mediated neutralization was performed in C57BL/6 mice (relatively resistant to inhalational infection by B. anthracis Sterne strain) and was achieved through intraperitoneal (i.p.) administration of anti-sera raised against individual interferon-inducible ELR- CXC chemokines or the NH2 terminus of CXCR3 [23], [24]. These antibodies were previously shown to be specific without cross-reactivity to a panel of cytokines and other chemokine ligands [23].
Neutralization of endogenous CXCL9 in B. anthracis spore-challenged animals was found to significantly increase host susceptibility to pulmonary infection (p = 0.012) resulting in approximately 30% mortality as compared to spore-challenged animals receiving control serum, <5% mortality (Figure 3); administration of CXCL9 neutralizing serum in the absence of infection was not found to cause death, with 12/12 mice surviving beyond 20 days. Neutralization of CXCL10 in spore-challenged animals resulted in decreased host survival (20% mortality) that approached statistical significance (p = 0.064) when compared to infected animals receiving control serum; similar mortality was observed in CXCL10-/- mice following spore challenge (data not shown). CXCL11 neutralization was not found to increase host susceptibility to inhalational anthrax. Combinatorial neutralization of CXCL9 together with CXCL10 or CXCL10/CXCL11 during pulmonary B. anthracis infection significantly increased host susceptibility to anthrax, with neutralization of all three CXC chemokines resulting in 50% mortality (p = 0.0003).
Importantly, antibody-mediated neutralization of CXCR3 (Figure S2) did not result in increased susceptibility to pulmonary B. anthracis infection (Figure 3), and survival among spore-challenged CXCR3-/- and wild-type animals was the same (data not shown). These data suggest that direct ligand-mediated effects not associated with CXCR3 contribute to limiting disease progression in this model of pulmonary infection. Indeed, the post-challenge induction of endogenous CXCL9, CXCL10, and CXCL11 previously associated with resistance to inhalational B. anthracis infection [22] was maintained in animals receiving CXCR3 neutralizing serum (Figure S3). Additionally, host inflammatory cell populations in the lungs of spore-challenged animals receiving CXCL9/CXCL10/CXCL11 or CXCR3 neutralizing sera were strikingly similar following challenge (Figure S4); the absence of significant differences in host cell populations indicate that CXCR3-dependent, cell-mediated effects are not responsible for the distinct differences in disease progression between these groups. Taken together, the above data demonstrate a novel antimicrobial role for the interferon-inducible ELR- CXC chemokines during pulmonary B. anthracis infection that is independent of CXCR3-mediated cellular recruitment to sites of infection.
To gain insight into disease progression associated with the neutralization of CXCL9, CXCL10, and CXCL11, and to confirm that host death resulted as a consequence of B. anthracis infection, we investigated two salient features of anthrax: bacterial dissemination and toxemia. The ability of B. anthracis to disseminate from initial sites of infection was examined by measuring B. anthracis CFU in the lungs, kidneys, spleen, and liver from moribund mice receiving CXCL9/CXCL10/CXCL11 neutralizing sera, as compared to those measured from paired, spore-challenged animals receiving control serum.
Consistent with previously published reports measuring bacterial dissemination during pulmonary B. anthracis infection [25], [26], B. anthracis CFU in the lungs of spore-challenged animals were approximately equivalent between treatment groups. However, in contrast to control animals, which showed little evidence of extrapulmonary dissemination, tissues harvested from animals receiving CXCL9/CXCL10/CXCL11 neutralizing sera demonstrated widespread bacterial dissemination with considerable CFU detected in the kidneys, spleen, and liver (Figure 4). This observed dissemination is consistent with bacterial dissemination previously reported for strains of mice highly susceptible to Sterne strain infection [20]. The ability of the CXC chemokines to participate in limiting disease progression prior to systemic invasion was also observed using in vivo imaging. C57BL/6 mice were challenged with a bioluminescent strain of B. anthracis (7702-lux) whose vegetative cells are constitutively luminescent and allow visualization of bacterial dissemination [27]. As above, only upon neutralization of CXCL9, CXCL10, and CXCL11 was systemic disease observed as evidenced by detection of luminescence in tissues distant to the host airways (Figure 5A,B). Also, extrapulmonary dissemination was observed to occur after the establishment of infection in the chest (Figure C,D) consistent with impaired host defense at local sites of infection.
Toxemia is characteristic of systemic anthrax and results from the secretion of a tripartite toxin consisting of the receptor binding component protective antigen (PA), and two catalytically active components, the metalloprotease lethal factor (LF) and the adenylate cyclase edema factor (EF) [18]. Several animal studies examining the production of PA and LF during infection have found PA to be detectable in the blood of infected animals only during the terminal stages of disease [28], an observation thought to reflect rapid binding of PA by host cells [29]. Conversely, LF has been shown to accumulate earlier in infection, consistent with delayed internalization (cellular entry of LF depends upon prior PA binding, activation, and heptamerization), providing a good measure of toxemia during disease progression [28]. As an index of toxemia, we used an established mass spectrometry-based method [30] to detect and measure the levels of biologically active LF in serum collected from spore-challenged mice. Whereas infected control animals showed low or undetectable levels of LF, serum collected from spore-challenged animals receiving CXCL9/CXCL10/CXCL11 neutralizing sera was found to contain concentrations of active LF ranging from 25–400 ng/ml (Figure 6), levels commensurate with concentrations measured from the sera of nonhuman primates that have succumbed to inhalational anthrax [30]. These data indicate that the interferon-inducible ELR- CXC chemokines help protect against pulmonary B. anthracis infection in a murine model of infection, and that disruption of innate, ligand-mediated roles in host defense increases susceptibility to invasive disease and toxemia.
Exposure of the host lungs to potentially pathogenic microorganisms represents a significant immunological challenge for host defense. Initial encounters between inhaled microbes and components of pulmonary innate immunity initiate a dynamic set of interactions that ultimately determines whether disease will occur [31]. The host response to infection is coordinated, in part, through the production of chemokines that allow controlled cellular accumulation and activation during an immune response [32]. Some chemokine ligands display direct antimicrobial activity in vitro [33] raising the possibility of a multifunctional role for chemokines in host defense that includes microbial killing at local sites of host-pathogen interaction. Here, we investigated the ability of the interferon-inducible ELR- CXC chemokines to directly contribute to host defense against pulmonary B. anthracis infection. We found that CXCL9, CXCL10, and CXCL11 each exert direct antimicrobial effects against B. anthracis in vitro and that neutralization of endogenous CXCL9, individually or together with CXCL10 or CXCL10/CXCL11, but not CXCR3, significantly increases host susceptibility to inhalational anthrax in a murine model of infection. Our data support a novel, CXCR3-independent role for the interferon-inducible ELR- CXC chemokines in the innate host response against pulmonary B. anthracis infection that is consistent with direct chemokine-mediated antimicrobial activity at local sites of infection.
Each murine CXC chemokine examined in vitro for antimicrobial activity was found to exert direct antimicrobial effects against B. anthracis Sterne strain spores and bacilli. In vitro analysis also demonstrated the ability of murine CXCL9 and human CXCL10 to exert direct antimicrobial effects against fully virulent B. anthracis Ames strain spores and encapsulated bacilli. Interestingly, while Ames strain spores were found to be fully susceptible to murine CXCL9, direct killing of encapsulated bacilli by murine CXCL9 was reduced compared to unencapsulated organisms. As CXCL9 contains a relatively extended C-terminal region [5], the relative reduction in antimicrobial activity may result from greater exclusion of CXCL9 by the poly-D-glutamic acid capsule thereby preventing the chemokine from reaching the presumed site(s) of action at the bacterial surface. While this difference may impact the defensive role of CXCL9 during infection with fully virulent B. anthracis, the in vitro data presented here demonstrate that chemokine-mediated antimicrobial activity is applicable to both strains of B. anthracis examined. Given that the activity of many antimicrobial chemokines and host peptides is disrupted by the presence of serum and/or physiological concentrations of ions including Na+, K+, and Mg2+ [33], [34], it is important to note that the in vitro antimicrobial activity of CXCL9, CXCL10, and CXCL11 against B. anthracis was tested in culture medium containing physiologically relevant concentrations of serum proteins and ions. In addition, these ion concentrations are similar to those found in airway surface fluid [35], supporting the potential of the interferon-inducible ELR- CXC chemokines to mediate antimicrobial activity in the host airways.
Previous work by our laboratory has demonstrated that the induction of the interferon-inducible ELR- CXC chemokines within the lungs following B. anthracis spore challenge is associated with significant reductions in spore germination and resistance to pulmonary infection [22]. In the present study, we investigated the consequences of selectively neutralizing CXCL9, CXCL10, and/or CXCL11 during pulmonary B. anthracis infection, and whether potential ligand-mediated contributions to host defense were independent of interactions with CXCR3. Consistent with its potent antimicrobial activity in vitro and its sustained induction within the lungs following spore challenge in vivo (Figure S3; [22]), neutralization of endogenous CXCL9 resulted in significantly increased host susceptibility to inhalational anthrax. While individual neutralization of CXCL10 or CXCL11 was not found to result in significantly increased mortality among spore-challenged animals, combined neutralization of CXCL9 together with CXCL10 or CXCL10/CXCL11 indicated potential additive effects in promoting host defense against pulmonary B. anthracis infection, with neutralization of all three CXC ligands resulting in widespread bacterial dissemination, toxemia, and the highest mortality of any spore-challenged group examined in this study. Importantly, CXCR3 neutralization, which disrupts receptor-mediated cellular recruitment in response to these CXC chemokines, was not found to increase host susceptibility to inhalational anthrax. These results demonstrate the ability of the interferon-inducible ELR- CXC chemokines, in particular CXCL9, to contribute directly to host defense through activities not associated with CXCR3. Moreover, these observations support the potential of an efficient, multifunctional role for host chemokines that may represent a more generalized mechanism of the innate host response against infection.
While the data presented here are consistent with direct chemokine-mediated antimicrobial activity in vivo, chemokine ligand concentrations measured from lung homogenates of spore-challenged animals are not as high as those required to achieve antimicrobial effects in vitro. In fact, with few exceptions, most known antimicrobial chemokines and host peptides, including many defensins, exert direct bactericidal effects in vitro at relatively high concentrations; minimal inhibitory concentrations typically range from 0.1–100 µg/ml [36]. Numerous studies, however, have identified roles for antimicrobial host peptides in pulmonary defense against bacterial infection suggesting biologically relevant concentrations do occur during infection [37]. The ability of the interferon-inducible ELR- CXC chemokines to mediate direct antimicrobial activity in vivo is most likely relevant at local sites of host-pathogen interaction. At these inflammatory foci, the elaboration of chemokine production by host cells can be expected to result in substantial chemokine concentrations capable of mediating direct contributions to host defense [1]. This notion is supported by the ability of epithelial [38], [39] and mononuclear cells [5] to produce significant amounts of CXCL9, CXCL10, and/or CXCL11 in response to inflammatory stimuli, with concentrations of CXCL9 and CXCL10 reaching several hundred nanograms per milliliter [39]. Furthermore, tonsil fluid collected from patients with Streptococcus pyogenes pharyngitis contains CXCL9 concentrations exceeding those required to kill S. pyogenes in vitro, and the inhibition of CXCL9 expression reduces antimicrobial activity against this organism at the surface of inflamed pharyngeal cells [38]. CXCL9 may be of particular importance in promoting host defense against bacterial infection as it is strongly induced in several murine models of pulmonary infection including Klebsiella pneumoniae and Mycobacterium tuberculosis [9], [40]. Similarly, and of particular relevance to the current study, adults exposed to B. anthracis spores (based on positive nasopharyngeal swab cultures) in the U.S. Capitol building during the 2001 anthrax attacks demonstrated elevated levels of several inflammatory mediators including CXCL9 [41].
The ability of CXCL9 to mediate a multifunctional role in host defense is supported by observations that S. pyogenes and the opportunistic pathogen Finegoldia magna each release specific virulence factors believed to promote immune evasion by disrupting the integrity or availability of the C-terminal region of CXCL9, thereby reducing or abolishing direct antimicrobial activity [38], [42]. Interestingly, while these factors limit CXCL9-mediated antimicrobial activity, the ability of CXCL9 to signal through CXCR3 is largely retained, demonstrating separate and distinct chemokine-mediated functions independently disrupted by pathogens. That other antimicrobial chemokines are similarly targeted [43] further indicates that endogenously produced host chemokines mediate multifunctional roles in host defense that likely represent a more generalized mechanism of the innate host response to infection. Indeed, murine CCL6 and its human homologs were recently found to be highly expressed in the intestinal mucosa and capable of mediating antimicrobial effects against a subset of intestinal bacteria ex vivo [44]. In addition, the antimicrobial chemokine CCL28 has been found to be constitutively expressed and highly concentrated in mucosal secretions [6], and CXCL9 from seminal plasma possesses antimicrobial activity against the urogenital pathogen Neisseria gonorrhoeae [45]. These observations are each consistent with direct chemokine-mediated roles in host defense and support the notion of host chemokines as multifunctional effectors of innate immunity.
It remains to be determined at what point in pulmonary B. anthracis infection the interferon-inducible ELR- CXC chemokines mediate their contribution(s) to host defense. CXCL9 and CXCL10 are each induced to relatively high levels within the lungs following spore challenge suggesting that antimicrobial activity may act early in infection against the spore form of the organism. Antimicrobial activity against B. anthracis spores during infection is consistent with the previously reported association between CXCL9, CXCL10, and CXCL11 induction and decreased spore germination in vivo [22], as well as observations that the reduction of spore burden on resident macrophages is important in preventing intracellular vegetative outgrowth and subsequent disease progression [46]. Spore challenge with toxigenic, unencapsulated B. anthracis results in spore germination and the establishment of infection at local sites within the host airways [14], [15]. Infection is initially contained here providing an opportunity for chemokine-mediated antimicrobial activity against vegetative bacilli prior to extrapulmonary dissemination [20]. While the observations reported here are consistent with direct antimicrobial effects similar to those found in vitro, they do not preclude ligand-mediated immunomodulatory activity; CXCL9 has recently been reported to induce gene transcription and chemokine production in peripheral blood mononuclear cells, independent of interactions with CXCR3 [47]. As inhalational anthrax is an acute disease capable of abrogating host immune responses suggests that host chemokines mediate important roles in the innate host response of naïve hosts and help to limit infection early in disease progression.
The continuing emergence of antibiotic resistance [48] and the potential of engineered resistance in the weaponization of biological agents [49] represent serious areas of concern. The ability of host defense peptides to exert direct antimicrobial effects and promote protective immunity has been suggested as a template for the development of novel therapeutic strategies capable of addressing these challenges [50]. Certain chemokines (including CXCL9, CXCL10, and CXCL11) share many structural and functional relationships with host defense peptides, suggesting that these mediators have overlapping roles in host defense and similar therapeutic potential [51]. The ability of type 1 (IFN-α/β) and type 2 (IFN-γ) interferons to strongly induce ELR- CXC chemokine production supports the administration of exogenous interferon as a therapeutic strategy for treating pulmonary B. anthracis infection. Indeed, both IFN-α/β and IFN-γ have been found to promote protection against B. anthracis challenge in vitro [52] and in vivo [53]. While neither of these studies examined CXC chemokines, each supports the potential therapeutic application of exogenous chemokine induction in post-exposure prophylaxis or the treatment of anthrax. Furthermore, the exogenous induction of host chemokines capable of activating cellular immunity, promoting immune mediator production, and directly killing pathogens may apply more broadly to the development of innovative therapeutic avenues for the treatment of pathogenic and potentially, multidrug-resistant bacterial infections.
In summary, our findings provide strong evidence for an important CXCR3-independent role for the interferon-inducible ELR- CXC chemokines in the innate host response against pulmonary B. anthracis infection, and indicate that CXCL9, in particular, may function as one of the major antimicrobial components of the inflamed host airway. Neutralization of the CXC chemokine ligands, but not their shared cellular receptor, was found to disrupt the host's ability to limit disease progression and contain B. anthracis at initial sites of infection, resulting in increased susceptibility to inhalational anthrax characterized by systemic dissemination, toxemia, and death. While further studies are required to define the biologically relevant contributions of the interferon-inducible ELR- CXC chemokines to host defense, the ability of an intact host chemokine response to directly promote the innate host response against inhalational anthrax is consistent with direct antimicrobial activity as observed for these chemokines in vitro. Direct chemokine-mediated antimicrobial activity at the interface of host-pathogen interaction may represent an important mechanism in host defense, and supports the consideration of host chemokines in the development of novel, immunomodulatory therapeutic strategies.
Animal studies were carried out in strict accordance with the US Public Health Service Policy on the Humane Care and Use of Laboratory Animals (PHS Assurance #A3245-01), the US Department of Agriculture Animal Welfare Act (USDA Registration #52-R-0011), and the US Government Principles for the Utilization and Care of Vertebrate Animals Used in Testing, Research, and Training. Animal protocols were reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of the University of Virginia (Protocol #3677).
B. anthracis Sterne strain 7702 spores were prepared using a liquid culture method [54] with modification. Briefly, Difco Sporulation Medium [55] was inoculated with B. anthracis 7702, and cultures were incubated 4–5 d at 37°C with shaking. After sporulation, cultures were washed in cold, sterile dH2O and heat treated at 65°C to kill any remaining vegetative cells. Spores were purified over a Percoll gradient (GE Healthcare Biosciences, Piscataway, NJ, USA) washed, and enumerated. B. anthracis bacilli were prepared in brain heart infusion (BHI) broth (Becton, Dickinson and Company, Franklin Lakes, NJ, USA) and subcultured prior to use. Luminescent B. anthracis 7702-lux was kindly provided by Dr. T. Merkel (Food and Drug Administration, Bethesda, MD) and is described in detail elsewhere [27]. All work involving B. anthracis Sterne strain 7702 was performed using appropriate BSL-2 precautions. B. anthracis Ames strain was obtained through the NIH Biodefense and Emerging Infections Research Resources Repository, NIAID, NIH: Bacillus anthracis, Strain Ames (A0462), NR-411. The original stock was grown on capsulation (CAP) agar plates (0.3% yeast extract, 0.8% nutrient broth, 1.5% agar, 5% horse serum, and 0.8% sodium bicarbonate) overnight at 37°C, 5% CO2 in order to isolate phenotypically encapusulated organisms. Ames strain spores were prepared on agar slants as previously described [14], and bacilli were prepared fresh from CAP agar plates. All experiments with B. anthracis Ames strain were performed under BLS-3 precautions in a Select Agents approved laboratory following guidelines established by the Centers for Disease Control and Prevention, the US Department of Agriculture, and the University of Virginia Institutional Biosafety Committee.
For CFU determination and Alamar Blue analysis, spores (0.4−1×106 total) or bacilli (0.4−3×105 total) were added to Dulbecco's modified essential medium (Invitrogen, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (Hyclone, Logan, UT, USA) and containing 48 µg/ml of recombinant murine or human CXCL9, CXCL10, CXCL11, CCL2, or CCL5 (Peprotech, Rocky Hill, NJ, USA) stabilized with 0.3% human serum albumin (ZLB Bioplasma AG, Berne, Switzerland), or an equal volume of albumin alone (untreated control); recombinant murine chemokines were used unless otherwise specified. Endpoint analyses were performed 6 h post-treatment (after spore germination and/or vegetative outgrowth, but before bacterial overgrowth in untreated samples) as previously described [22].
Wild-type C57BL/6 mice, as well as CXCL10-/- and CXCR3-/- animals were obtained from The Jackson Laboratory (Bar Harbor, ME, USA). Antibody-mediated neutralization of CXCL9, CXCL10, CXCL11, and CXCR3 was achieved using published protocols [23], [24]. Briefly, C57BL/6 mice (female, 6–8 weeks old) were administered i.p. injections of goat serum raised against recombinant CXCL9, CXCL10, or CXCL11 (R&D Systems, Minneapolis, MN, USA) or a peptide constituting the NH2 terminus of murine CXCR3; control animals received an equal volume of donor herd normal goat serum (SeraCare Life Sciences, Milford, MA, USA). Neutralization was begun 24h prior to spore challenge, and performed daily throughout the study period (≤20 d). For single ligand or receptor neutralization, animals received approximately 6 mg of total goat IgG daily; for multiple ligand neutralization, animals received equal amounts of the indicated neutralizing sera, approximately 15 mg of total IgG. Antibody neutralizing capacity and selectivity have been described previously [23]. Intranasal B. anthracis spore challenges were performed following sedation with ketamine/xylazine (60/6 mg/kg body weight, i.p.). Twenty microliters of spore suspension (1−6×107 spores total) was placed drop-wise onto the nares of mice, and the animals were kept upright until breathing returned to normal. Animals were monitored for signs of illness according to an IACUC approved scoring system taking into account activity level, posture, and respiration; animals determined to be moribund were euthanized with an overdose of ketamine.
All tissues used in CFU determination were harvested following euthanasia and homogenized by hand on ice in sterile PBS. Sample dilutions were prepared in duplicate, and subsequently plated on BHI agar (Remel, Lenexa, KS, USA); sample plates were incubated overnight at room temperature before colony enumeration. All tissue samples were plated ± heat treatment at 65°C for 30 min to distinguish between spore and vegetative forms of B. anthracis.
Images of spore-challenged mice and luminescent signals were acquired using the In Vivo Imaging System (IVIS) Spectrum (Caliper Life Sciences, Hopkinton, MA, USA). For imaging, mice were anesthetized with 2.5% isofluorane mixed with oxygen and delivered by the XGI-8 gas anesthesia system supplied with the IVIS Spectrum. Images were acquired according to the manufacturer's recommendations, and the emission of photons from live animals was analyzed using Living Image 2.5 software.
Functional anthrax toxin LF was measured in animal sera prepared from whole blood collected via cardiac puncture. Quantification was based on matrix-assisted laser desorption/ionization (MALDI) time-of-flight (TOF) mass spectrometry (MS) as previously described [30]. Briefly, MALDI-TOF MS was used to detect specific peptide products generated following LF-mediated cleavage of a synthetic peptide substrate; LF concentrations were subsequently determined by isotope-dilution MS.
Significant differences among in vitro treatment groups were determined using one-way ANOVA with a Bonferroni multiple comparison post test; logarithmic (log10) transformation of CFU values was performed prior to statistical evaluation. The reported half maximal effective concentration (EC50) values were determined using the sigmoidal dose-response equation of nonlinear regression and are presented as EC50 ±95% confidence interval. Significant differences in bacterial counts and LF concentrations among animal treatment groups were determined using the Mann-Whitney rank-sum test for non-parametric data. Host survival was analyzed according to the Kaplan-Meier product limit method; pair-wise comparisons were made using the log-rank test.
|
10.1371/journal.pntd.0002112 | Spatiotemporal Transmission and Determinants of Typhoid and Paratyphoid Fever in Hongta District, Yunnan Province, China | Typhoid and paratyphoid fever are endemic in Hongta District and their prevalence, at 113 per 100,000 individuals, remains the highest in China. However, the exact sources of the disease and its main epidemiological characteristics have not yet been clearly identified.
Numbers of typhoid and paratyphoid cases per day during the period 2006 to 2010 were obtained from the Chinese Center of Disease Control (CDC). A number of suspected disease determinants (or their proxies), were considered for use in spatiotemporal analysis: these included locations of discharge canals and food markets, as well as socio-economic and environmental factors. Results showed that disease prevalence was spatially clustered with clusters decreasing with increasing distance from markets and discharge canals. More than half of the spatial variance could be explained by a combination of economic conditions and availability of health facilities. Temporal prevalence fluctuations were positively associated with the monthly precipitation series. Polluted hospital and residential wastewater was being discharged into rainwater canals. Salmonella bacteria were found in canal water, on farmland and on vegetables sold in markets.
Disease transmission in Hongta district is driven principally by two spatiotemporally coupled cycles: one involving seasonal variations and the other the distribution of polluted farmland (where vegetables are grown and sold in markets). Disease transmission was exacerbated by the fact that rainwater canals were being used for disposal of polluted waste from hospitals and residential areas. Social factors and their interactions also played a significant role in disease transmission.
| Typhoid and paratyphoid epidemics are serious events in low-income countries; these diseases are notorious for their high infection rate, long duration, and heavy health burden. In China, typhoid and paratyphoid are considered to be under control, although the situation varies considerably from place to place. During 2010 the disease incidence was 1.2 per 100,000 at the national level. The highest incidence, 113 per 100,000, occurred in the Hongta District of Yunnan province, in southwestern China. We used quantitative spatiotemporal analysis to explore the relationship between the incidence of these diseases and a number of factors suspected of influencing their occurrence. We found that cases tended to occur near discharge canals and polluted farmland where vegetables are grown for sale in local markets. The spatial characteristics of disease transmission were associated with the seasonal variations common to all intestinal infectious diseases. The findings of this work could inform local public health planners and the health directorate and help to improve public health intervention programs in regions with the highest incidence of these diseases.
| Typhoid and paratyphoid fever are serious infections, particularly in low-income countries, causing about 16 million cases and 600,000 deaths annually worldwide [1]. These diseases are notorious for their high infection rate, long duration, and heavy health burden. In China, typhoid and paratyphoid have been recorded daily since 2004 by the National Infectious Diseases Reporting Information System (NIDRIS) of the Chinese CDC, which enables all health care institutes across the country to report individual case information of important infectious diseases in real time using the Internet. Typhoid and paratyphoid fever are now regarded as being under control nationwide, and during 2010 the national disease incidence was 1.2 per 100,000.However, prevalence, varied considerably from place to place with the highest incidence of 113 per 100,000,occurringinHongta district of Yunnan province in southwestern China [2], [3].
Water and food sanitation and environmental awareness generally can effectively reduce food and water borne diseases. A total of 3538 cases of typhoid and paratyphoid fever were reported in Hongta district during the period 2006–2010. Sanitation conditions in general are better for Hongta residents than for people in other parts of the province: a previous investigation showed that 50% of the population did not drink unboiled water, 91% washed their hands before dinner and 79% washed fruit before eating it. The case-control study [4], which included80 pairs of cases and controls, showed that adding fresh mint (OR = 2.17, 95%CL: 1.04–4.54) to breakfast, eating uncooked vegetables (OR = 2.29, 95%CL: 1.24–4.24) at restaurants or roadside food sites, and eating flavoring that contained fresh caraway and mint (OR = 2.38, 95%CL: 1.00–5.69) are all risk factors for typhoid and paratyphoid fever in the Hongta district. The incidence reaches its highest peak during the June–October period, and then decreases, reaching its lowest value during February. This pattern is consistent with the seasonal characteristics of other intestinal infectious diseases. During the high incidence period the weather is hot and humid, allowing more rapid bacteria reproduction; also during this weather people often eat more raw and cold food. The cycle has been repeated for many years and the situation has become chronic. Furthermore, in the Hongta district, infection is often left untreated and only 42.6%–47.9% of carriers receive medical care. This leads to a large number of infection sources and carriers, and contributes significantly to the persistence of these diseases in the region.
In recent years, intervention has become a crucial determinant of disease transmission [5]–[7]. During 2008, the typhoid and paratyphoid fever control program (2008–2010) in the Hongta district focused on the restriction of fresh caraway and mint eating in restaurants and roadside food sites, the surveillance and hospital treatment of patients and carriers using appropriate antibiotics, improvement of the environment, health education, food hygiene(especially for fresh vegetables), drinking water disinfection and vaccination. As a result, the morbidity in the Hongta district dropped from 232 per100,000 during 2000–2007 to 113 per 100,000 in 2010, a decrease of about 51%, but still high compared with other regions of China.
Intervention would be much more efficient if the disease sources, transmission paths and factors were accurately identified. According to other studies [8]–[16], and our present work, the following disease transmission hypotheses are proposed: (1) Polluted water from hospitals and residential sewage discharged into canals is an important disease determinant. (2) Polluted vegetables sold in the markets constitute another determinant. (3) Social and environmental factors interact together to influence disease transmission.
The environmental health processes and variables involved in the above hypotheses are distributed across space and time [17]. Understanding this pattern should provide valuable clues about disease sources and determinants. The disease pattern can be rigorously addressed using stochastic spatiotemporal analysis [18]–[20] and powerful GIS technology [21]–[25] available at a relatively lower cost than the mainstream epidemiological techniques. A non-random disease distribution will always display spatial disease clusters [22]. The concordance between the spatial pattern of a disease and that of a contributing factor or determinant usually indicates the power of that determinant [23], [24]. Integration of GIS with statistical models of disease distribution forms a nationwide web-based automated system for early disease outbreak detection and rapid response in China [25]. The dataset and tools have already been used in the analysis of typhoid transmission [26]–[28]. The main objective of the present study is to identify the sources, transmission processes, and determinants of typhoid and paratyphoid fever in the Hongta District, based on the space-time analysis of the available epidemiological and notification data.
Data for the typhoid and paratyphoid cases for the Hongta district, and the factors associated with it were collected and a conceptual model was developed to study the proposed transmission hypotheses based on the available data. Transmission sources and population exposure were studied using geographical detectors and epidemiological surveys.
The Hongta district is located in the center of the Yunnan province, with coordinates 24°08′N–24°32′N, 102°17′E–102°41′E, at an elevation of 1630 m, (see Figure 1). The district's climate is subtropical and sub-humid. The total area is 1004 km2, 85% of which is mountains and the rest is irrigated farmland. The total population is 420,553, of which 268,635 are rural people. There are 11 towns and 81 administrative villages. The per capita income of the farmers is 4432 CNY, which lies in the middle of the range of farmers' income country-wide.
Data on incidence included both probable and laboratory-confirmed cases as defined by the national Salmonella typhi and paratyphoid standards. Illness caused by S.typhi is often characterized by insidious onset of sustained fever, headache, malaise, anorexia, relative bradycardia, constipation or diarrhea, and non-productive cough. Laboratory criteria for diagnosis are based on positive isolation of S. typhi or paratyphi in blood, stool, or other clinical specimen. A confirmed case is defined as a clinically compatible case that is laboratory-confirmed. It was estimated that the proportion of unreported cases was below 5%. These cases generally concerned mild illnesses identified in both hospitals and community. Daily numbers of confirmed typhoid and paratyphoid cases during the period 2006–2010 were obtained from the national infectious diseases reporting information system of the Chinese CDC. The system covers 95% of the country's population and provides certain patient information, including personal identification number (ID), specimen used for testing paratyphoid pathogen, and family address defined by GPS. Typhoid and paratyphoid cases were diagnosed according to clinical symptoms and blood culture Salmonella-positive test (the national diagnostic criteria for typhoid and paratyphoid fever were used [29]). All study data were stripped of personal information.
Demographic and socio-economic data were obtained from the local Statistics Bureau and included GDP per capita, number of medical institutions, population, percentage of farmers in the population, and geographical distribution of the villages.
Annual precipitation and other meteorological data were available from the China meteorological data-sharing network, and were interpolated using the spatial kriging technique to cover the entire study area.
Geomorphic data were obtained from the China geomorphy map (1∶1,000,000 scale), and seven categories were defined: plain, terrace, hill, low-relief mountain, middle-relief mountain, high-relief mountain and extremely high-relief mountain. Geomorphic data were provided by the state key Laboratory of Resources and Environmental Information Systems from the Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Science.
The Normalized Difference Vegetation Index (NDVI), acquired by satellite remote sensing imaging, reflects greenness of vegetation, photosynthetic rate, and vegetation metabolic intensity and measures seasonal and inter-annual changes. This makes NDVI an efficient indicator of vegetation and eco-environment in the locations where typhoid/paratyphoid Salmonellae occur. We used monthly NDVI values with spatial resolution 1 km by 1 km.
The spatial distribution of canals, markets, water wells, streets, and landuse in Zhoucheng town (Hongta district) were provided by the Yuxi city Bureau of Mapping and Survey. Field surveying showed that hospital disposal and residential sewage are connected and discharged into the rainwater canal system.
Water and vegetable samples were collected from markets and farmland, canals irrigating the farmland, hospital disposal products and residential sewage. The samples were tested by pulsed field gel electrophoresis, a technique used for the separation of and identification of Salmonella DNA.
The data were organized either into excel tables or handled by means of ENVI and ArcGIS 10.1 software.
Figure 2 outlines the study design. The observed yearly disease incidence for towns located in the Hongta district is shown in the top part of the diagram. The bottom part of the diagram depicts the infected patient and the route taken by the patient's unsterilized excretion products to water sources that are then used to irrigate vegetables grown on the farmland. Transmission to susceptible patients may occur either directly via flies or polluted water or via vegetables grown on the polluted farmland. Infected individuals may recover and become susceptible, and then become infected again. Untreated individuals may recover but remain carriers of the disease. As noted earlier, disease transmission mechanisms depicted in the bottom part of diagram and observations shown along the top, have suggested certain study hypotheses involving disease sources and factors (x), transmission (y∼x), and spatiotemporal patterns (y). These hypotheses can be investigated by means of rigorous spatiotemporal analysis and laboratory testing. Cases and suspected determinant data were collected for this purpose.
Spatial modeling goes beyond standard data analysis by providing new dimensions that can process additional information, discriminate factors that may not be perceivable (in simple time series analysis for example), and locate sources and factors for accurate intervention. We examined clustered and geographically correlated cases, spatial patterns linked to suspected sources, and case distribution-disease determinant consistencies suggesting transmission mechanisms. These were analyzed by means of, local indicator of spatial association (LISA), spatial buffering, and geographical detector techniques [24] respectively. Multivariate analysis was performed by the interaction detector technique, measuring the spatial consistency between disease distribution and a distribution formed by overlaying the distributions of the multivariate of the determinants. Multivariate regression was not suitable because either the sample size was too small (9 towns in Hongta district) or the associated explanatory variables were not available (in villages and streets).
Local Moran's I, i.e. LISA measures the spatial autocorrelation of an attribute as [22]where xi is the attribute value in the spatial cell i; c is the matrix of spatial weights, e.g., cij denotes the weight of relationship between cells i and j; and s2 denote the mean and variance of x, respectively. A positive Ii implies that the cell i and its neighbors are similar; e.g., high-high or low-low cells. A negative Ii implies that the cell i and its neighbors are dissimilar; e.g., high-low or low-high cells. LISA was performed using the publically available software GeoDA software.
An object's buffer refers to a zone covering a specified distance around it. For a single object a series of buffers may be considered at increasing distances. If the object is a point (e.g. a market), its buffers are a series of circles centered at that point. If the object is a line (e.g., a canal), its buffers are a series of belts parallel to the line. If the prevalence in the buffers exhibits a trend with distance from an object, it implies that the specified object is associated with disease transmission. Buffer statistics were calculated using ARCGIS software.
The geographical detector [24] is used to assess environmental risks to human population. The method is different from conventional mathematical correlation or regression, which require that the determinants or their proxies are quantitative. It is assumed that the disease will exhibit a spatial distribution similar to that of an environmental factor if the environment contributes to disease transmission, as measured by the Power of Determinant (PD),where ℜ and σ2 denote the area and the dispersion variance of disease incidence of the study area, respectively. The study area is stratified into L stratums, h = 1, …, L [23] according to spatial heterogeneity of a suspected determinant or its proxy of the disease. Note that spatial heterogeneity is defined as an attribute whose statistical properties (e.g., mean and standard deviation) change in space. PD ∈[0, 1], where 1 indicates that the determinant completely controls the disease, and 0 indicates that the determinant is completely unrelated to the disease. In other words, PD expresses the extent to which a determinant explains disease incidence. In addition, disease transmission is often determined by multiple factors. The PD, combined with spatial overlay techniques of GIS and set theory, yields a geographical detector that can effectively assess the relationship between multiple factors: Are these factors independent or interacting? Are they enhancing or weakening each other? Is their interaction linear or nonlinear? In the present study, the geographical detector was implemented using the software www.sssampling.org/geogdetector (publicly available).
The overall disease distribution as a function of time is shown in Table 1. Figure 3 shows that the disease incidence is in inverse phase with NDVI (Figure 3a) and in the same phase as precipitation (Figure 3b).
Figure 4 displays yearly incidence in 9 towns of the Hongta district during the period 2006–2010, with the exception of 2009, together with population densities and landuse. Towns are categorized as having low (L), medium (M) and high (H) population densities. Zhoucheng town is located at the south-eastern and central area of Hongta district and is made up of 51 villages. It has the highest population density and much higher incidence than towns in the western and northern hill woodland, which is why it was made the focus of this study. LISA statistic (P = 0.01) in Figure 5, shows that low incidence villages are close to other low incidence villages in the outskirts, whereas villages forming high incidence clusters are located close to the town centre. Higher disease incidence occurs in areas with a larger proportion of farmers in the population. Data relating to distribution of canals and markets are also available.
Figure 6 displays the associations between disease incidence and suspected determinants in Zhoucheng town. The x-axis represents the distance between sewer and confluence (in Figure 6a) and between market and well (in Figure 6b), shown with their respective buffer zones. The y-axis denotes years (2006–2010), and the z-axis the population incidence. Figure 6a shows that the population incidence increases with decreasing distance from the canals. Interestingly, the incidence reduces rapidly away from confluence canals (note: blue bars rapidly shorten along buffer-axis), whereas it reduces rather slowly away from sewage canals (red bars slowly shorten along buffer-axis).Figure 6b shows that disease incidence reduces with distance from markets and groundwater wells, the decline being more rapid from the markets (red bars rapidly shorten along buffer-axis) than from the wells (blue bars slowly shorten along buffer-axis). Figures 6a and b, show that the abnormally high incidence during 2009 occurred within a 500 m buffer zone of sewers and confluences and an 800 m buffer zone of markets and wells, suggesting that the suspected disease sources are not only a cause of the observed spatial disease distribution but they were also responsible for the 2009 disease outbreak.
After being standardized as (y−min)/(max−min), the time series of incidence, precipitation and NDVI were analyzed. The typhoid fever and the paratyphoid fever both fluctuated in time, experiencing two peaks per year; the first peak from April to June and the second from September to November. The incidence was positively correlated with the precipitation time series and usually peaked about one month earlier or one month later than the precipitation peak. NDVI is an index reflecting biological mass, and its time series appears to be negatively correlated with disease incidence.
Laboratory testing indicated that the Salmonella strains found in water samples from hospitals, canals, farmland, and market vegetables belonged to closely related clones of S. typhi. Vegetables sold in the markets came from farmland irrigated by water contaminated by hospital and residential sewage. The findings of our spatiotemporal analysis, that hospital and residential sewage discharge and food markets are the most probable routes of disease transmission, support previous surveys carried out in this area.
In addition to these direct disease sources, other environmental factors are believed to exert an influence on disease transmission. These include factors such as economic and social status, types of disease intervention, climate-biology types and physical conditions. However, these factors are rather difficult to measure in a straightforward manner. Instead, they are represented by their proxies: GDP per capita (economic determinant), proportion of farmers in total population (social determinant), number of medical institutions (intervention), NDVI (climate-biology type), and geomorphic types (environmental conditions).
Using the geographical detectors it was found that social factors have the highest disease impact of all those measured (Table 2). The combined impact of two factors A and B on disease incidence generally differs from the simple linear summation of the separate A and B impacts, i.e., A ∩ B≠A+B. The spatial incidence distribution is highly correlated with that of farmer proportion and population density. The proportions of spatial prevalence variation explained by these two factors reach PD = 78.3% (for A) and 73.8% (for B). The farmer proportion and population density together explain 84.7% of the variation of disease incidence, higher than that explained by any single factor.
The incidence is weakly associated with the economic condition (GDP per capita) and the climate-biological type (NDVI), with PD values of 9.7% and 1.8%, respectively. However, the incidence is more strongly linked with the spatial combination of GDP and NDVI. More than half (53.8%) of the spatial variation of the incidence can be explained by the spatial distribution of economic condition (GDP per capita) combined with availability of medical intervention (number of medical institutions). The individual determinant impact of GDP and NDVI are clearly low (8.7% and 18.7% respectively). Geomorphology explains only a small part of spatial disease transmission, but its impact increases when it is combined with population density or GDP per capita. These findings jointly imply that a policy instrument based on analysis of a combination of the most relevant determinants,(in this case, population density and farmer proportion), would be highly efficient in controlling spatial disease transmission in a region.
Based on the results of rigorous spatiotemporal analysis and GIS technology, disease transmission processes can be identified, modeled, and used for population health management purposes. The findings of the present Hongta district study include the following: Unsterilized water from hospital disposal and residential sewage are discharged into the rainwater canal system to irrigate the region's farmlands where local vegetables are grown. Polluted vegetables are sold in the markets and consumed without being adequately washed. Infection is magnified during the rainy seasons, when more contaminated water irrigates the farmlands and the polluted water, mixed with additional garbage spreads over a larger area of the inhabited region. Over time, the combination of the farmers' poor hygiene conditions and low cure rates lead to considerable pathogen reserves among the population. The general seasonal-intestinal infection cycle, driven mainly by annual temperature fluctuations and varying volume of polluted canal water in the irrigation system driven mainly by the rainy seasons, combine to cause the persistent high endemic transmission of the disease in the Hongta district.
The availability of spatiotemporal data and the development of health information systems facilitate the GIS-spatiotemporal exploration of public health issues. The physical and social determinants of health are usually spatially distributed: surface and subsurface contaminated water, polluted water emission from workplaces, disease prevention methods, and nutrition and food habits. More importantly, the susceptible and the infected populations are always geographically distributed. Health determinants may be detected when the disease incidence shares similar spatial features with environmental features suspected of being associated with the disease. When disease cases and environmental features do not share similar spatial features, however, this does not necessarily mean that they are unrelated. Instead, the geographically dispersed disease cases may have the same casual sources as features which may be elucidated by alternative techniques such as molecular tracking in food or serum samples, and food electronic label tracking systems.
This study identified some of the disease sources, determinants, and transmission cycles, which affect the distribution of typhoid and paratyphoid cases in the Hongta district which currently exhibits the highest incidence in China. This knowledge can be used to design intervention measures to reduce and hopefully eradicate the disease in this area. In addition to the measures common to the control of enteric infectious disease intervention, such as washing hands before eating and environmental sanitation, particular attention should be paid to the sterilization of hospital waste prior to discharge into canals, especially during the rainy season.
|
10.1371/journal.pgen.1002127 | Parallel Adaptive Divergence among Geographically Diverse Human Populations | Few genetic differences between human populations conform to the classic model of positive selection, in which a newly arisen mutation rapidly approaches fixation in one lineage, suggesting that adaptation more commonly occurs via moderate changes in standing variation at many loci. Detecting and characterizing this type of complex selection requires integrating individually ambiguous signatures across genomically and geographically extensive data. Here, we develop a novel approach to test the hypothesis that selection has favored modest divergence at particular loci multiple times in independent human populations. We find an excess of SNPs showing non-neutral parallel divergence, enriched for genic and nonsynonymous polymorphisms in genes encompassing diverse and often disease related functions. Repeated parallel evolution in the same direction suggests common selective pressures in disparate habitats. We test our method with extensive coalescent simulations and show that it is robust to a wide range of demographic events. Our results demonstrate phylogenetically orthogonal patterns of local adaptation caused by subtle shifts at many widespread polymorphisms that likely underlie substantial phenotypic diversity.
| Identifying regions of the human genome that differ among populations because of natural selection is both essential for understanding evolutionary history and a powerful method for finding functionally important variants that contribute to phenotypic diversity and disease. Adaptive events on timescales corresponding to the human diaspora may often manifest as relatively small changes in allele frequencies at numerous loci that are difficult to distinguish from stochastic changes due to genetic drift, rather than the more dramatic selective sweeps described by classic models of natural selection. In order to test whether a substantial proportion of interpopulation genetic differences are indeed adaptive, we identify loci that have undergone moderate allele frequency changes in multiple independent human lineages, and we test whether these parallel divergence events are more frequent than expected by chance. We report a significant excess of polymorphisms showing parallel divergence, especially within genes, a pattern that is best explained by geographically varying natural selection. Our results indicate that local adaptation in humans has occurred by subtle, repeated changes at particular genes that are likely to be associated with important morphological and physiological differences among human populations.
| Although the predominant population genetics model of adaptation assumes novel advantageous alleles sweep to fixation [1], [2], most putative examples of adaptive divergence between human populations lack the full signature of a classic hard sweep [2]–[10]. In fact, classic sweeps may have played a negligible role in the evolutionary changes that have occurred since the most recent common human ancestor [10], prompting the question of whether the moderately large allele frequency differences observed among modern human populations at a small proportion of loci are indeed adaptive. These divergent loci may indicate more complex and subtle modes of selection, but it is often difficult to demonstrate with statistical confidence that they are not merely the tail end of a stochastic neutral distribution. However, if selection acts independently on the same loci in different geographical locations, data from multiple populations can be leveraged to provide strong evidence for non-neutral evolution. Such parallel adaptation among populations of the same species has been identified in sticklebacks [11], [12], whitefish [13], Drosophila [14]–[16], and other taxa [17], [18]. The extent to which the same genes underlie repeated adaptive events is unclear, but a growing number of observations suggest that parallel evolution at the molecular level may be quite prevalent [19]–[21], and from first principles it seems especially likely when selection acts on the same standing genetic variation in closely related populations [2], [22]. Demonstration of parallel evolution among populations provides strong support for the hypothesis that repeated selection of the same alleles in distinct environments is an important mechanism of local adaptation. Studying this evolutionary process can also generate a list of candidate sites that likely have functional phenotypic consequences and provide insight into which environments present similar selective pressures.
In this study, we develop a novel approach to test the hypothesis that parallel adaptive evolution has shaped extant patterns of human genomic variation. Our method evaluates a set of independent SNPs genotyped in populations that can be clustered into at least four groups forming two or more phylogenetically distinct, allopatric group pairs (Figure 1). The goal is to test whether the same SNPs show high divergence in phylogenetically independent contrasts [23] between pairs of groups more often than expected under neutrality. To this end, we calculate pairwise FST for all SNPs between all pairs of groups. For each pair of groups, we identify divergent SNPs that exceed the 95th percentile of FST values, and define parallel divergent SNPs as those that are divergent in two independent group pairs. A significant excess of parallel divergent SNPs is interpreted as evidence for parallel adaptive divergence. We demonstrate substantial parallel adaptation in human populations and we characterize the genomic and geographical patterns of the parallel divergent SNPs.
We applied our method to the Human Genome Diversity Project (HGDP) data, which consists of approximately 1,000 individuals genotyped for over 650,000 SNPs [24]. We only analyzed a subset of these data that met our assumptions of independence, chosen as follows. We combined 19 of the HGDP populations, consisting of 343 individuals, into six ecologically and genetically distinct groups (Figure 1A). Fifteen divergence comparisons are possible in this dataset, where each divergence comparison contrasts FST values calculated in two phylogenetically independent pairs of groups. Although it is difficult to completely rule out cryptic gene flow, we carefully examined these groups for any evidence of admixture. Previous population structure analysis indicates that the particular samples that constitute these groups are distinct lineages [25]. Possible gene flow between Pygmies and West Africans is irrelevant to this analysis as these sister populations are never assigned to different groups in a divergence comparison. The only other detectable gene flow signature among these populations is low-level introgression from East Asia into Oceania [25]; therefore the three divergence comparisons that could potentially be confound by this introgression, all comparing East Asia–South America FST against FST between Oceania and another population, are evaluated with this caveat in mind. To supplement these prior population structure results, we confirmed that the established phylogenetic topology has the significantly highest likelihood (Shimodaira and Hasegawa tests; p<0.001). Furthermore, after removing divergent SNPs exceeding the 95th percentile, we regressed the ranks of FST values at remaining SNPs for each group pair against the corresponding ranks for other group pairs and found that non-divergent FST values are uncorrelated (Bonferroni-corrected p>0.05 for all divergence comparisons). This result suggests that admixture has had a negligible effect on these populations, since high levels of gene flow would cause SNPs to show relatively similar FST values in multiple group pairs.
To address potential confounding factors in our analysis, we only considered a subset of all HGDP SNPs to meet the assumptions of our neutral model (Figure 1B). One confounding factor is variation among SNP frequencies in the common human ancestor population, since the expected divergence due to genetic drift depends on the ancestral minor allele frequency; therefore SNPs with skewed global frequency, a proxy for the ancestral frequency, were excluded. Another confounding factor is mutation rate; therefore only globally polymorphic SNPs showing variation within all group pairs were used, to maximize the probability that alleles are identical by descent and not new mutations. We identified 111,724 HGDP SNPs that were intermediate frequency and globally polymorphic. Within this set, linked SNPs are not independent and ascertainment bias may be more pronounced in some populations [26]; therefore only SNPs showing low linkage disequilibrium with each other and collectively similar site frequency spectra in all groups were analyzed together (Figure S1). We randomly selected 1000 subsets of SNPs (mean = 26,864 SNPs; range = 25,371 to 27,770 SNPs) meeting these criteria. Thus, the theoretically expected number of parallel divergent SNPs in a divergence comparison is 67.16 ( = 26,864*0.052; Text S1). An ideal dataset for this analysis would employ full sequence data rather than SNPs and potentially larger sample sizes than the 28 to 116 individuals per population that we used. In the absence of such a dataset, it is unlikely that any sampling scheme will produce a truly independent set of SNPs or precise estimates of allele frequencies, but our method represents a way to eliminate the most egregious potential errors while still retaining a larger number of SNPs for analysis.
In the HGDP dataset, fourteen of the fifteen divergence comparisons showed more parallel divergent SNPs than the expected value of 67.2 (Figure 2). The mean was 82.0 parallel divergent SNPs (range = 58 to 96), a 22% increase over the expected value. Eight divergence comparisons showed a mean of 81.0 or more parallel divergent SNPs, a significant excess according to our simulation results (see below) (Table S1; Figure 2; Figure 3). Four divergence comparisons showed a mean of 91 or more parallel divergent SNPs and therefore were significant even after a conservative Bonferroni correction (Fisher's exact tests, p<0.05). The highest levels of parallel divergence were observed when divergence between Pygmies and Europeans was compared to divergence between Oceanians and South Americans or East Asians. As the threshold for divergence was dropped to more stringent values below 5%, the excess of parallel divergent SNPs increased (Figure 4). There is no comparable excess of parallel conserved SNPs with extremely low divergence in multiple group pairs, suggesting that purifying selection is not the cause of correlations among lineages (Figure 3). The only divergence comparison showing fewer than expected parallel divergent SNPs compares Europe-Oceania FST to East Asia-South America FST. This is the only divergence comparison that does not include an African population, suggesting that similar selection pressures in African and non-African populations may be driving most of the observed patterns. In addition, as extensive migration between Asia and Oceania could lead to false positives in this divergence comparison, the relative dearth of parallel divergent SNPs suggests that such migration does not have a large effect on our results.
Many SNPs fell into narrow regions that were represented in a large number of the 1000 independently generated subsets (Table S2). The mean percentages of both genic and nonsynonymous parallel divergent SNPs were higher than mean for the set of all SNPs in a replicate, a relative excess that increased as the threshold was dropped to more stringent values below 5%, suggesting that positive selection is driving parallel divergence (Figure 4). Genes overlapping parallel divergent SNPs were modestly enriched for diverse functional categories associated with various cell types including neurons, lymphocytes, cancer, and epithelium (Table S1; Table S3; Figure 5). Among the most extreme parallel divergent genes (observed at a threshold of 0.5%) were the skin keratinization gene ABCA12 [27] (Figure 6); SH2B1, which controls serum letpin levels and body weight [28]; GRM5, a glutamate receptor associated with schizophrenia [29] and with pigmentation via the closely linked TYR [30]; ATP2A2, which causes a neuropsychiatric/keratinization disorder [31]; F13A1, a coagulation factor linked to numerous cardiovascular diseases and to Alzheimer's [32]; and IFIH1, associated with antiviral defense, type 1 diabetes, and psoriasis [33]–[35] (Figure S2). The pleiotropic nature of many of these genes suggests that selection on one trait may have affected the evolution of other traits.
For all divergence comparisons, we calculated the orientation skew, defined as follows. For any divergence comparison, there are two possible allele frequency orientations for a given SNP. In the first pair of groups, one allele is at relatively higher frequency in the first group compared to the second. In the second pair, this allele can either be at relatively higher frequency in the first group or the second, defining the two orientations. The skew is simply the frequency of the more common, or major, orientation for all parallel divergent SNPs, which should be near 0.5 under the neutral expectation that both orientations are equally probable (expected skew is 0.55 for 67 parallel divergent SNPs, following the binomial distribution).
Eleven divergence comparisons showed a significant skew in the mean number of parallel divergent SNPs displaying each orientation across all replicates (major orientation frequency >60%; p<0.05, one-sided comparison to simulation results; Figure 2; Table S1). These results suggest similar selection pressures acting on disparate groups. For example, the most extreme skew (major orientation frequency = 85%) has Europeans diverge from Pygmies in the same direction as East Asians diverge from Oceanians, and there are a myriad of climatological, dietary, and disease variables that could exert similar selective pressures on the two temperate groups relative to the two tropical groups. In fact, four of the five divergence comparisons showing the most parallel divergence show this same general pattern, if Pygmies are considered interchangeable with West Africans and East Asians are considered interchangeable with South Americans (Table S1; Figure 2; Figure 3). This pattern suggests similar selective pressures on Africans and Oceanians relative to Europeans, East Asians, and South Americans. South Americans may carry alleles adapted to temperate climates due to their ancestral migration across Beringia, and they may have lacked adequate time and/or genetic variation to completely re-adapt to a tropical environment. One SNP that fits this hypothesis lies in DDB1, which protects the skin from solar UV exposure [36], and is one of the strongest examples of this parallel divergence pattern, with one allele fixed in South America, over 90% in Europe and East Asia, and less than 40% in African and Oceania. Another possible selective pressure driving this pattern may be parasites historically restricted to the Old World tropics; indeed, several immune-related genes like LRBA and LAT show this parallel divergence pattern even at thresholds well below 5% (Table S2).
To calculate the probability that neutral evolution would generate the observed number of parallel divergent SNPs, we performed nine sets of coalescent simulations, each comprised of 100 full SNP datasets equivalent to our empirical data, under various demographic scenarios including bottlenecks, migration, and growth (Table S4). Some simulations used realistic sets of parameters while others were not realistic for these human populations but tested the effects of diverse demographic events. For each simulated divergence comparison, we calculated the number of parallel divergent SNPs and the orientation skew.
Our primary model (standard) employed realistic, previously calibrated demographic parameters [37] featuring recent growth in all populations, no migration among populations, and bottlenecks in all populations, including several population sizes under 1000 individuals over 50 generations. Under standard, the mean number of parallel divergent SNPs among all simulated divergence comparisons was 67.22 (standard deviation among 1500 divergence comparisons = 7.85; standard deviation among 100 dataset means = 2.37; range of dataset means = 61.27 to 72.28), nearly identical to the theoretical expectation of 67.16 (standard deviation = 7.79; standard deviation of means = 2.01) (Figure 2; Figure 7; Text S1; Figure S3). Out of the 1,500 standard simulated divergence comparisons, replicated 10 times each, a mean of 81.0 or more parallel divergent SNPs was observed 2.47% of the time, suggesting that empirical divergence comparisons with at least 81 parallel divergent SNPs exceed the 95% confidence interval. A Bonferroni-corrected significant excess of parallel divergent SNPs (91 or more, theoretically expected 5% of the time if divergence comparisons are independent) was never observed in standard (p<0.01). The mean orientation skew was 0.54, and 4.7% of divergence comparisons showed an orientation skew greater than 0.60.
Eight variations on standard revealed that our method is in general robust to demographic perturbations, but that extreme demographic events can cause slight excesses of false positives, defined here as values at least 10% greater than the theoretical expectation (Table S4; Figure 7). Eliminating bottlenecks in some or all populations does not cause an excess of false positives, nor does moderate introgression between populations of up to 10 migrants per generation. False positives are produced by extreme bottlenecks of 150 individuals over 50 generations (3tightestbottle), extreme migration of 100 migrants per generation (OC-EAmig100), or low long-term effective size (constantlow). However, no simulation model produced enough false positives to match the empirical results, as mean simulated values for both parallel divergent SNP counts and orientation skew were universally significantly lower than the empirically observed mean results (Figure 7). In the results from two models implemented to represent unrealistically extreme demographic events, 3tightestbottle and OC-EAmig100, a minority of individual simulated divergence comparisons (2 or 3, respectively, out of 15) were comparable to our most significant empirical divergence comparisons (>90 parallel divergent SNPs and/or >0.70 orientation skew) in more than 5% of simulated datasets (Table S4). Presumably, bottlenecks or migration events would have to be even more extreme or prevalent than in these simulations in order to be solely responsible for the mean empirical values across all divergence comparisons. Thus, demographic events are unlikely to be the cause of our empirical results.
The geographic distribution of human genetic diversity shows a pattern of parallel divergence, such that the same variants have undergone exceptionally high divergence repeatedly (Figure 2; Table S2). Both the counts of parallel divergent SNPs and the orientation skews are higher than expected under a neutral framework, which suggests that spatially varying selective pressures are partially responsible. Because observed parallel divergent SNPs did not exceed the expected value at a 5% threshold by more than twofold, the majority of parallel divergent SNPs are probably neutral false positives and not adaptive variants. The relative excess of parallel divergent SNPs increases as the threshold is lowered (Figure 4), so candidates at lower thresholds are more likely to be real adaptive variants, but even at these lower thresholds the evidence for selection on any given SNP reported here (Table S2) is relatively weak in the absence of additional experimental data. Rather than pinpointing specific candidates with high confidence, the main strength of our results is instead to emphasize the broad genomic trend of parallel adaptive divergence.
Our simulations indicate that particular demographic models can cause patterns of parallel divergence similar to those that could be caused by selection. Although this caveat means that our results must be interpreted with caution, demography alone is unlikely to responsible for the empirically observed patterns. None of our models produced mean counts of parallel divergent SNPs or mean orientation skews as high as we observed in the empirical data (Figure 7), and those that did produce unusually high counts of parallel divergent SNPs represented very extreme demographic events (Table S4; Figure 7). Specifically, we only observed an increase in false positives if gene flow occurred on the order of one hundred migrants per generation (OC-EAmig100), if bottlenecks occurred with an effective size of 150 individuals over 50 generations in multiple populations (3tightestbottle), or if effective size was consistently low with no growth (constantlow). None of these three models is particularly realistic for the human populations examined here. While previous analysis suggests that there has been a small amount of admixture between East Asia and Oceania, which could potentially confound three of our divergence comparisons [25], it is unlikely to have been as high as one hundred migrants per generation [38]. Migration between Pygmies and West Africans, or between East Asians and South Americans, would not confound our assumptions of population independence because these populations never appear in different group pairs within the same divergence comparison. Other potential instances of intercontinental gene flow, such as introgression from Europe to South America, do not appear to have contributed to the ancestry of these particular samples, which show negligible evidence of such admixture [25]. Bottlenecks have certainly played an important role in human history, and their magnitude is difficult to estimate with precision, but multiple bottlenecks as extreme as those simulated in 3tightestbottle would have been unlikely. For example, the Pygmy bottleneck effective size is estimated to have been from ∼500 to several thousand individuals [39], the effective female population size during the colonization of America from Asia has been estimated as ∼1000 individuals [40], and although Oceania may have been founded by very few individuals [38], its cumulative effective size as suggested by modern genetic diversity has been slightly higher than for American populations [25]. As recent human population growth is readily apparent, the constant low effective size simulated by constantlow is overly simplistic. Finally, patterns such as an excess of genic and nonsynonymous SNPs (Figure 4) cannot be attributed to demography and imply a prominent contribution by natural selection. Thus, even with the demographic caveats, the highly significant signal of parallel divergence across multiple group pairs suggests that parallel adaptation is an important feature of at least some of these SNPs in some of these populations.
The patterns of parallel adaptive divergence we identified do not reflect classic selective sweeps and therefore suggest more complex modes of selection have shaped human genomic diversity [10]. Our results are consistent with fluctuations in the allele frequencies of standing variation, as in soft sweeps [2], [22], although alleles have rarely actually swept to fixation. The tempered changes at most loci suggest that they encode quantitative polygenic traits that have reached new optima [2], that fitness landscapes fluctuate too rapidly, that the sweeps are still ongoing, and/or that gene flow prevents fixation. Our approach is based on the hypothesis that adaptive variants, or variants closely linked to them, were present in the ancestral human population. Parallel adaptation may also occur via independent adaptive mutations [5], [41]. It is also conceivable that a favorable mutation arising in one population could spread via adaptive introgression to other populations even if gene flow were too low to be detectable at neutral markers. Our method does not preclude the possibility of such newly arisen alleles, although they are less likely to be globally polymorphic. These scenarios still require similar selective pressures acting in multiple distinct populations and thus represent alternate forms of parallel adaptive divergence.
Our method complements existing strategies for detecting intraspecies non-neutral divergence. Most previous studies have focused on identifying loci that differentiate individuals on one continent from all others [4], [6], [8], and thus contribute to the same patterns of population structure generated by neutral processes [2], [7]. While these unique dramatic adaptive events have undoubtedly been important in human evolution, we have shown that phylogenetically orthogonal patterns are also a major component of geographically varying selection. One promising approach for detecting local adaptation is to compare allele frequencies to environmental variables, while controlling for population structure [9]. Our method differs in that we allow for parallel divergence among any group pair regardless of any obvious environmental similarities, and thus we can detect the effects of more cryptic selection pressures. We anticipate that both approaches will be fruitful in uncovering fundamental patterns of local adaptation at the molecular level. Finally, in contrast many other genomic scans for selection [6], a strength of our method is that we do not merely identify the most extreme outliers; rather, we test whether outliers showing parallel divergence are significantly more frequent than expected under neutrality. Future studies on parallel divergence could infer the haplotype backgrounds of selected SNPs via sequencing or denser SNP genotyping, in order to estimate the lengths of the chromosomal regions affected by selection, to further pursue evidence of cryptic gene flow, and to pinpoint causal adaptive SNPs.
In summary, we have demonstrated a statistically significant excess of parallel divergent SNPs in a set of human populations, relative to both the theoretical expectation under neutrality and the values observed in neutral simulations. Although it is difficult to completely rule out the effects of demography on genomic patterns, our simulations and the inferred histories of these populations indicate that a non-adaptive explanation is unlikely. Thus, our results provide statistical support for a major feature of the human evolutionary process: that the same genes are selected independently in multiple environments. Feasible adaptive solutions to selective pressures are therefore limited and are reused in separate lineages [20], [42]. Our approach may gain additional power and lead to new insights with the coming availability of full sequence data from numerous human populations, as well as data from non-human species.
We united HGDP populations into six phylogenetically independent groups of at least twenty-five unrelated individuals based on climatic and subsistence designations [9] and population substructure analysis [25]: Pygmy tropical hunter-gather (PY; Biaka and Mbuti), West African tropical horticultural (WA; Mandenka and Yoruba), European temperate agricultural (EU; French, Basque, North Italian, Orcadian, Sardinian, Tuscan), East Asian temperate agricultural (EA; Han, Japanese, Miaozu, Tujia), Oceanian tropical horticultural (OC; Papuan and Melenesian), and South American tropical horticultural (SA; Colombian, Karitiana, Surui). The remaining HGDP populations were excluded either because they showed evidence of admixture between disparate sections of the phylogeny (e.g. Middle Eastern and Central Asian populations), or because they were ecologically or genetically distant from the other populations and the sample size was insufficient for a new group (e.g. San, Pima, Tu). We used an adjusted autosomal SNP dataset for which SNPs with missing data were either abandoned or had missing values estimated based on Hardy-Weinberg equilibrium [43]. We conducted all analyses using only SNPs showing polymorphism in all eleven pairs of analyzed groups (“globally polymorphic”) and showing a global minor allele frequency greater than 0.4 (“intermediate frequency”).
In order to confirm the previously published phylogeny [24], we used the contml package in PHYLIP [44] to determine the evolutionary relationships among our six groups based on allele frequencies at globally polymorphic, intermediate-frequency SNPs. We tested whether the topology with the highest likelihood was significantly better than alternate topologies using Shimodaira and Hasegawa tests [45] with α set at 0.001. Furthermore, for each divergence comparison, we tested whether FST values were correlated among SNPs that were not divergent; such a correlation would suggest that the pairs were not independent due to gene flow. After ranking each SNP in each divergence comparison according to FST values, we removed all divergent SNPs (exceeding the 95th FST percentile) in order to eliminate SNPs under divergent selection. We then regressed the remaining ranks for each pair against each other. We tested whether any divergence comparison showed a significant correlation after a Bonferroni correction (α = 0.05/15 = 0.0033).
For the purpose of calculating linkage disequilibrium in large, panmictic demes, we chose the one population from each group with the largest sample size (Biaka, Mandenka, Sardinian, Karitiana, Papuan, and Han). In each population, we calculated composite pairwise linkage disequilibrium (D) among all globally polymorphic, intermediate frequency SNPs within 10 Mb of each other. We then calculated global D as the average among these populations. For each of 1000 replicates, we randomly selected a combination of globally polymorphic, intermediate frequency SNPs showing pairwise D less than 0.1 (“unlinked”). We did this by randomly selecting unlinked SNPs one by one, each time recalculating the site frequency spectra, defined as the count of SNPs with a minor allele frequency in each of ten equal bins (intervals of 5%), in each group. In order to minimize the population-specific effects of ascertainment and demography, we did not allow the size of any of the ten bins in the site frequency spectrum in any group to deviate from the corresponding bin in another group by more than 5%, unless the difference was under 100; thus, in our final random sample all groups had very similar site frequency spectra (“similar-spectrum”; Figure S1).
For each pair of groups, we calculated pairwise FST [46] between the two groups at all chosen SNPs and we ranked these FST values. We defined a SNP to be divergent for a particular group pair if its rank exceeded the designated threshold (top 5% for most analyses). If a SNP was divergent between two phylogenetically independent pairs of groups (a divergence comparison), we considered it to be a parallel divergent SNP. For each divergence comparison, we designated all SNPs as either parallel divergent, divergent in only one pair of groups, or not divergent, and tested the significance of these categories using both Fisher's exact test and comparisons to our simulated results (Text S1). We calculated the mean number per replicate of genic SNPs, nonsynonymous SNPs, and SNPs in genes associated with Gene Ontology (GO) or KEGG pathway terms [47]. We tested whether these categories were enriched for parallel divergent SNPs by using the random sample of globally polymorphic, intermediate-frequency, unlinked, similar-spectrum SNPs as the background, and the parallel divergent SNPs themselves, rather than the genes or regions they overlap, as the test group; this approach should preclude any effect of variation in SNPs per gene or in the probability that a SNP is included for analysis.
We tested for a bias in allele frequency orientation, a binary variable defined by which groups in each divergence comparison have relatively similar allele frequencies at parallel divergent SNPs. In order to examine whether observed orientations strayed from the neutral expectation of equal probability for both orientations, we calculated the mean number of parallel divergent SNPs with each orientation in each divergence comparison, across all replicates. We assessed the deviation from the binomial expectation using both Fisher's exact test and comparisons to our simulated results.
For each of nine demographic models, we used ms [48] to generate 100 independent coalescent-simulated datasets equivalent in size to the empirical HGDP dataset. Each simulation consisted of at least 12,000 unlinked regions representing 100 kb of DNA each, with a per-site mutation rate of 2×10−9 and a per-site recombination rate between 0 and 11.25×10−8, drawn from a distribution based on the empirical distribution of recombination rates in humans. Our standard model was a modified cosi demographic model calibrated using African, European, and Asian HapMap populations [37], splitting the African and Asian populations into two (representing PY and WA), and three (representing EA, OC, and SA) populations, respectively. Additional models were variations on this standard model (Table S4). We excluded all SNPs that were not globally polymorphic and intermediate frequency. For each of ten replicates for each of the 900 independent simulations, we randomly selected 26,864 SNPs to analyze, to match the empirical mean, controlling for LD and site frequency spectra in each random sample as we did with the empirical SNPs. Thus, we analyzed 9000 unique sets of 26,864 simulated SNPs using the same methods we used on the empirical dataset, including the analysis of fifteen distinct divergence comparisons, for a total of 135,000 simulated divergence comparisons.
|
10.1371/journal.pgen.1005075 | Clonality and Evolutionary History of Rhabdomyosarcoma | To infer the subclonality of rhabdomyosarcoma (RMS) and predict the temporal order of genetic events for the tumorigenic process, and to identify novel drivers, we applied a systematic method that takes into account germline and somatic alterations in 44 tumor-normal RMS pairs using deep whole-genome sequencing. Intriguingly, we find that loss of heterozygosity of 11p15.5 and mutations in RAS pathway genes occur early in the evolutionary history of the PAX-fusion-negative-RMS (PFN-RMS) subtype. We discover several early mutations in non-RAS mutated samples and predict them to be drivers in PFN-RMS including recurrent mutation of PKN1. In contrast, we find that PAX-fusion-positive (PFP) subtype tumors have undergone whole-genome duplication in the late stage of cancer evolutionary history and have acquired fewer mutations and subclones than PFN-RMS. Moreover we predict that the PAX3-FOXO1 fusion event occurs earlier than the whole genome duplication. Our findings provide information critical to the understanding of tumorigenesis of RMS.
| To decipher the dynamic mutational process and identify the causative genomic events in rhabdomyosarcoma (RMS), we developed a systematic method that incorporates multiple types of genomic information to estimate normal cell contamination, tumor clonality, and a timeline of somatic events that occurred prior to the tumor presentation. Our results demonstrate two distinct evolutionary paths resulting in PAX-fusion-negative-rhabdomyosarcoma (PFN-RMS) and PAX-fusion-positive-rhabdomyosarcoma (PFP-RMS): (1) In PFN-RMS, genomic loss of heterozygosity on chromosome 11p15.5 and non-synonymous mutations in RAS pathway and cell cycle genes (FGFR4, KRAS, NRAS, HRAS and CCDN1), as well as several genes not previously known to be drivers of RMS, including PKN1, CUL2, and TTK, occurs early in the evolutionary history of tumor; (2) In contrast, the PAX gene fusion event in PFP-RMS tumors is an early detectable event which consistently occurs prior to a whole genome duplication event. These findings provide new insights into the biology and molecular events that initiate RMS tumorigenesis and may help identify actionable drivers for targeted therapies.
| Cancer development is driven by dynamic mutational processes and selective pressures which allow a tumor to adapt over time from the initiating oncogenic lesion towards clinical presentation [1–4]. High coverage, next-generation sequencing technologies have provided an unprecedented view of the mutational landscape of whole cancer genomes and demonstrated that cancer genomes have typically acquired thousands of somatic alterations by the time they are clinically detected [5–8]. Although a majority of these alterations do not have clear biological consequences, some alterations are recurrently found; implicating them as critical events in that tumor’s evolution. Subsequently, external factors such as the tumor microenvironment and therapy can confer selective advantages that allow successful clones to eventually supersede one another [1,2]. Two of the major findings of next generation sequencing studies are that biopsies taken at the time of tumor presentation already contain a significant amount of genetic heterogeneity [5,6] and importantly that often a rare subclone in the primary tumor subsequently becomes the founding clone of a metastatic or relapse tumor [5,6,9,10]. The remarkable accuracy and read coverage depth of whole genome sequencing technology now enables the inference of intra-tumor heterogeneity and cancer evolutionary history that is encrypted in its mutational profile [2,9,11–14]. The study of the evolutionary history of a cancer provides insight into which mutations are cancer-driving from the numerous passenger mutations and sheds light on the mechanism of tumorigenesis.
Rhabdomyosarcoma (RMS) is the most common soft-tissue sarcoma of childhood. Despite a growing understanding of the molecular mechanism underlying RMS, the disease continues to have significant mortality and morbidity especially when the tumor is metastatic or recurrent [15–17]. RMS tumors can be subdivided into two major subtypes: PAX fusion positive (PFP) and fusion negative (PFN), characterized by the presence or absence of a oncogenic fusion between the PAX3 or PAX7 and FOXO1 genes [18]. Fusion positive tumors tend to occur in adolescence and are associated with an adverse outcome. Fusion negative tumors typically occur at a younger age and have been associated with significant aneuploidy, loss of heterozygosity (LOH) at chromosome 11p15.5 [19] and mutations of NRAS, KRAS, HRAS [20], PIK3CA, CTNNB1 [21,22] and FGFR4 [22]. We have recently reported that PFP- and PFN-RMS have distinct landscapes of somatic genomic alterations that activate a common molecular pathway [23]. Another effort, using high coverage targeted re-sequencing of two relapsed fusion negative tumors, showed that indeed RMS relapse tumors are derived from a minor subclone discovered in the primary tumor [24].
In this study, we decipher for the first time, the evolutionary history of RMS using high coverage whole-genome sequencing of 44 primary tumors with their matched normal samples. We applied a framework of algorithms that enabled the prediction of the sequential order of mutational events and subclonality using chronological molecular information encoded in somatic mutations and allelic copy number (including copy number changes and allelic imbalance). The major inputs to our method include the variant allele fraction (VAF) of somatic mutations, the variant allele fraction of germline single nucleotide variants in tumor samples and somatic copy number status. We verified the accuracy of mutation detection and VAF estimation by deep sequencing on independent platforms. Our framework first estimates the rate of normal cell contamination and corrects its effect on VAF and copy number. Tumor subclones with different genomic profiles are then identified using the distribution of VAF of somatic mutations as well as the allelic copy number status. Finally, we estimate the temporal progression of the observed somatic events in each tumor genome based on the fact that somatic mutations that occur before and after aneuploidy events will have different values of VAF.
Through our analysis, we discovered that the initiating common lesions of PFN-RMS are the combination of loss of heterozygosity of chromosome 11p15.5 and point mutations in members of the RAS pathway in majority of cases. In a small number of PFN-RMS tumors where no obvious RAS pathway mutations were present, we identified alternative genes that were mutated early in the tumors progression, indicating their potential roles in oncogenesis. In PFP-RMS, we discover that a whole genome duplication event which results in tetraploidy consistently occurs in the middle or late stage of the development of these tumors and that the PAX3-FOXO1 fusion event occurs prior to the whole genome duplication event, making it probable that the fusion is an early event in the evolutionary history of these fusion-positive RMS. Finally, we find that in general RMS tumors are universally composed of a dominant clone, although each tumor contains subclonal populations with a unique mutational profile, which may provide selective advantage to a relapse or metastatic tumor. Because we sequenced only one tumor sample for each patient, a limitation of our study, despite a moderately high coverage of the whole genome sequencing (average 105X), is that we may miss some subclonal mutations and clones especially for subclonal mutations with a VAF<0.1. Nevertheless our findings allow us to propose a developmental model of how this devastating pediatric cancer initiates and evolves prior to presentation.
We used data from whole-genome sequencing (WGS) of 44 primary RMS tumors (19 PAX- fusion-positive, and 25 PFN tumors) with paired blood samples (providing germline status) with an average of 105x coverage per genome base. The accuracy of somatic mutation detection has been estimated as 93% by experiments in which 604 non-silent mutations were verified by whole-exome sequencing and targeted sequencing on independent platforms [23]. To verify the observed VAF of somatic mutations, we re-sequenced the somatic mutations discovered in two RMS samples using multiplex PCR and deep sequencing (1997x coverage) with an orthologous platform. The verification rate of VAF was high by the targeted sequencing (90% accuracy respectively, S1 Fig.). In order to accurately estimate the timing of somatic alterations and dissect intra-tumor subclonality, we first estimated the portion of normal cells contaminating the tumor sample by surveying allelic copy number status and VAF of somatic mutations across the genome (S1 Text). The normal cell contamination rate was generally low in the 44 tumor samples, with a range of 0–33%, a median of 16% and a standard deviation of 9% (S1 Table). The effect of normal cell contamination was numerically corrected for in all subsequent analyses so that the VAF and copy number status are purely for tumor population. The efficacy of normal cell contamination correction is illustrated in S2–S4 Figs., where the observed non-integer allelic copy number was corrected to integers and the observed VAF distribution of somatic mutations was corrected to the expected distribution, e.g. VAF = 0.5 for heterozygous mutations on chromosomes without aneuploidy.
Multiple studies have shown that subclones exist in individuals with cancer which may become a major clone at relapse or progression and this phenomenon has been reported in two fusion negative RMS tumors [24]. To identify those events which may provide resistance to therapy and allow recurrence, we searched for genomic alterations that specifically were confined to the subclonal population. Subclonal copy number alteration events and mutational events were identified using an in-silico method (for which the workflow is shown in S5 Fig.). Fifty point mutations were identified where the variant’s allele frequency was greater than 10% from the expected full clonal allele frequency (S2 Table). Included in this gene list were 7 COSMIC genes including ABL1, BUB1B, CDK12, ERBB2, IGF2, KDR, and SMARCA4. Gene ontology showed enrichment for genes involved in cell adhesion (GO:0007155 p = 0.002), negative regulation DNA repair (GO:0006281, p = 0.05) and cell migration (GO:0030335 p = 0.026). Across the population, several chromosomal events were recurrently found in the subclonal population, including gain of chromosomes, 5, 8, 11, 13, 14, 18 and 19 (S1 Table).
We illustrate our method in more detail using one PAX-fusion negative RMS tumor (RMS2110). Genome wide, this tumor had 3,889 somatic mutations including oncogenic mutation KRAS G13D and FGFR4 V550L [23]. Using the procedure described in the previous section, we estimated that this sample had 12% normal cell contamination (S1 Table). Copy number analysis showed that RMS2110 had large-segment allelic imbalance (including copy neutral LOH) and copy number alteration on eleven chromosomes (Fig. 1A).We employed the combination of somatic mutations, allelic imbalance and copy number alterations to dissect tumor clones that have unique genomic profiles. First, VAF of somatic mutations was determined to find subclonal mutations, based on the fact that a subclonal mutation (present only in a part of the tumor cells) usually has lower VAF than full-clonal mutations (present in all tumor cells) [2]. A typical scenario is on chromosomes without aneuploidy or allelic imbalance, where heterozygous full-clonal mutations have VAF equal to 0.5 while subclonal mutations have VAF<0.5. Analysis of RMS2110 showed that more than half of the chromosomes are without aneuploidy or allelic imbalance (Fig. 1A and 1B). On these chromosomes, a small portion of somatic mutations have observed VAF significantly lower than others indicating the presence of subclones (Fig. 1C, where there are two distinct VAF clusters, one is centered on VAF = 0.5 and the other is centered on VAF = 0.2; for individual chromosomes see Fig. 1D). Using a clustering algorithm with cluster-number-selection procedure (S1 Text) we can identify subclonal mutations as those with lower VAF. Given the depth of sequencing coverage, we estimated that we could detect subclonal mutations with VAF as low as 0.1 (S6 Fig.). Second, allelic copy number status is used to detect subclonal copy number alterations, based on the fact that subclones of different copy number will result in a non-integer allelic copy number for the whole tumor sample. The joint status of total copy number and lesser allele fraction (LAF—the ratio between the less allelic copy number and total copy number, estimated by germline single nucleotide variants, see Methods) reflects whether the allelic copy number is an integer—whether the observed allelic copy number (red dots) is on the expected position (blue crosses) in Fig. 1D. Therefore our approach predicts the subclonal copy number alterations and the fraction of tumor cells possessing these changes (S1 Text).
We thus predict that the subclonal somatic mutations made up only 5% of all the somatic mutations in RMS2110 tumor. Subclonal copy number alterations were detected on chromosomes 5, 7, 8, 12, 13, 17 and 20.
The method was then applied to all the 44 RMS tumors. Looking across all 44 genomes revealed that a dominant clonal lineage was present in each tumor sample (Fig. 2A). The dominant clone carried a large proportion (from 81–96%) of somatic mutations regardless of PAX-fusion status of the tumor. Meanwhile, a small percentage of the cells in each tumor did display evidence of subclonal changes (Fig. 2B). In addition, we observed more subclonal aneuploidy events in PFN tumors (more than half of the tumors have detectable subclones) than PFP tumors (Fig. 2C). The inferred subclonal aneuploidy and mutational events were listed in S1 and S2 Tables, respectively.
The timing of genomic alteration events can be derived using chronological molecular information encoded in the somatic mutations and copy number alterations, because a specific order of genomic event will result in different VAF [2]. To explain the workflow of our method, we again take sample RMS2110 as an example.
The majority of this cancer genome was the expected diploid status with LAF = 0.5, however regions with aneuploidy or allelic imbalance provided the opportunity to identify the timing of genomic events, by comparing the allelic copy number status with the VAF distribution of somatic mutations (S7 Fig.). For example, chromosome 9p and 11 have 2 copies with LAF of 0 (Fig. 1B), which is likely the loss of one allele followed by a duplication of the remaining allele although we cannot formally exclude the possibility of the duplication of both alleles with subsequent loss of the 2 copies of one allele. This observation was confirmed by the fact that most germline single nucleotide variants on chromosome 9p and 11 had VAF near 1 (Fig. 3A). In this case, the somatic mutations occurring before the “LOH+duplication” event must be present on both copies, with an expected VAF of 1, whereas those occurring after the “LOH+duplication” event would be present on only one copy, with an expected VAF of 0.5. The data confirmed this prediction with the VAF displaying a bi-modal distribution with two peaks at 0.5 and 1, respectively (Fig. 3B). The ratio between the numbers of mutations in the two clusters reflects the fraction of “molecular time” it undergoes to accumulate mutations before and after the “LOH+duplication” event, assuming a constant accumulation rate [2]. We acknowledge that the somatic mutation accumulation rate varies among small genomic segments [5,25–27], but for chromosome-level segments used in this study, the average accumulation rates of somatic mutation were observed to be consistent with one another (r2 > 0.98, S8 Fig.). Therefore, the molecular timing inferred for different aneuploidy events were comparable among the segments within the same tumor sample.
This analysis was extended genome-wide and allowed the inference of the "phylogenetic tree" of tumor RMS2110 (Fig. 3C). Somatic mutations occurring before and after chromosome aneuploidy events had distinct VAF values, which allowed inference of the timing of two candidate driver mutations in this tumor, the nonsynonymous mutation of FGFR4 V550L and the nonsynonymous mutation of KRAS codon G12. The mutation of FGFR4 V550L has a VAF = 0.95, indicating that the mutation happened before the uniparental disomy event of chromosome 5, which is estimated to occur at 26% of molecular cancer lifetime (S1 Table). Therefore FGFR4 mutation should happen between 0 to 26% of molecular time. Similarly, the mutation of KRAS codon G12 has a VAF of 1, indicating that it occurred before the copy gain of chromosome 12p, timed at 21% of molecular cancer lifetime. Other early events included trisomy of chromosome 8 which happened at 8% of the molecular cancer lifetime and uniparental disomy of chromosomes 11 and 9p which occurred at 14% of the molecular cancer lifetime. Note that on both copies of chromosome 9p, there is a somatic focal deletion of CDKN2A (chr9:21981721-21952010), which is also verified by RNAseq (FPKM = 0).
Other aneuploidy events in this sample included: trisomy of chromosomes 12 and 19 which occurred at 21% of molecular cancer lifetime; uniparental disomy of chromosomes 5, 7 and 17 which happened at 26% of molecular cancer lifetime; tetrasomy of a segment of chromosome 3q which occurred around 50% of molecular cancer lifetime; trisomy of chromosome 13 which occurred at 77% of molecular cancer lifetime; hexasomy of 12p which happened at 85% of molecular cancer lifetime; and trisomy of chromosome 20 which occurred near the time of tumor presentation (Fig. 3C).
This method was applied to all the 44 tumor-normal sample pairs to build the evolutionary history of RMS (Fig. 4, S9 Fig., and S1 Table). In summary, our results showed three major findings. First, LOH of 11p15.5 was a consistent early founding event (in average occurred at 35% of molecular cancer lifetime) in PFN-RMS. In comparison, other highly recurrent aneuploidy events such as the copy gain of chromosome 8 and 2 were not consistent early occurring events (time ranges from 1%∼95% and 16%∼96%, respectively). Second, mutations in RAS pathway genes, including FGFR4, KRAS, NRAS and HRAS, were recurrent early events in PFN-RMS. In addition, mutations in other genes (PKN1, CCND1, CUL2, and TTK) occurred early suggesting their role in tumorigenesis. Third, PFP-RMS tumors in general had much fewer somatic alterations and few of them occurred early in the tumor’s molecular lifetime. Of note, a whole-genome duplication event consistently occurred at the middle or late point in the molecular lifetime of these tumors. The high recurrence suggests that this event might be crucial for the presentation of this cancer subtype. We will discuss these findings in more details in the following sections.
The LOH of 11p15.5 is a critical event in fusion-negative tumors [19,28] and frequently results from uniparental disomy and trisomy. In our study, the LOH of 11p15.5 occurred in a total of 26 RMS tumors. As previously described, there was a distinct enrichment for the event in the PFN-RMS population (24/25 PFN samples). Of note, the remaining PFN sample had a small deletion event (3 bps) in the 3-prime non-coding region of IGF2 of undetermined significance. In contrast, only 2 out of 19 PFP-RMS samples had 11p15.5 LOH. The observed LOH usually was accompanied by a copy gain (23 out of 24 PFN samples with LOH on 11p15.5) of the remaining allele resulting in uniparental disomy (n = 8 in PFN and n = 1 in PFP), trisomy (n = 4 in PFN), tetrasomy (n = 8 in PFN and n = 1 in PFP), or pentasomy (n = 3 PFN) (S10 Fig.).
Applying our evolution-history-inference method across the RMS population revealed that the 11p15.5 LOH event universally occurred at an early time point in these tumors development (S11 Fig.). In these tumors, 11p15.5 had a LAF of 0 (S12 Fig. (A)-(C)), meaning that most germline single nucleotide variants present in the tumor, were homozygous. In contrast, the somatic mutations can be homozygous (before 11p15.5 copy gain) or not (after 11p15.5 copy gain). For instance, on chromosome 11 (uniparental disomy) of sample RMS2110, the majority of somatic mutations distributed around a VAF of 0.5 with a minority distributed around VAF of 1 (Fig. 3B). The cluster of somatic mutations discovered around a VAF of 0.5 indicated that these mutations occurred after chromosomal duplication. Comparison of the relative size of the mutations at VAF of 0.5 to those at VAF of 1 indicated that the duplication of chromosome 11 happened early in the cancer evolutionary history. On average, the “LOH+duplication” event of 11p15.5 occurred at 35% of molecular time of the 23 PFN-RMS tumors that had this event. The early occurrence of 11p15.5 “LOH+duplication” event was especially prominent in the patients diagnosed at a later age—the mean occurrence molecular time is 29% of lifetime and the variance is 10% in ≥3 year old patients (S1 Table).
In contrast to the high recurrence (92%) of 11p15.5 LOH in PFN tumors, 11p15.5 LOH was found in only 11% of our PFP tumors. Moreover, in contrast to the consistent early occurrence in PFN tumors, the LOH event did not consistently happen at early lifetime of PFP tumors—37% and 64% of lifetime for two PFP tumors, respectively.
Beyond 11p15.5, the LOH of chromosome 17 (6 out of 25 PFN-RMS) was identified as a recurrent alteration event, typically occurring at an early time point in the evolutionary history of RMS. The LOH region on chromosome 17 includes two small genomic regions that encompass the TP53 and NF1 locus, respectively (S10 Fig.). The observed lesions occurred at an average molecular timing of 38% of cancer evolutionary history. The LOH of 9p, was present in 3 out of 25 PFN-RMS and was found to occur at an average 41% of cancer evolutionary history. Additional accumulation of aneuploidy, such as gain of chromosome 2, 8 and 13, frequently followed the founding events and had no consistent timing pattern.
Unlike PFN-RMS tumors which have significant chromosomal rearrangements, PFP tumors typically have an LAF at the expected 0.5 and a consistent copy number status across the genome. The VAF distribution of somatic mutations revealed that many samples have genome-wide tetraploidy as evidenced by a cluster of somatic mutations with VAF around 0.25 (Fig. 2B, PFP tumors). In total, we found similar VAF distributions in 17 out of 19 PFP tumors. While it is possible that there were two distinct subclones each occupying 50% of tumor cells, given that the cluster with VAF around 0.25 is consistently found across multiple tumors, tetraploidy is a much more likely assumption. Interestingly, when we analyzed these 17 PFP cancer genomes chromosome by chromosome, we found that the inferred tetraploidy occurred in one apparent event (S13 Fig.). This event typically occurred around the mid-point of the molecular lifetime of the tumor (62%±16%) (S1 Table).
In order to identify the sequential order of the PAX3-FOXO1 fusion event and the whole genome duplication event we interrogated the number of the sequencing reads across the PAX3-FOXO1 junction and found that the junction is duplicated. In contrary to PAX7-FOXO1, PAX3-FOXO1 is not known to be focally amplified [29], thus the observed duplication of PAX3-FOXO1 is likely due to the whole genome duplication. Our analysis demonstrated that the PAX3-FOXO1 fusion was consistently duplicated in the whole genome duplication and thus was likely to have occurred prior to the whole genome duplication event (S4B Fig., S3 Table).
In an effort to obtain a comprehensive picture of the evolutionary history of the RMS tumor genomes, we applied our method to estimate the timing of all the aneuploidy events and mutational events that occurred prior to each tumor’s presentation (Fig. 4). In addition to the recurrent and early 11p LOH event in fusion negative tumors, mutation of FGFR4, KRAS, NRAS and HRAS frequently occurred at an early time point in the cancers evolutionary history (S1 Table). Interestingly, many of the early-mutated genes belong to the RAS pathway (Fig. 4). In total 13/25 fusion negative samples carried the combination of early loss of LOH of 11p15.5 and the mutation of RAS, NF1 or FGFR4 which form the founding events in the evolutionary history of these tumors.
Timing and expression analysis of all the observed somatic mutations was used to discover potential driver mutations that occurred early during tumor development (S4 Table). These mutations included recurrent alteration in two PFN RMS of PKN1 (<30% and <42% of molecular lifetime) which encodes a kinase belonging to the protein kinase C superfamily, CCDN1 at <33% of molecular lifetime in RMS2032, CUL2 at <33% in RMS2034, PAX2 at <77% in RMS 2117, TTK at 17–78% in RMS202.
Recurrent and early mutation of PKN1 in two of the evaluated tumors led us to hypothesize that these changes were involved in myogenic differentiation. To test our hypothesis and assess the potential of our method in identifying driver mutations, we conducted a cell line study described in this section. PKN1 is a member of the AGC-subfamily of serine/threonine kinases. The protein product of the PKN1 gene is composed of a C-terminal kinase domain with significant homology to that of the protein kinase C isoforms, but a unique auto-inhibitory N-terminus made up of 3 homologous stretches of anti-parallel coiled-coil folds (ACC1–3) known to bind to Rho-family GTPases in a nucleotide-dependent manner followed by a C2-like region, known to bind phospholipids and fatty acids. Rho-GTPase binding to the ACC regions causes a conformational change that allows PKN1 to be phosphorylated and activated by PDK1. Active PKN1 plays a role in diverse cellular processes such as regulation of the actin cytoskeleton, cell adhesion, vesicular transport and glucose metabolism [30]. PKN1 represses WNT/CTNNB1 signaling [31] and stimulates the ATF2 and MEF2A transcription factors via a signaling pathway that involves MAP2K3/MAP2K6 and MAPK12 [32]. The observed mutations in RMS202 (E216K) and RMS2035 (A298T) occurred prior to 30% and 42% of molecular lifetime of the tumors, respectively. Both mutations occur in the region of the third ACC domain and could potentially interfere with regulation of the kinase activity of PKN1. To test the functional consequences of the PKN1 E216K mutation, wild type and mutant PKN1 viral constructs were made and transduced into the mouse skeletal muscle precursor cell line C2C12. Using a differentiation assay described by [33], defects in terminal differentiation, reflected in expression of myosin heavy chain (MHC) were observed with the mutated version of PKN1, (Fig. 5A-C), suggesting that the mutant PKN1 prevented the C2C12 cell differentiation. C2C12 cells expressing wild type PKN1 could be induced to express MHC although cell fusion, as determined by the number of nuclei per MHC positive cell, was significantly inhibited. Since PKN1 is known to regulate the activity of several transcription factors known to play a role in myogenic differentiation, we performed expression analysis of the constructed cell lines to determined differentially expressed genes among the constructs. Interestingly, gene set enrichment analysis showed that YAP1 target genes were induced in myoblasts and skeletal muscle genes were repressed in myotubes (Fig. 5D) when the PKN1 mutation was present.
While a growing number of studies have used next generation sequencing to examine the evolutionary development of adult tumors [2,9,13], this study is the first application of whole-genome sequencing to the study of a pediatric solid tumor’s subclonality and evolutionary history. The implication from adult studies is that progression of a tumor cell follows a long course, with subsequent lineages of cells acquiring mutations until a particular alteration allows expansion of one clone; ultimately leading to clinical detection. Studies of multiple cancer types including melanoma [34] and colon [35] have shown that even potent oncogenic mutations may be relatively quiescent and thus remain in an undetected premalignant state until released by an additional genetic hit. The patients in our study had a median age at diagnosis of 6.5 years, indicating that the tumors in these children were developed over a relatively short time course, perhaps hinting at the relative potency of the observed somatic changes. Despite a shorter window of development, our study shows that the basic pattern of sequential mutation accumulation is maintained.
Given that similar analyses of adult tumors find the presence of a dominant clone [2], we anticipated that RMS tumors might have a similar finding. Our results show that a dominant clone accounts for more than 80% of cancer cells in a typical tumor sample and genome wide carries many hundreds or even thousands of accumulated point mutations. Even with our genome-wide sequencing and deep coverage, the subclones are frequently observed only at the limits of detection. Admittedly, our study may miss part of the subclonal mutations due to the limit of coverage and the complex nature of genomic subclonality, especially for subclonal mutations with a VAF<0.1. Nevertheless GO analysis of subclonal mutations showed enrichment for potential resistance and progression pathways including cell adhesion, DNA repair and cell migration. While our DNA extraction included relatively large tumor sections, it is also possible that further sampling of the tumor may yield additional subclonal populations as have been appreciated in the study of distant metastases [10]. Even with these constraints, we can conclude that one clone typically dominates these samples, and the majority of tumor cells share most of the detectable genomic alterations. In addition, our findings indicate that a significant proportion of the clonal heterogeneity in RMS is found in gain or loss of copy number, mirroring similar findings in a study of pediatric acute lymphoblastic leukemia [36]. Future efforts will be needed to determine if these alterations might play a role in defining a "cancer stem cell" which resists the selective pressures of the microenvironment and therapy to survive as a recurrence or metastatic lesion.
From the analyses described here, we can begin to understand the dynamics of RMS development. LOH of 11p15.5 has long been described as recurrent feature of several pediatric conditions including the overgrowth phenotype of Beckwith-Wiedemann Syndrome [37], Wilms tumor [38] and embryonal RMS. The proposed mechanism for oncogenesis of this lesion is loss of imprinting control over the IGF2 locus resulting in over-expression of this developmentally regulated growth factor. In this study, we find that not only is this lesion highly recurrent (>90%), it also appears to be the key early landmark in the evolution of fusion-negative tumors. The discovery of a somatic mutation of IGF2 within a fusion negative sample that does not harbor LOH of 11p15.5 (RMS2037) provides additional support to the role of dysregulation of IGF2 in PFN-RMS. In evolutionary terms, the presence of 11p15.5 LOH defines the “most recent common ancestor” when combined with a mutation in a gene in the RAS pathway (NRAS, KRAS, HRAS, FGFR4). While the progression we describe in this study indicates a possible common sequence of events (S1 Table), in some tumors it is equally as likely that the oncogenic mutation of a RAS pathway gene is the founding lesion. This corroborates the finding that patient’s with Costello Syndrome (HRAS germline mutation) [39], Noonan Syndrome (NRAS, KRAS, PTPN11 germline mutations) [40] and Neurofibromatosis (NF1 germline mutation) [41] all have increased risk of developing fusion-negative RMS. Interestingly, while RMS is certainly described in patients with Beckwith-Wiedemann Syndrome (germline uniparental disomy of 11p15.5), these patients appear to have a higher relative risk of developing Wilms’ tumor and hepatoblastoma than RMS [42]. Regardless of which lesion comes first, the combination of LOH of 11p15.5 with a RAS pathway mutation appears to set a clone on the course towards developing a fusion negative RMS tumor (Fig. 6A).
By examining other mutations that occur early in the molecular history of fusion negative tumors, especially those tumors that had no mutation in a candidate “RAS” pathway member, it is possible to nominate other potential founding lesions. In this analysis, mutation of PKN1 stands out as both recurrent (occurring in tumor RMS202 and RMS2035) and occurring early in the evolutionary history of these tumors (<30% and <42% respectively). PKN1 is a member of the protein kinase C family and has been implicated as a repressor of WNT/CTNNB1 signaling [43], a mediator of insulin signaling to the actin cytoskeleton [44] and an activator of MEF2A dependent transcription [32]. The observed mutations (E216K and A298T) occurred within the third ACC domain and may relieve N-terminal auto-inhibition of PKN1 kinase activity (S14 Fig.). Our functional assessments of the PKN1 mutation demonstrated that the observed mutation can inhibit terminal differentiation of skeletal muscle in a dominant-negative fashion. Gene expression analysis of PKN1 E216K expressing C2C12 cells grown in low serum media reveals that this differentiation block occurs as a result of repression of skeletal muscle specific genes and genes with MEF2A binding sites in the promoter. Expression of MEF2 proteins is induced by myogenin during normal skeletal muscle differentiation, and MEF2 factors, in complex with other transcriptional activators, are known to play a role of in myogenic differentiation [45]. One possible explanation for these results is that PKN1-dependent signaling leads to replacement of MEF2A at skeletal muscle specific genes with a transcriptional repressor. The identity of this transcriptional repressor is not yet known.
GSEA analysis of PKN1 E216K expressing C2C12 myoblasts reveals that the gene expression signature of these cells is enriched for a YAP gene signature. The YAP gene signature is composed of genes activated by YAP overexpression in human mammary cells, mouse liver tissues and immortalized mouse fibroblasts, and thus represents a list of evolutionary YAP target genes [46]. YAP1 has recently been identified as a potent ERMS oncogenic driver, and YAP1-TEAD complexes repress expression of genes important for myogenic differentiation, in part by impinging upon the binding of myogenic transcription factors including MEF2 [47]. Intriguingly, YAP1 is able to functionally compensate for oncogenic KRAS in colorectal and pancreatic cancer mouse models [48,49], which is consistent with the fact that RAS mutation and PKN1 mutation are mutually exclusive in the set of ERMS tumors presented here. The mechanism by which PKN1 leads to increased YAP1 activity is currently unknown.
With whole genome sequencing from a single biopsy, we were unable to reliably estimate the molecular timing of some of the recurrent genetic lesions observed in RMS. These events include mutations in PIK3CA, FBXW7, NF1, and BCOR, and the focal amplification of MDM2, CDK4, and MYCN. Moreover, somatic events such as the copy gain of chromosomes 2, 8 and 13 were estimated to occur in an inconsistent temporal pattern among RMS patients (occurring at multiple time points throughout the evolutionary history). Certainly, the recurrence of these lesions in multiple large sample sets speaks to their importance, however since they can occur either early or late in the cancers evolutionary history they do not represent the initiating lesion in the tumor but rather modifiers of the tumor as it grows towards presentation.
Early flow cytometry studies to evaluate DNA content [50] and cytogenetic analysis [51] in ARMS noted the frequent presence of tetraploidy. We find evidence that fusion positive tumors demonstrate a very high recurrence of a whole genome duplication event that results in a tetraploid genome. In general, the duplication event occurs around the midpoint of the tumor’s evolutionary history and prior to additional accumulation of aneuploidy (Fig. 6B). The mechanism of a cell moving from a diploid state to a tetraploid state can be due to a cytokinesis failure, a cell fusion event or mitotic slippage and it has long been speculated that tetraploidy is an intermediate in cancer progression [52]. This is born out in at least one premalignant state (Barrett’s esophagus), where tetraploidy represents an early intermediate as the cancer precursor develops towards esophageal adenocarcinoma [53]. Further experimental evidence of a tetraploidy intermediate state is demonstrated in TP53 null mouse mammary cells where tetraploid but not diploid cells promote tumorigenesis in nude mice [54]. Of note, in the mouse model of PFP ARMS both the increase in allelic copy number of the PAX3-FOXO1 fusion and the loss of TP53 or CDKN2A were found to be a critical component in increasing the penetrance of the phenotype [55]. Given the role of loss of TP53 in allowing a permissive environment for the tetraploid cell to escape cell cycle arrest, this is consistent with our findings. It is interesting to speculate that in our tumor series the whole genome duplication event is an attempt by the tumor to increase the allelic dose of the PAX fusion. Another intriguing possibility is that the whole genome duplication event alters the telomere length and activates telomerase; a critical cooperating genetic event in the temporal sequence that produces alveolar rhabdomyosarcoma from human myoblasts [56].
In conclusion, based on deep whole genome sequencing, we developed a systematic method to infer the evolutionary history and identify the causative lesions of this pediatric solid tumor. From our data, we provide a model for how these tumors develop. Our results demonstrate two distinct evolutionary paths resulting in a convergent phenotype of this soft tissue cancer: 1) genomic loss of heterozygosity on 11p15.5, nonsynonymous mutations on RAS pathway and cell cycle genes, including FGFR4, KRAS, NRAS, HRAS and CCDN1, as well as several other genes, including CUL2, TTK and PKN1, occur early in the evolution history of PFN-RMS; 2) recurrent whole genome duplication occurs in the middle or late stage of the tumors evolution in the PAX-fusion positive RMS tumors. PAX3-FOXO1 fusion occurs before the whole genome duplication event. Intriguingly, a recent report of the clonal evolution in two cases of relapsed fusion-negative RMS tumors after chemotherapy and radiation demonstrated that recurrent tumors are derived from the founding tumor’s minor clone [24]. Nevertheless given that survival following relapse is 30% [57], our findings that LOH of 11p15.5 and mutation of RAS genes form the “trunk” of the fusion negative RMS tumors has important therapeutic implications; at least in theory it is preferable for treatment to target mutations that are present in all of the tumor cells. In addition, while further biologic validation is required, our analysis identifies candidate lesions that may underpin the ability of the minor clone to survive therapy and proliferate during relapse or metastasis. Thus given clonal evolution and heterogeneity, our data suggests that future efforts to understand the emergence of resistant clones should include deep sequencing analysis on all patients at presentation and relapse. These efforts will add to the growing understanding of the biology of RMS and identify actionable genetic aberrations for targeted therapies.
All patient sample collection was approved by the institutional review board of the participating facility. All de-linked and de-identified patient sample information was collected under the approved institutional review board of the National Cancer Institute protocol 10-C-0086. Samples were assembled from collections at the Pediatric Oncology Branch of the National Cancer Institute, Children’s Oncology Group, and the Tumor Bank at The Children’s Hospital at Westmead (New South Wales, Australia). All tumors were collected at initial diagnosis and prior to any therapy with the exception of samples NCI0040 and NCI0080 which were collected at relapse. Samples were de-identified and histologic diagnosis and clinical information were compiled. Quality control genotyping for the whole genome samples was performed to ensure the match of tumor normal pairs.
Approximately 6 micrograms of DNA was sequenced using the paired end sequencing method of Complete Genomics. Data analysis was accomplished using CGA tools package v2.0 as well as a number of in house tools described in [23].
A custom panel of oligonucleotides designed to incorporate somatic single nucleotide variants discovered in samples RMS2110 and RMS2107 was generated using Ion Ampliseq designer software. 150 base pair amplicon libraries were generated using multiplex PCR according to the Ion Torrent Ampliseq Library 2.0 kit. Individual samples were barcoded and the generated libraries were sequenced using a 318 chip on a Personal Genome Machine (Life Technologies). Compiled reads were mapped to Hg19 and the expected variants were analyzed for coverage and VAF. The targeted sequencing has a depth of 1997x. We used it to verify the VAF estimates of somatic mutations across whole genome of two samples. The result shows VAF estimation accuracy is as high as 90% (S1 Fig.).
Illumina Omni 2.5M (97 paired plus 30 unmatched tumors) or 5M (10 paired samples) were performed according to the standard procedure from the manufacturer (Illumina, San Diego, CA) at the National Cancer, Cancer Genomics Research Laboratory. The data were previously reported in [23].
PolyA selected RNA libraries were prepared for RNA sequencing on Illumina HiSeq2000. 100 bases long paired-end reads were assessed for quality and reads were mapped using CASAVA (Illumina, San Diego, CA). The generated fastq files were analyzed by TopHat2 [58] and Cufflinks [59] The data were previously reported in [23].
We used the somatic score from Complete Genomics (CG), a highly sensitive and specific somatic mutation-calling criteria [23,60,61], combined with principles previously used [25,62] to detect somatic signal nucleotide variants (SNVs). The somatic score is designed by systematically considering sequencing error, mapping error and read count and has been validated as an effective criterion by our previous studies as well as studies from other groups [23,61,63]. To ensure sensitivity and specificity of mutation calling, we set the somatic score cutoff as ≥0, with a set of additional filters for removing system artifacts and mapping errors, to select somatic mutations, based on the verification results from an independent sequencing platform [23]. By SOLiD whole exome sequencing (35x coverage) and Ion Torrent Ampliseq targeted sequencing (300x coverage) on 30 RMS tumor-normal sample pairs sequenced by CG, we verified that our filtering removes 99% of false positives while maintaining 80% of true positives [23] in CG’s comprehensive somatic mutation pool (called in a loose criterion to be as inclusive as possible) [23,60], which is premium given the less somatic mutations in pediatric tumors [3].
For the 44 matched tumor and normal sample pairs with whole genome sequencing, we used somatic copy number segmentation profile provided by Complete Genomics, with customized corrections. For a larger cohort, with 120 matched tumor and normal sample pairs measured by SNP array, we used the copy number profile provided by NEXUS copy number analysis.
In copy number profiling, an important quantity, lesser allele fraction (LAF), is estimated to represent the fraction of copies coming from each parental allele. Along with total copy number, LAF tells us the allelic copy number status. LAF is defined as the ratio between the copy number of the lesser allele (with fewer copies than the other allele) versus the total copy number. For example, a trisomic chromosome with allelic type “AAB” has LAF equal to 1/3. LAF is estimated based on germline heterozygous single nucleotide variants that are present in cancer genome. For each genomic segment of tumor sample, sites that have heterozygous single nucleotide variants in matched normal samples were selected and LAF was estimated by the ratio between the read count of lesser allele and the total read count for these sites.
In order to study the evolutionary history of RMS, we performed an integrated statistical analysis, using information including germline single-nucleotide variants, somatic mutations (single-nucleotide variants), somatic copy number alterations (CNA) and junctions (mapped breaking points), to estimate the normal cell contamination, intra-tumor heterogeneity (due to subclones) and the timing of somatic variants. The methods are detailed in the S1 Text.
Mouse myoblast cell line C2C12 was obtained as a generous gift from Dr. Marc Landanyi. The cells genotype was performed by the NCI Core Genotyping facility and the cell line was confirmed to be myocoplasm negative. All cell culture was performed in DMEM supplemented with 10% FBS. The differentiation assay was performed using 2% horse serum (Life Technologies) as previously described [33]. For retrovirus production, the pBabe vector system (Addgene) was used.
Plasmids encoding human PKN1 cDNA was purchased from Addgene, and constructs were subcloned into pBABE containing a N-terminal HA tag. A cDNA encoding the PKN1 E216K mutant was generated using the GeneART Site-Directed Mutagenesis kit (Life Technologies) and was subcloned into pBabe. The generated mutation was confirmed by Sanger sequencing. Retroviruses were generated by contransfection of pBabe constructs with pCL-10A1 into 293T cells (American Type Culture Collection (ATCC), CRL-3216) subsequently used to infect the C2C12 cell line as previously described; infected cells were selected with 2 μg/ml puromycin (Life Technologies).
Cells grown on Nunc chamber slides were fixed with 4% paraformaldehyde and permeabilized in PBS containing 0.5% Triton X-100 and blocked in Block-Aid (Life Technologies) for 1 h at room temperature. Cells were then incubated with MF20 monoclonal antibody (DSHB) against MHC (1:40 dilution;) overnight at 4 degrees. Secondary antibody Alexa Fluor 488-conjugated secondary antibody (1:200 dilution; Life Technologies) for 1 h at room temperature. Cells were mounted with ProLong Gold antifade reagent with DAPI (4′,6-diamidino-2-phenylindole; Life Technologies).
Cells were lysed in M-PER lysis buffer (Pierce Biotechnology). Lysates were denatured in 4× sample buffer at 70°C for 10 min, resolved on 4–12% NuPAGE gels (Life Technologies) and transferred onto PVDF (polyvinylidene fluoride) membranes. Membranes were blocked in 5% nonfat milk in TBST buffer (TBS with Tween-20) for 1 h at room temperature and probed with primary anti-HA antibody obtained from Covance (1:2,000 dilution). Bound antibodies were detected with peroxidase-labeled horse antibody to mouse IgG and visualized using enhanced chemiluminescence reagents (ThermoScientific).
Total cellular RNA was isolated using the RNAeasy mini kit (Qiagen). Cellular RNA (250ng) was in vitro transcribed, fragmented, hybridized and applied to Affymetrix Mouse 430A arrays according to the standard operating procedure of the Laboratory of Molecular Technology core facility (http://atp.ncifcrf.gov/genetics-and-genomics/laboratory-of-molecular-technology) and the manufacturer’s instructions (Affymetrix, Santa Clara, CA). For gene set enrichment analysis (GSEA), the normalized gene expression data were z-scored and ranked according to absolute fold-change expression over the control. GSEA analysis (http://www.broadinstitute.org/gsea/index.jsp) was performed using default parameter settings.
|
10.1371/journal.pntd.0004250 | Molecular Differentiation of the African Yellow Fever Vector Aedes bromeliae (Diptera: Culicidae) from Its Sympatric Non-vector Sister Species, Aedes lilii | Yellow fever continues to be a problem in sub-Saharan Africa with repeated epidemics occurring. The mosquito Aedes bromeliae is a major vector of yellow fever, but it cannot be readily differentiated from its non-vector zoophilic sister species Ae. lilii using morphological characters. Genetic differences have been reported between anthropophilic Ae. bromeliae and zoophilic Ae. lilii and between forest and domestic populations. However, due to the application of different molecular markers and non-overlapping populations employed in previous studies, interpretation of species delimitation is unclear.
DNA sequences were generated from specimens of Ae. simpsoni s.l. from the Republic of Benin, Tanzania and Uganda for two nuclear genes apolipophorin 2 (apoLp2) and cytochrome p450 (CYPJ92), the ribosomal internal transcribed spacer region (ITS) and the mitochondrial cytochrome c oxidase (COI) barcoding region. Nuclear genes apoLp2 and CYPJ92 were unable to differentiate between species Ae. bromeliae and Ae. lilii due to ancestral lineage sorting, while ITS sequence data provided clear topological separation on a phylogeny. The standard COI barcoding region was shown to be subject to species introgression and unable to clearly distinguish the two taxa. Here we present a reliable direct PCR-based method for differentiation of the vector species Ae. bromeliae from its isomorphic, sympatric and non-biomedically important sister taxon, Ae. lilii, based on the ITS region. Using molecular species verification, we describe novel immature habitats for Ae. lilii and report both sympatric and allopatric populations. Whereas only Ae. lilii is found in the Republic of Benin and only Ae. bromeliae in Tanzania, both species are sympatric in Uganda.
Our accurate identification method will allow informed distribution and detailed ecological studies that will facilitate assessment of arboviral disease risk and development of future targeted vector control.
| In Africa, epidemic outbreaks of yellow fever continue despite the availability of an effective vaccine. Effective understanding of disease epidemiology and control requires the ability to reliably identify vectors of yellow fever. The mosquito Ae. bromeliae, a competent vector of yellow fever virus, cannot be reliably morphologically differentiated from its sister species Ae. lilii, which does not bite humans and so does not transmit yellow fever. DNA sequencing of four molecular markers allowed comparisons of how they perform at distinguishing these species. We found that the mitochondrial cytochrome c oxidase (COI) barcoding region and nuclear apolipophorin 2 (apoLp2) and cytochrome p450 (CYPJ92) were unable to reliably distinguish these species. Conversely, genetic variation at the internal transcribed spacer region (ITS) was able to confirm the vector Ae. bromeliae and non-vector Ae. lilii as distinct species. Based on ITS sequence differences, we developed a robust molecular method to identify the vector Ae. bromeliae from its sister species Ae. lilii. Consequently, we find that these species use the same larval habitats including banana, cocoyam and Dracena spp. in Uganda. Whereas only Ae. lilii appears to be present in Benin and only Ae. bromeliae in Tanzania, we confirm that both species occur in Uganda. Reliable species designation will promote more detailed studies of distribution, ecology and vector status essential for disease risk assessment and mosquito control.
| Correctly identifying the vector species involved in mosquito-borne disease transmission is fundamental to predicting disease outbreaks, ascertaining general risk to the human population and targeting control efforts. Despite this necessity, the reliable identification of mosquitoes is problematic in many cases, including the medically important Aedes simpsoni complex. This complex comprises three known species including Ae. simpsoni, Ae. lilii and Ae. bromeliae. Among these three species, Ae. bromeliae is an important vector of yellow fever virus (YFV) and potentially other arboviruses [1]. Yellow fever has increased in incidence as a result of urbanisation and changes to public health policy [2, 3]. The burden of yellow fever in Africa is estimated to be 130,000 cases a year, 85,000 of which result in deaths, despite the availability of a vaccine [4]. After mass immunisation campaigns at the beginning of the 20th Century, YFV was successfully reduced in targeted countries. However, YFV outbreaks are causing renewed attention. In West Africa, 13 out of 14 countries known to host YFV now report cases regularly and have experienced epidemics since 2000 [5, 6]. Further concern has arisen over identification of a novel YFV genotype implicated in recent outbreaks including the first reported outbreak in Kenya, East Africa (1992–1993) [7]. Related genotypes have also been reported in Sudan in 2003 and 2005 and more recently in northern Uganda in 2010 [8–10]. Although much attention has been given to understanding yellow fever disease epidemiology [11], relatively little is known about the mosquito vectors.
Ten species are currently known within the wider Simpsoni Group (Huang 2004). Within this group, Theobald [12–14] originally described three species belonging to the Simpsoni Complex (Ae. simpsoni, Ae. bromeliae and Ae. lilii). Despite these separate species designations, many entomologists referred to all simpsoni-like mosquitoes within this complex as the nominotypical species, Ae. simpsoni [15–21]. In Uganda, it was found that some populations of Ae.simpsoni s.l. were anthropophilic and attracted to human bait, whereas others were not attracted to man despite local abundance [20]. Furthermore, mosquitoes collected from zoophilic and anthropophilic populations showed different feeding preferences in the laboratory with a preference for rodents or humans, respectively [20]. These findings led to a re-examination of mosquito morphology and subsequently Huang [22, 23] provided a full description of the component members of the Simpsoni Complex, reviewed their ecology and drew attention to the incorrect use of former nomenclature. Aedes simpsoni, which has only been reported from South Africa and Swaziland, is not implicated in human disease transmission [22, 23]. The anthropophilic yellow fever vector Ae. bromeliae is widespread on the African continent [23]. In contrast, Ae. lilii, only previously reported from Sudan, Ethiopia and Uganda, has never been reported biting man and is thus not considered to be involved in disease transmission [22, 23]. Although we know these three species have different distributions, biting behaviour and vectorial abilities, these need to be fully characterised. Understanding ecology and epidemiology requires that species can be distinguished from one another. However, controversy over mosquito taxonomy means that the anthropophilic disease transmitting Ae. bromeliae cannot be reliably distinguished from its zoophilic sister species Ae. lilii [24, 25].
Large scale detailed studies of the Simpsoni Complex by Huang [22, 23, 26], reported that Ae. simpsoni s.s. can be distinguished from Ae. bromeliae and Ae. lilii in that it has simple claws on the mid tarsi, as opposed to toothed mid-tarsal claws present in the latter two taxa. Aedes simpsoni can also be easily distinguished from conspecifics by its distinct tarsomere scaling pattern [22, 25]. However, Jupp & Kemp [24] reported variation in tarsal claw morphology of Ae. simpsoni and Ae. bromeliae, and questioned the reliability of this diagnostic character. In this study, we focus on the more widespread taxa of the complex, Ae. bromeliae and Ae. lilii. These sister taxa can sometimes be reliably identified morphologically based on tarsomere banding patterns, but only when this character is exhibited at the extremes of its range as banding patterns overlap between the species [25]. Lutwama & Mukwaya [25] questioned the usefulness of tarsomere banding patterns as a diagnostic character for Ae. bromeliae and Ae. lilii. They observed variation in scale ornamentation on an almost continuous scale and found that progeny from the same mother could be identified both as Ae. bromeliae or Ae. lilii based on this morphological character.
The unreliability and practicality of using morphological based methods for routine field-based identification in the Simpsoni Complex has led to attempts to delimit species boundaries through molecular methods. In a study of the Simpsoni Complex, excluding the southerly distributed Ae. simpsoni, Mukwaya et al. [27] found that anthropophilic populations from Kenya and Uganda form a distinct genetic clade separate from non-sympatric and non-anthropophilic populations from Uganda and Nigeria based on the non-coding internal transcribed spacer (ITS) region of ribosomal DNA. In that study, species were designated according to both blood feeding preference (because Ae. lilii is zoophilic whereas Ae. bromeliae are human vectors) and on tarsomere banding patterns which were distinctive for some specimens [22, 23, 27]. Recently, Walter et al. [28] inferred the presence of Ae. bromeliae and Ae. lilii in sympatry in Rabai, Kenya based upon genetic differences between domestic/peri-domestic and forest populations. However, how these putative species relate to those characterised by Mukwaya et al. [27] remains unclear as they used different molecular markers; two nuclear genes, apolipophorin 2 and cytochrome p450 (CYPJ92).
A 658 bp region of the cytochrome c oxidase (COI) gene has been widely adopted as a DNA barcoding standard for species identification because it shows high utility in discriminating between closely related taxa as well as resolving phylogeographic groups within species [29–31]. However, sole use of mitochondrial DNA to delimit species has been questioned because of the potential for pseudogene development and introgression, which may limit the ability of mtDNA markers to resolve closely related species [32, 33]. Combined analysis of both mitochondrial and nuclear genes can improve phylogenetic resolution since these markers evolve at different rates and so target different levels of the phylogenetic tree [34]. Another barcoding candidate is the nuclear internal transcribed spacer (ITS) regions of the ribosomal gene cistron that comprises the 18S, 5.8S, and 28S genes, an external spacer region and two internal spacer regions ITS1 and ITS2 [35]. The ITS regions evolve at a rapid rate in the absence of functional constraint [36, 37]. Because they are tandemly repeated in the genome, the ITS spacer regions are also subject to concerted evolution whereby paralogues are homogenised by genetic exchange [36, 37]. Consequently, paralogues remain genetically similar within species while showing high levels of interspecific divergence [37]. Although not useful in all taxonomic groups, ITS can differentiate between sister species in a large number of cases and has been widely used to delimit closely related mosquitoes in Anopheles complexes [35, 38–42].
We seek to expand previous work on genetic differentiation of Ae. bromeliae and Ae. lilii in the Simpsoni Complex by determining how the putative species identified by Mukwaya et al. [27] and Walter et al. [28] relate to one another. We achieve this by sequencing the same mosquito samples at previously used molecular markers (apoLp2, CYPJ92 and ITS). In addition, we were able to evaluate the utility of various molecular markers, including the ITS and COI regions, in determining species bounds. We apply these findings to develop a molecular identification method based on variation in the ITS region to distinguish the disease vector Ae. bromeliae from its non-vector sister species Ae. lilii. Applying this method to mosquitoes from Benin, Uganda and Tanzania enabled us to generate reliable findings of their ecology and distribution.
Mosquitoes of the Simpsoni Complex were collected as larvae from natural breeding sites in Tanzania (n = 36), Uganda (n = 50) and the Republic of Benin (n = 24) from locations detailed in Fig 1 and S1 Table from 2009 to 2014. Immature habitats sampled included the leaf axils of Musa spp. (banana), Colocasia spp. (cocoyam/taro), Dracaena spp. and in tree holes. To avoid biasing the dataset with siblings, each discrete habitat was treated as a separate collection and only one individual per collection was taken for genetic analyses. Where possible, immatures were reared though to adults; otherwise, larvae destined for DNA analysis were stored in 95% ethanol. Adults were desiccated with silica for optimal DNA preservation and either stored in BEEM capsules or pinned. Larvae were preserved in ethanol for later extraction of DNA. All adult mosquitoes were identified as belonging to the Simpsoni Complex using the morphological identification key in Huang [26], and a subsample of these, and outgroup taxa including Ae. aegypti and Ae. aegypti formosus, were morphologically verified by Dr. Yiau-Min Huang.
DNA was extracted from whole larvae or a single leg of an adult mosquito using the modified phenol-chloroform method in Surendran et al. [44]. Forty individuals of the Simpsoni Complex were amplified and sequenced for a region of the mitochondrial COI gene with universal primers LCO1490 (5’GGTCAACAAATCATAAAGATATTGG’3) and HCO2198 (5’TAAACTTCAGGGTGACCAAAAAATCA’3) [45] using a protocol recommended by the Consortium for the Barcode of Life (http://barcoding.si.edu/dnabarcoding.htm). Peridomestic Aedes aegypti collected from an artificial container in Tanzania was also sequenced at COI as an outgroup.
Internal transcribed spacer regions 1 and 2 were amplified from six individuals from Uganda and three individuals from Tanzania that were selected to represent the genetic diversity we observed at the COI gene. This was achieved with the 18SFHIN and CP16 primers (5’-GTAAGCTTCCTTTGTACACACCGCCCGT-3’ and 5’-GCGGGTACCATGCTTAAATTTAGGGGGTA-3’, respectively) [46], as used by Mukwaya et al. [27]. To generate PCR products, 1 unit of high fidelity MyFi DNA Polymerase (Bioline, UK), 2X MyFi Reaction Buffer, 0.8 μM forward and reverse primer and 1–10 ng of template DNA were used under the following conditions at 30% ramp speed; 95°C for 3 min followed by 30 cycles of 95°C for 30 sec, 58°C for 45 sec and 72°C for 45 sec with no final extension.
Thirteen Individuals, including eight of the same individuals sequenced at the ITS region, were also sequenced for regions of the nuclear genes apolipophorin 2 (apoLp2) and cytochrome p450 (CYPJ92) first described by Brown et al. [47] and used by Walter et al. [28] in the Simpsoni Complex. This included verified Aedes aegypti formosus collected from a tree hole in Tanzania, for use as an outgroup as was the case in Walter et al. [28]. Another marker, short-chain dehydrogenase-reductase (SDR) also used by Walter et al. [28] in the Simpsoni Complex was not used because it produced multiple nonspecific bands on amplification. PCR products were generated as in Walter et al. [28]. PCR products were purified with the GenElute PCR clean up kit (Sigma-Aldrich, UK) and Sanger sequenced in forward and reverse directions using the amplification primers. Sequences were generated with BigDye Terminator v3.1 cycle sequencing kit (Applied BioSystems, UK) on an Applied BioSystems 3730 automated sequencer.
ITS products were cloned using the P-GEM cloning kit (Promega, UK) as per instructions. Transformants were blue/white screened and colonies with inserts stored in TE buffer for PCR amplification. Universal M13 primers (5’-TGTAAAACGACGGCCAGT-3’ and 5’-CAGGAAACAGCTATGAC-3’) [48] were used to amplify cloned ITS products in the following 13μl reaction; 1.25 units of BIOTAQ DNA Polymerase (BioLine, UK), 1 X NH4 Reaction Buffer, 2mM MgCl2 solution, 0.8mM dNTP and 0.5 μM forward and reverse primer. Thermocycler conditions were 95°C for 2 min followed by 95°C for 15 sec, 60°C for 30 sec and 72°C for 30 sec for 35 cycles and a final extension of 72°C for 10 min. A minimum of two and a maximum of four clones were forward sequenced for each individual.
Cloned ITS sequences were aligned with Mukwaya et al.’s [27] from several locations in Africa (S1 Table) and Le Goff et al.’s [43] sequence data from two Ae. bromeliae originating from the Indian Ocean island of Mayotte available from GenBank (KF135509-10) using the program Geneious v5.4.7 [49]. Primers were designed based on this alignment to discriminate between species. Several putative species-specific primers were trialled under a wide range of PCR conditions. However, the only primers that worked effectively generated species-specific PCR products that differed by ~30 b.p. Due to this small size difference we recommend that the primer pairs specific for each species are run in separate PCR reactions. Non-specific banding was a feature of all primers tested which we suspect may result from variable indel length among ITS copies within an individual. Although PCR protocols were optimised to reduce non-specific banding, some extra banding can be visible but does not obscure amplification/non-amplification of the species-specific PCR product.
The developed primers are nested within the ITS1-2 region; therefore PCR products encompassing this region were first generated with the 18SFHIN and CP16 primers as described above. These PCR products were purified with the GenElute PCR clean up kit (Sigma, UK) and 0.5 μl was used as template in a 25 μl reaction with 0.6 units of Go Taq Hot Start polymerase (Promega, UK), 1 X NH4 reaction buffer, 1 mM MgCl2 solution, 0.8 mM dNTP and 0.5 μM forward and reverse primer. Primers developed for amplification of 591 bp in Ae. bromeliae only were BRO-F (5’-CCTGGCCAGTGGCCA-3’) and BRO-R (5’-GTGCACACCACTGA-3’). Amplification was achieved with a touchdown PCR protocol; initialisation step of 95°C for 3 minutes followed by 95°C for 30 seconds, 82°C for 45 seconds and 72°C for 1 minute for 5 cycles, followed by 30 cycles of 95°C for 30 seconds, 64°C for 45 seconds and 72°C for 1 minute and then a final extension of 72°C for 7 minutes. Primers developed for amplification of a 620 bp region in Ae. lilii only were LIL-F (5’CTGATGCACTGGCCTCAAAG’3) and LIL-R (5’TCAACCGCCGTGCGTG’3). The thermocycling conditions were 95°C for 3 minutes followed by 95°C for 30 seconds, 78°C for 45 seconds, 72°C for 1 minute for 10 cycles followed by 95°C for 30 seconds, 70°C for 45 seconds, 72°C for 1 minute for 20 cycles and a final extension step at 72°C for 7 minutes. Amplified products were run on a 1.2% agarose electrophoresis gel to determine the presence or absence of DNA bands of the expected size. Positive and negative species controls were used in all PCR reactions. Sequences are available on GenBank (KT998333- KT998452).
Sequence alignment was achieved with Geneious v5.4.7 [49]. For the apoLp2 and CYPJ92 datasets, files were prepared using seqPHASE [50] and PHASE v2.1 was then used to infer haplotypes [51]. The haplotypes of Ae. aegypti formosus outgroups were determined through alignment with the relevant datasets for Ae. aegypti in Brown et al. [47] and using the program PHASE. Sequences for two Ae. aegypti from Brown et al. [47] were used as outgroups in conjunction with data generated during the present study. The COI dataset was aligned with two COI sequences of Ae. bromeliae from the Indian Ocean island of Mayotte available from GenBank (KF135496-97) [43].
Neighbour joining (NJ) trees were constructed in MEGA 6 [52]as in Mukwaya et al.[27] and Walter et al. [28] using the best available substitution model as chosen by JModelTest [53, 54]. The Tamura-Nei model with uniform rates among sites was used to construct NJ trees for COI. The Kimura 2 parameter model (K80) with uniform rates among sites was used for ITS sequences and K80 with 0.8 gamma-distributed sites was used to construct trees for apoLp2 and CYPJ92. For all genetic markers, missing data including indels were excluded from analysis. Topological support was determined through 1000 bootstrap replications. The ITS sequence tree was constructed without an outgroup because the high level of divergence between the outgroup and the ingroups presents a challenge for sequence alignment. The two ITS sequences of Ae. bromeliae from Mayotte [43] were not included in the NJ tree because they did not overlap with the sequences generated here.
A hierarchical AMOVA was performed on the ITS sequence data generated in this study together with Mukwaya et al.’s sequences [27] (n = 69) in Arlequin v3.5 [55]. A hierarchical AMOVA was also performed on the apoLp2 (n = 13) and CYPJ92 (n = 13) sequences generated in this study.
Four fixed point substitutions and eight indels of varying length were observed between species in the first 300 bp of the ITS sequence alignment. The neighbour joining tree of ITS sequences revealed two major clades with a bootstrap support of 98% (Fig 2). As reported previously, the 46 sequences from Mukwaya et al. [27] clustered into one or other of the two clades depending on their inferred host feeding preference and which they accordingly designated as Ae. bromeliae (anthropophilic) and Ae. lilii (non-anthropophilic). Sequences of all nine Simpsoni Complex individuals from this study also belonged to one or other of these clades, and we therefore identified them as Ae. bromeliae or Ae. lilii with reference to Mukwaya et al.’s [27] ITS-based species designation.
Mosquitoes collected from Kanyawara, Bundibugyo (Bwamba region) and Najjembe in Uganda represented both ITS clades. In comparison, individuals collected from Tanzania strictly aggregated into the ITS lineage of Ae. bromeliae. AMOVA revealed that almost all sequence differences can be explained by variation between species (92.65%), while there is little variation between populations within groups and within populations (0.25% and 7.19%, respectively).
Sequencing of individuals at both ITS and nuclear genes apoLp2 and CYPJ92 allowed us to compare the ITS-based species designation of Mukwaya et al. [27] with that of Walter et al. [28]. Two well-supported lineages were observed in NJ trees for the nuclear genes apoLp2 and CYPJ92 with 100% bootstrap support (Figs 3 and 4). Individuals confirmed as Ae. bromeliae or Ae. lilii using the ITS region with reference to Mukwaya et al. [27], all fall into a single clade for each of the nuclear genes, apoLp2 and CYPJ92. For both gene trees, the other lineage comprises only the sequences from the forest species in Walter et al. [28] that they referred to as Ae. lilii. A hierarchical AMOVA revealed that nuclear genes apoLp2 and CYPJ92 could not effectively distinguish between Ae. bromeliae and Ae. lilii with only 0.86% and 18.94% of genetic variance between species groups, respectively. Conversely, there was relatively high genetic variation among populations within species groups (37.04%, 10.28%, respectively) and within populations (62.10%, 70.79%, respectively).
There are two phylogenetic clusters with high bootstrap support (90%, 90%) and two with weak bootstrap support (56% and 43%) in the NJ tree for mitochondrial COI with individuals tending to cluster according to geographic origin (Fig 5). The sequences from Mayotte cluster with Ae. bromeliae from Tanzania. Individuals designated as Ae. bromeliae or Ae. lilii according to ITS sequence variation tend to group according to species within the same clades, but there are three exceptions. One individual from Uganda, exhibiting the ITS sequence of Ae. bromeliae clusters within a grouping of Ae. lilii while two individuals from Uganda and Benin, identified as Ae. lilii, cluster within a grouping that is otherwise comprised of Ae. bromeliae. A hierarchical AMOVA showed there is more genetic variation within (42.83%) and between populations (55.18%) than between species (1.99%) at the COI gene.
Based on fixed differences between species in the ITS sequences, species specific primers were designed to amplify PCR products in Ae. bromeliae or Ae. lilii. Primers designed to amplify in Ae. bromeliae did not amplify a PCR product from Ae. lilii and vice versa (Fig 6). Application of this method to 110 specimens of Ae. simpsoni s.l. positively identified all individuals as either Ae. bromeliae or Ae. lilii, as detailed in Table 1. Both species use the same breeding habitats in domestic and peridomestic habitats including the leaf axils of Musa spp., Colocasia spp. and Dracena spp. Despite focussed collection attempts, neither species was found utilising tree holes for immature development. Both species occur in sympatry in Uganda, while only Ae. lilii was collected from the Republic of Benin and Ae. bromeliae was the only species detected in Tanzania (Fig 1).
Here we used multiple markers to confirm there are two closely related species of the Simpsoni Complex that occur both allopatrically and sympatrically across the sampled range in sub-Saharan Africa. These species correspond to Ae. bromeliae and Ae. lilii of the Simpsoni Complex as characterised by Mukwaya et al. [22, 27]. We infer that a third forest taxon originally reported as Ae. lilii in Walter et al. [28] relates to another species, possibly a another member of the Simpsoni Group. The development of a molecular species identification method for the YFV vector Ae. bromeliae and non-vector Ae. lilii using variation at the ITS region allowed us to make for the first time reliable inferences about mosquito ecology and distribution. Such information is vital for a complete understanding of disease transmission by these species.
All the Ae. simpsoni s.l. mosquitoes tested fell into one or other of two genetically divergent ITS lineages. Based on sequence similarity these correspond directly to the two lineages found by Mukwaya et al. [27] and we follow their previous designation as Ae. bromeliae and Ae. lilii. We consider their species designation to be reliable as it was based on both morphology and feeding preference. Firstly, even though some individuals could not be identified due to overlap in morphological characters, Mukwaya et al. [27] were able to identify some individual specimens from the defining characters of leg tarsomere banding pattern and claw morphology described by Huang [22, 23]. Secondly, Mukwaya et al. [27] identified species based upon host feeding behaviour; human landing catches were used to distinguish anthropophilic Ae. bromeliae from non-anthropophilic Ae. lilii which were collected as larvae where human biting members of the Simpsoni Complex were absent. We confirm the existence of Ae. bromeliae and Ae. lilii as two distinct species since, using ITS sequence variation, they remain genetically distinct in sympatry at several locations in Uganda.
Our study enabled us to link the work of Mukwaya et al. [27] with that of Walter et al. [28]. Whereas Muwakya et al. [27] focused on feeding behaviour, Walter et al. [28] found divergence between forest and peridomestic populations in coastal Kenya at three nuclear genes. Our phylogenetic analysis reveals that the genetic variation in the ITS region does not correspond to the two distinct lineages found in Walter et al. [28]. Both Ae. bromeliae and Ae. lilii fall into a single phylogenetic clade at nuclear genes apoLp2 (apolipophorin 2) and CYPJ92 (cytochrome p450). This is consistent with these more slowly evolving nuclear genes being unable to resolve recently diverged species due to incomplete lineage sorting [56]. The forest lineage from Kenya detected by nuclear genes apoLp2 and CYPJ92 is clearly distinct and has no shared ancestral polymorphism with either Ae. bromeliae or Ae. lilii indicating a more distant relationship. Morphological identification of mosquito species is notoriously difficult and there are many species within the wider Simpsoni Group which share morphological characteristics [26]. We therefore suggest that the forest species reported by Walter et al. [28] represents a member of the wider Simpsoni Group, rather than the Simpsoni Complex. Possible candidates for this species that are morphologically similar and known from Kenya include Ae. gandaensis, Ae. woodi, Ae. subargenteus and Ae. sampi [26].
The mitochondrial COI species barcode marker was also unable to distinguish these taxa with most genetic variation occurring between geographic populations rather than between species. Whilst this inability of COI to distinguish Ae. bromeliae and Ae. lilii could be due to incomplete lineage sorting, the phylogeny suggests it is most likely due to mtDNA introgression; there are three cases in which Ae. bromeliae clusters with Ae. lilii or vice versa, two of which occur in sympatry in Uganda. Mitochondrial introgression is fairly common between closely related mosquito taxa [33]. Given this, and that Aedine mosquitoes may be particularly prone to NUMT’s due to their large genome size [57] it would seem wise to not rely on mitochondrial markers to distinguish species in this genus without prior confirmation from an additional marker. We found that the ITS region was the only marker able to reliably separate species of the Simpsoni Complex and therefore used it to develop a PCR mediated species identification method. This method is widely applicable across the range of Ae. bromeliae and Ae. lilii except in southern Africa. For use in this region, the method should be modified to accommodate the presence of Ae. simpsoni s.s. Our PCR mediated species diagnostic method removes the difficulties imposed by morphological identification of field specimens and therefore provides a valuable asset to medical entomologists studying arbovirus transmission.
Our identification tool in conjunction with larval sampling from natural habitats has provided further information on the ecology of the Simpsoni Complex. Contrary to the wide range of plant species utilised by Ae. bromeliae as breeding habitats, the immatures of Ae. lilii have only been reported to date from the axils of Sansevieria spp., suggesting a narrower range of immature habitats [22, 23]. However, we found that Ae. lilii, like Ae. bromeliae, utilised the plant axils of Musa spp., Colocasia spp. and Dracena spp., suggesting that larval breeding sites are not as restricted as previously reported [22, 23]. Walter et al. [28] hypothesised that selective pressure for use of domestic larval habitats may have driven speciation in the Simpsoni Complex. This is not supported by our findings that showed no obvious differences in larval habitat between species, although our characterisation of larval habitats was not exhaustive. As host choice appears to be an ecological difference between these species, it is possible that divergent selection for anthropophily, and the reliable blood source it provides, could have driven species divergence [28, 58].
We have shown here that Ae. bromeliae and Ae. lilii are common in Uganda where they can be found in sympatry in peridomestic habitats. Ae. lilii was the only member of the Simpsoni Complex collected from the Republic of Benin where it is described for the first time, while only Ae. bromeliae was collected in Tanzania. In addition, Mukwaya et al.’s [27] sequence data shows that mosquitoes from Nigeria are Ae. lilii whereas those from Kenya are Ae. bromeliae. Molecular evidence therefore agrees with the earlier morphological data of Huang [22, 23] that Ae. bromeliae is prevalent in East Africa (Tanzania, Kenya, Uganda, Mayotte; Fig 1). This morphological data also indicates a wider distribution across sub-Saharan Africa for Ae. bromeliae (Fig 1) [22, 23], but it would be wise to confirm this using molecular identification. However, given that Ae. bromeliae tends to be readily collected when it is present, the molecular data suggest that only Ae. lilii is present in West Africa (Fig 1). This difference would be consistent with the early reports that West African populations are predominantly non-anthropophilic [59]. Based on our preliminary distribution data, Ae. bromeliae may be an important vector of yellow fever in East Africa where YFV has been isolated from the Simpsoni Complex previously and implemented in disease epidemics including the Ethiopian outbreak of 1960–61 [59–61]. In comparison, other mosquito vectors including Ae. aegypti, Ae. luteocephalus, Ae. furcifer and Ae. taylori may be more important for YFV disease transmission in West Africa where yellow fever has been isolated from these vectors and implicated in outbreaks [59, 62–65].
Emerging/re-emerging arboviruses such as Chikungunya and Zika are causing great concern since they are increasingly responsible for catastrophic epidemics worldwide [66–72]. Members of the Simpsoni Group transmit a range of arboviruses including yellow fever, Babanki and Ngari viruses [1]. A limited number of arboviral studies have focused on disease transmission in this species group, progress of which is hampered by incomplete taxonomic understanding. There is therefore a great need to resolve the molecular systematics of the wider Simpsoni Group that should in turn be used to develop methods of species identification. In addition to studies of arboviral risk, assessment of mosquito species ranges is required in order to relate these to differences in the distribution of arboviruses and/or arboviral genotypes such as that observed for yellow fever [59]. Improved identification methods would also facilitate studies on feeding behaviour and genetic introgression which are important for understanding the risks of zoonotic disease emergence. For example, the introgression we observed between Ae. lilii and Ae. bromeliae could increase the propensity of Ae. lilii to feed on humans, resulting in the increased transfer of zoonotic disease into humans. A precedent for this can be seen in Culex pipiens in North America where mixing of the molestus and pipiens genetic forms (predominantly human and bird feeding, respectively) increases transmission of West Nile virus to humans. A wide range of inferences on ecology and epidemiology can now be made with our molecular identification tool that should be used to assess disease risk and provide basic information for vector population control.
|
10.1371/journal.pntd.0006342 | Development and evaluation of a novel high-throughput image-based fluorescent neutralization test for detection of Zika virus infection | Zika virus (ZIKV) is an emerging arbovirus belonging to the genus flavivirus that comprises other important public health viruses, such as dengue (DENV) and yellow fever (YFV). In general, ZIKV infection is a self-limiting disease, however cases of Guillain-Barré syndrome and congenital brain abnormalities in newborn infants have been reported. Diagnosing ZIKV infection remains a challenge, as viral RNA detection is only applicable until a few days after the onset of symptoms. After that, serological tests must be applied, and, as expected, high cross-reactivity between ZIKV and other flavivirus serology is observed. Plaque reduction neutralization test (PRNT) is indicated to confirm positive samples for being more specific, however it is laborious intensive and time consuming, representing a major bottleneck for patient diagnosis. To overcome this limitation, we developed a high-throughput image-based fluorescent neutralization test for ZIKV infection by serological detection. Using 226 human specimens, we showed that the new test presented higher throughput than traditional PRNT, maintaining the correlation between results. Furthermore, when tested with dengue virus samples, it showed 50.53% less cross reactivity than MAC-ELISA. This fluorescent neutralization test could be used for clinical diagnosis confirmation of ZIKV infection, as well as for vaccine clinical trials and seroprevalence studies.
| Since 2015, DENV’s cousin known as ZIKV has been in the spotlight. It caught researchers’ attention because it rapidly spread worldwide and ZIKV infection has been associated with Guillain-Barré syndrome cases and congenital brain abnormalities in newborn infants. For being so closely related, differentiation between DENV or ZIKV infection is challenging. Among the assays used in viral serological diagnosis, the plaque-reduction neutralization test (PRNT) that was described in the 1950s seems to be more specific, although longstanding, very laborious and not capable to test large number of samples. Therefore, we developed an image based neutralization test for ZIKV that overcomes restrictions presented by PRNT. This new test is faster, robust and able to test many samples simultaneously. It was successful in distinguish ZIKV infection from other infections, such as dengue and yellow fever. This may be especially relevant to solve cases such as congenital disorders in newborns and also to elucidate the agents involved in neuropathological outcomes such as Guillain-Barré syndrome. It also can be useful in serological surveys and vaccine studies.
| Zika virus (ZIKV) is a mosquito-borne flavivirus that belongs to the Flaviviridae family, and is closely related to dengue virus (DENV). Flavivirus virions present a positive single-stranded RNA genome of approximately 11 Kb with a single open reading frame that encodes one polyprotein, which is further cleaved in 3 structural (C, prM and E) and 7 non-structural proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B and NS5) [1].
ZIKV was first isolated from a sentinel monkey in Uganda in 1947 [2] and, until 2007, it was considered endemic to Africa and Asia, when a small epidemic was reported in Yap State, Federated States of Micronesia [3]. In 2013, another ZIKV outbreak was reported in French Polynesia [4]. In 2015, ZIKV emerged in Brazil, and rapidly spread. By 2017, 48 countries and territories in the Americas had confirmed autochthonous ZIKV transmission [5–7].
In previous outbreaks, the illness was characterized by rash, conjunctivitis, subjective fever, arthralgia, and arthritis; infection appeared relatively mild, self-limiting, and nonlethal [3]. However, in recent outbreaks, an association with Guillain-Barré syndrome and congenital brain abnormalities in newborn infants of mothers infected with ZIKV during pregnancy has been observed [6,8,9]. These evidences indicate that an unequivocal diagnosis of the illness is of utmost importance for correct clinical management, especially in the case of pregnant women.
ZIKV diagnosis is based on clinical, epidemiological and laboratorial criteria. When samples are collected up to 5–7 days after the onset of symptoms, viral RNA can often be identified in serum or urine, and RT-PCR is the preferred test for ZIKV, and also for DENV and chikungunya virus (CHIKV) detection [10]. After this period, IgM antibodies may be detected by ELISA; however, flaviviruses have strong cross-reactivity, which may generate false positive results in serological tests [4,11]. This makes diagnosis of ZIKV infections quite a challenge, especially because the disease emerged in regions where other flaviviruses are endemic. Therefore, plaque-reduction neutralization test (PRNT) is indicated to measure virus-specific neutralizing antibodies and may be able to determine the etiology of infection [12].
Classical virus PRNT was first described in the 1950s and is considered the gold standard to measure neutralizing antibodies against viruses. Although being more specific, it is laborious and therefore not readily amenable to high-throughput, making it difficult to use for large-scale surveillance and vaccine trials.
In this study, we describe a fast and robust test to measure neutralizing antibody against ZIKV, which is suitable for high-throughput screening of large collections of serum specimens. This new assay is based on quantitative immunofluorescence, allying the classical PRNT format with a modern readout method.
C6/36 Aedes albopictus cells (ATCC CRL-1660) were grown in Leibovitz L-15 medium (Gibco/Invitrogen, Grand Island, NY, USA) supplemented with 5% fetal bovine serum (FBS) (Gibco/Invitrogen, Grand Island, NY, USA), 0.26% tryptose (Sigma-Aldrich, St. Louis, MO, USA) and 25 μg/mL gentamicin (Gibco/Invitrogen, Grand Island, NY, USA) at 28°C. Human-derived hepatoma cells (Huh 7.5, ATCC PTA-8561) were grown in Dulbecco’s Modified Eagle Medium: Nutrient Mixture F-12 (DMEM/F-12 medium) (Gibco/Invitrogen, Grand Island, NY, USA) supplemented with 10% FBS and 100 IU/μg/ml penicillin/streptomycin (Gibco/Invitrogen, Grand Island, NY, USA) at 37°C in a humidified, 5% CO2-controlled atmosphere. D1-4G2-4-15 hybridoma was cultivated in RPMI-1640 medium (Gibco/Invitrogen, Grand Island, NY, USA) with 25 mM HEPES and supplemented with 10% FBS, 1 mM sodium piruvate, 250 ng/ml amphotericin B and 100 IU/μg/ml penicillin/streptomycin. ZIKV strain ZV BR 2015/15261 was isolated from a patient with zika fever from Northeast Brazil in 2015. Dengue viruses from four serotypes were used. DENV1- FGA/89 was isolated from a South American patient with dengue fever in 1989 (GenBank: AF226687). DENV2- ICC 265 and DENV3- BR DEN 97–04 (GenBank: EF629367) were isolated in Brazil. DENV4- LRV13/422 (GenBank: KU513441) was isolated from a non-fatal case of dengue with hemorrhagic manifestation. To obtain viral stocks, virus were propagated in C6/36 at the multiplicity of infection (MOI) of 0.01 and titrated by focus forming assay in C6/36.
A total of 226 sera were used in this study, which was approved by Fiocruz and the Brazilian National Ethics Committee of Human Experimentation (CAAE: 42481115.7.0000.5248), as well as the waiver of the Informed Consents. Specimens were divided as follows: 29 positive sera for ZIKV were confirmed by IgM ELISA and/or real time RT-PCR; 30 IgG sera positive for DENV confirmed with Panbio IgG indirect ELISA (Alere, Brisbane, Australia); 95 IgM sera positive for DENV (from all serotypes), confirmed by IgM capture ELISA and RT-PCR; 5 sera from yellow fever virus vaccinated volunteers; and 14 negative sera. Additionally, a panel of 53 samples positive for other acute infections was tested. This panel included sera positive for Toxoplasmosis (5 samples), Epstein-Barr virus (EBV) (10), Venereal Disease Research Laboratory test (VDRL) (17), Cytomegalovirus (CMV) (10), CMV/EBV (2), Leptospirosis (7), Hantavirus (2). With exception of Zika positive sera, all samples were collected prior to ZIKV emergence in Latin America.
Zika positive sera have been received in our laboratory since ZIKV outbreak in Brazil, when it was designated as a Sentinel Laboratory by the Brazilian Ministry of Health, thus working on ZIKV diagnosis in the South region.
Huh 7.5 cells were plated in 24 well plates at a density of 1x105 cells, 16h previous to infection. Serum samples were inactivated at 56°C for 30 min, and then diluted 1/20 (followed by serial 1/3 dilutions). An equal volume of virus suspension containing 210 plaque-forming units (pfu) was mixed with diluted samples and incubated at room temperature for 1h. After this step, each mixture was inoculated onto plates with cells and after incubating at 37°C for 1h; inoculum was discarded and an overlay (1.6% CMC and 10% FBS in DMEM/F-12 medium) was added. Plates were left at 37°C for 6 days and then, cells were fixed with 3% paraformaldehyde and stained with 0.75% crystal violet. Plaques were counted and antibody titer was determined as the serum dilution that inhibited 90% of the tested virus inoculum (PRNT90).
Huh 7.5 cells were plated in 96 well plates at a density of 1.5x104 cells, 16h previous to infection. Serum samples were inactivated at 56°C for 30 min, and then diluted as described above. An equal volume of virus suspension (MOI of 0.4–300 pfu) was mixed with diluted samples and incubated at room temperature for 1h. Then, each mixture was inoculated onto plates with cells and incubated at 37°C for 1h. Inoculum was replaced with fresh medium and plates further incubated at 37°C for 48h.
Cells were fixed with cold methanol/acetone (v/v) and immunostained. Monoclonal antibody 4G2 (1/100) was used to stain virus envelope protein. It was diluted in blocking buffer (PBS with 1% BSA) and incubated at 37°C for 1h. Wells were washed three times with washing buffer (PBS with 0.05% tween 20) and incubated with secondary antibody anti-mouse IgG Alexa Fluor 488 (1/400) (Molecular Probes) in blocking buffer. Cell nuclei were counterstained with 5 μM DRAQ5 (Thermo Fisher Scientific) and washed three times with washing buffer.
Images were obtained with the Operetta High-Content Imaging System (PerkinElmer) with the objective 10x long WD. The number of images necessary to be representative for the entire well was defined and analyzed with Harmony High-Content Imaging and Analysis Software (PerkinElmer) (S1 Fig). Percentage of infected cells were obtained and normalized in relation to positive and negative controls; antibody titer was determined as the serum dilution that inhibited 90% of viral infection (NT90).
Zika IgM antibody capture enzyme-linked immunosorbent assay (MAC-ELISA) was performed accordingly to the guidelines from CDC [13] with minor modifications. A humanized monoclonal antibody (mAb) anti-flavivirus kindly provided by CDC was used as positive control. Antigens (ZIKV or Mock) were derived from β-propiolactone inactivated cell-culture supernatant from non-infected and ZIKV infected cells.
For ZIKV genome detection, viral RNA was extracted from 140 μL of samples using QIAamp viral RNA mini kit (Qiagen, Hilden, Germany). Real-time RT-PCR was performed as described by Lanciotti et al. (2008) [14], using 5 μL of RNA and Go-Taq Probe 1-Step RT-qPCR System (Promega). Assays were performed in the LightCycler 96 instrument (Roche, Mannheim, Germany) and human RNAse P was used as endogenous control [15].
Assay quality was assessed by Z’ = 1– [3(σp+σn)/(μp- μn)], where σ is the standard deviation, μ is the mean of both positive (p) and negative (n) controls. Results were considered when Z’ was higher than 0.5 [16]. Neutralization curves were obtained using the software Prism (GraphPad version 6, USA) and PRNT90 and NT90 were calculated by the log (agonist) vs. response–Find ECanything curve, with a hillslope of 1.
To develop and validate the newly proposed fluorescent neutralization test as a potential substitute to the low throughput and labor intensive classical PRNT, we tested several parameters seeking for reproducible and faster results.
ZIKV strain ZV BR 2015/15261 was chosen because it is a recent Brazilian clinical isolate and therefore a good representative to test serum samples from this region. Viral stocks were obtained from the second viral passage and by using low multiplicity of infection (MOI of 0.01) in C6/36 cell line, due to its good infection rates and low cytotoxicity. A kinetic of virus growth was performed between the third and tenth day after infection (Fig 1A) to determine the time point to recover culture supernatants. Viral stocks were harvested at the fifth day after infection during the middle to end of the exponential phases of growth, to avoid high concentrations of defective interfering particles that could lead to falsely low neutralization titers.
Huh 7.5, a human-derived hepatoma cell line, was chosen for the neutralization assays, because it is permissive to ZIKV and other flavivirus infection and also can be automatically well segmented with a software tool. The appropriate cell seeding density was defined as 1.5x104 cells per well (34 mm2), since it has a sufficiently high number of cells but with enough spatial distribution for proper identification and accurate analysis. A MOI of 0.4 was used for all experiments because this condition yielded around 70% of infected cells after 48h (Fig 1B). Cell infection was visualized by an indirect immunofluorescence assay, with detection of ZIKV E protein by the 4G2 mAb and secondary anti-mouse IgG Alexa Fluor 488; nuclei were counterstained with DRAQ5. Four images per well (representative of the whole well) were acquired with the Operetta High-Content Imaging System and analyzed with the Harmony Analysis Software (PerkinElmer).
After the standardization step, we proceeded to the neutralization assay. Serum specimens were heat inactivated to reduce the effects that complement factors may have on final results. Serum and virus samples were mixed to allow neutralization. After the incubation period the mixture was added to cells so infection could occur by non-neutralized virus. The neutralization titer that inhibits 90% of viral infection (NT90) was used to analyze results (Fig 2).
Some criteria were followed in order to accept a valid assay. Among them, a uniform number of cells per well, appropriate percent of infection of controls, no serum toxicity observed with low serum dilutions, and a Z’ higher than 0.5.
In order to observe inter-assay variability, one negative and one positive sample were tested in three independent assays. It was observed a low variation for the negative (0.93 ± 0.16), and for the positive samples (188.46 ± 3.01), showing the robustness of the test. The average Z’ observed for all plates was 0.61.
PRNT is the gold standard for measurement of flavivirus neutralization. Therefore, neutralizing titers obtained from 12 serum samples by using either the new fluorescent neutralization test and or classical PRNT were compared. Similar neutralization results were obtained with the two approaches, with a correlation of 0.88 (Fig 3). This demonstrates the robustness of the newly developed test and that it could be used as a replacement of the traditional test, using the same interpretation guidance suggested by CDC [17].
The new proposed test was validated with a set of serum samples previously tested. This panel included sera positive for flavivirus and non-flavivirus acute infections and negative serum from healthy donors (Table 1). Zika positive samples were collected during the disease outbreak in Brazil; all the other samples were collected previous to the ZIKV emergence in the country. All samples were submitted to the fluorescent neutralization test and the NT90 was calculated (Fig 4).
Following recommended interpretation for neutralization results [17], a titer higher than 10 is supposed to be considered positive. However, it was observed that several IgM positive samples for dengue would be erroneous considered positive for ZIKV. Therefore, a more restrictive result interpretation was employed as follows: samples were considered negative when NT90 <10, inconclusive when NT90 ≥10 and <20, and positive when NT90 ≥20.
Even with the higher cutoff value, it was possible to observe six DENV IgM samples that cross-reacted in ZIKV neutralization assay; while another 10 samples were inconclusive. No cross reactivity was observed when samples of DENV IgG, other acute infections or YFV vaccine were analyzed.
Regarding the Zika positive panel (Table 2), samples were tested by Zika MAC-ELISA and/ or real time RT-PCR, and then divided into two groups: early infection (serum PCR positive and variable anti-ZIKV IgM) and late infection (PCR negative and anti-ZIKV IgM positive samples). PCR positive samples presented low neutralizing titers (<23), while NT90 of PCR negative /IgM positive samples ranged from 20.98 to 581.80.
Paired samples (presented in blue in Fig 4) were obtained from four patients; first collections were all RT-PCR positive and presented low NT90 titers, while second collections obtained 3 to 6 months after the first one had neutralization titers increased to levels a lot higher than the cut off value (Fig 5).
To assess the cross reactivity between ZIKV and DENV infections in MAC-ELISA format assay and the fluorescent neutralization test, 95 DENV well-characterized positive samples were tested (Table 3). Neutralization titers of these samples are presented in Fig 4. This sample panel was obtained between the years of 2004 and 2006 in Venezuela, thus before the emergence of ZIKV in the region. It is worth mentioning that the panel is composed by paired samples and viral isolation during acute phase of infection was used as “gold standard” for DENV infection.
Among the 95 IgM DENV positive samples tested by MAC-ELISA, 39 cross-reacted and presented false positive results for ZIKV and 25 were inconclusive or undetermined. On the other hand, only six samples presented false positive results for ZIKV and 10 were inconclusive in the neutralization test (Table 4). Thus, the novel neutralization test presented 50.53% less cross reactivity than MAC-ELISA, and the rate of correct identification of ZIKV negative serum increased from 32.63% to 83.16%.
The fluorescent neutralization test format can be expanded to other diseases. As a proof of concept, the test was adapted to identify neutralization antibodies to dengue virus. The same standardization steps used previously were employed to develop a test for the four serotypes of DENV. Optimal harvest time for viral stocks was between the 6 and 7th day after infection and, in order to obtain around 70% of infection after 48h, a MOI of 0.1 was used. DENV fluorescent neutralization test was able to identify neutralization antibodies against the four serotypes of the virus in all DENV IgM positive samples tested (Table 5). However, it was not possible to identify which DENV serotype was responsible for the current infection according to fluorescent neutralization assay results, indicating a probable secondary DENV infection.
Since ZIKV emerged in South America causing a number of outbreaks with reported cases associated with Guillain-Barré syndrome and congenital brain abnormalities in newborn infants, a great effort to develop specific and reliable diagnosis tests has been made [18–20].
A definitive ZIKV diagnosis is achieved by detecting viral RNA in patient serum, or other samples like urine, semen and placenta. Although RT-PCR assay is trustworthy and with good sensitivity and specificity, viremia among ZIKV-infected patients are relatively low and detectable for only a few days after the onset of symptoms [14].
For individuals beyond this viremia window, a serologic test must be employed. The most common used method is the detection of reactive IgM antibodies by ELISA. A disadvantage in using this option for ZIKV is the high number of false positive results due to cross-reaction with antibodies against DENV, and the low sensitivity of most existing immunoassays [11,21]. A novel ELISA based on recombinant ZIKV non-structural protein 1 (NS1) was able to eliminate cross-reactions with antibodies to DENV and other flaviviruses, although it presented low sensitivity in the IgM format [18].
In order to overcome this issue, a CDC diagnostic guideline recommends that presumptive positive or equivocal MAC-ELISA result for ZIKV needs to be verified with a confirmatory PRNT [10].
As previous stated, although PRNT is the gold standard for flavivirus serological test, a number of limitations prevents its use in large scale to test a great number of samples, as required during outbreaks or to perform serological surveys. In addition, there is a recommendation for serological testing of asymptomatic pregnant women with history of travel to ZIKV endemic regions or those living in areas with active viral transmission [22].
This study describes the development and validation of a novel image based neutralization test for ZIKV that overcomes restrictions presented by PRNT. Previous studies have developed assays for replacement of DENV PRNT. Vorndam and Beltran (2002) developed and evaluated a microneutralization test to measure anti-dengue antibodies using an in situ ELISA [23]. Additionally, a 96-well format flow cytometry-based neutralization assay was proposed, and similar neutralization patterns were observed when compared to classical PRNT [24]. The disadvantages observed included the high intra-assay variability and the need to remove adherent cells from wells.
Recently some alternative assays to ZIKV PRNT have also been reported. A MTT-based cell viability assay for ZIKV neutralizing antibodies quantification has been developed, and although it does not require expensive equipment or costly reagents, it depends on virus-induced cytopathic effect [25]. Shan et al. (2017) developed a reporter virus neutralization test (RVNT), based on ZIKV and DENV luciferase reporter viruses. The assay maintained relative specificity of traditional PRNT and was further evaluated with 258 clinical serum specimens, displaying a 93.1% agreement with the traditional ZIKV PRNT titers [26,27]. Furthermore, a neutralization assay in which the endpoint is measured by real-time PCR was proposed [28].
The novel fluorescent neutralization assay developed here combines the classical neutralization protocol with a new automatized readout method, employing a high-content imaging system. From seeding cells to obtaining results, the new test takes around 72h, in contrast to PRNT that can take up to 8 days [29], and also depends on manual counting of plaques, which can vary from person to person. Besides that, this new assay is able to test at least ten serum samples against a virus on a single 96-well plate, with dilutions performed via multichannel pipetting devices that increase assay capacity.
Maistriau et al. (2017) also proposed a fluorescent neutralization test using a high-throughput image acquisition system. However, it is based on the translocation of the transcription factor IRF3 in response to infection [30], thus requiring a careful selection of cell lines according to the virus of interest. In contrast, we propose a robust and simple method that can be easily set up to investigate other flavivirus infections.
In this study, a curve fitting method from several serum dilutions was used to calculate neutralization titers, which allows a more precise result, in contrast to simply report the reciprocal of the last serum dilution that shows 50 or 90% reduction of infection. The neutralization titer which inhibits 90% of viral infection (NT90) was used because it is indicated for epidemiological studies or diagnostic purposes in endemic areas, decreasing background serum cross-reactivity among flaviviruses [29].
It has been reported that people exposed to secondary DENV infections develop broadly neutralizing antibodies that neutralize different serotypes other than the one responsible for current infection [31], as it was also demonstrated in the DENV fluorescent neutralization test (Table 5). Additionally, sera from patients with secondary DENV infection exhibit potent cross-reactivity against ZIKV [11]. In this context, cross reactivity between ZIKV and DENV is quite expected, as the viruses envelope proteins share a high degree of homology with a sequence identity of 54% and nearly identical structures. The fusion loop, that is an important antibody target, is 100% conserved between the two viruses [11]. Therefore, increasing specificity of serological tests is particularly relevant, since ZIKV emerged in flavivirus endemic regions.
Aiming to reduce false positives results, samples were considered ZIKV positive when NT90 ≥20, while NT90 <10 samples were scored as negatives. When NT90 ranged from ≥10 to <20, results were recorded as inconclusive. Other studies have also used a higher cut off PRNT90 value [3,28]. Based on those parameters, higher specificity was achieved when compared to MAC-ELISA, yielding in less ZIKV false positive results for DENV IgM positive serum samples. Only 10.53% of inconclusive and 6.32% of false positive results were observed with these settings.
This result is particularly remarkable, since in another assay, up to 100% of cross reaction with ZIKV was observed when acute and convalescent sera from nine Thai patients with confirmed DENV infection by RT-PCR were tested, both in binding and in neutralization assays [11]. Another study using RVNT for anti-ZIKV antibodies detection, showed 20% of erroneous results in the presence of anti-DENV antibodies, although no false positive results with Yellow fever and West Nile positive samples were observed [27]. The real-time PCR neutralization assay also reported significant cross-reactivity when testing a serum specimen from a patient with proven current ZIKV infection which had a background of DENV infection [28].
It is noteworthy that higher cut off values may reduce assay sensibility, i.e, some samples of ZIKV early infections can become inconclusive. In those cases, a molecular diagnosis can be employed and/ or a second serum collection should be tested, since antibodies might not have yet reached detectable levels. This was observed when paired samples were tested and an increase in neutralization titers was observed.
As a conclusion, the developed fluorescent neutralization test offers significant advantages over classical PRNT. It is faster, prompt to high throughput adaptation, has automated reading of results, and is more specific than MAC-ELISA assay. As expected it also presents some limitations, as it does not discriminate between antibody classes, requires expensive equipment and can be performed only in selected laboratories. Nevertheless, it will make it possible to test simultaneously a large number of samples and against different viruses, assisting the correct management of suspected patients or asymptomatic pregnant woman and be employed in seroprevalence surveys.
|
10.1371/journal.pbio.1001506 | Neuronal Expression of Glucosylceramide Synthase in Central Nervous System Regulates Body Weight and Energy Homeostasis | Hypothalamic neurons are main regulators of energy homeostasis. Neuronal function essentially depends on plasma membrane-located gangliosides. The present work demonstrates that hypothalamic integration of metabolic signals requires neuronal expression of glucosylceramide synthase (GCS; UDP-glucose:ceramide glucosyltransferase). As a major mechanism of central nervous system (CNS) metabolic control, we demonstrate that GCS-derived gangliosides interacting with leptin receptors (ObR) in the neuronal membrane modulate leptin-stimulated formation of signaling metabolites in hypothalamic neurons. Furthermore, ganglioside-depleted hypothalamic neurons fail to adapt their activity (c-Fos) in response to alterations in peripheral energy signals. Consequently, mice with inducible forebrain neuron-specific deletion of the UDP-glucose:ceramide glucosyltransferase gene (Ugcg) display obesity, hypothermia, and lower sympathetic activity. Recombinant adeno-associated virus (rAAV)-mediated Ugcg delivery to the arcuate nucleus (Arc) significantly ameliorated obesity, specifying gangliosides as seminal components for hypothalamic regulation of body energy homeostasis.
| Obesity is a growing health threat that affects nearly half a billion people worldwide, and its incidence rates in lower income countries are rising dramatically. As obesity is a major risk factor for type II diabetes and cardiovascular disease, significant effort has been put into the exploration of causes, prevention, and potential treatment. Recent research has demonstrated that a region of the brain called the hypothalamus is a major integrator of metabolic and nutrient signals, adapting food intake and energy expenditure to current metabolic needs. Leptin or insulin receptors located in the plasma cell membrane of neurons sense energy signals from the body. They transmit this information inside the cell, which then regulates neuronal function. In this study, we show that leptin receptors interact with gangliosides, a class of plasma membrane lipids. This interaction is a prerequisite for proper receptor activation. Consequently, ganglioside loss in hypothalamic neurons inhibits leptin receptor signal transduction in response to energy metabolites. Furthermore, mice lacking gangliosides in distinct forebrain areas, amongst them the hypothalamus, develop progressive obesity and hypothermia. Our results suggest a previously unknown regulatory mechanism of plasma membrane lipids for hypothalamic control of body weight.
| The investigation of pathogenetic mechanisms underlying obesity has attained significant interest, as obesity has become an endemic metabolic disturbance worldwide. Elevated peripheral energy storage can develop as a consequence of alterations in the neuronal feedback circuits regulating energy homeostasis. The hypothalamus is the main CNS integrator of peripheral energy signals, matching energy intake to energy expenditure for body weight maintenance [1].
Among the most extensively studied peripheral molecules involved in regulating energy homeostasis and feeding behavior in the CNS are the adipocyte-derived hormone leptin as well as insulin [2],[3]. Among numerous leptin- and insulin-sensitive brain areas, the hypothalamic Arc is one of the main regions integrating peripheral energy signals and initiating adaptive metabolic and behavioral responses [4].
Recently, several CNS regions targeted by leptin have emerged that are involved in the regulation of energy metabolism, such as the brain stem nucleus of the solitary tract (NTS) and reward circuits involving the ventral tegmental area [5],[6]. Still, leptin is suggested to exert anti-obesity effects by signaling through “long form” leptin receptors (ObR) abundantly present on both orexigenic neuropeptide Y (NPY)/agouti-related peptide (AgRP) neurons and anorexigenic pro-opiomelanocortin (POMC) neurons in the Arc. Excess NPY signaling abates sympathetically mediated thermogenesis, thereby reducing energy expenditure [7]. NPY and AgRP expression is attenuated upon ObR-induced phosphatidylinositol-3-OH-kinase (PI3k) signaling [8]. Conversely, leptin stimulates the expression of the POMC-derived neurotransmitter α-melanocyte-stimulating hormone (α-MSH) through the Janus kinase/signal transducer and activator of transcription (Jak-Stat) pathway [9]. Alpha-MSH, a potent agonist of melanocortin receptors, inhibits food intake and stimulates the expenditure of excess energy in the body, thus preventing obesity development [10].
Insulin exerts its anorexigenic effects in hypothalamic neurons by directly stimulating insulin receptor autophosphorylation and activation of PI3k. Even though both insulin and leptin receptor stimulation leads to activation of PI3k and subsequent formation of phosphatidylinositol (3,4,5)-triphosphate (PIP3) [11], it has been shown that both hormones exert converging direct actions on POMC neurons, while having opposite effects on AgRP/NPY neurons [12].
GCS is the key enzyme for the biosynthesis of glycosphingolipids (GSLs) and gangliosides, a class of acidic GSLs abundantly expressed by neurons and glial cells [13],[14]. Ganglioside-depleted neurons are viable and show apoptosis rates comparable to wild-type neurons [15]. GSLs including gangliosides contribute to the formation of membrane microdomains, which are important mediators of intracellular signal transduction [16]. GCS expression is crucial for initial postnatal brain maturation and Ugcgf/f//NesCre mice with constitutive Ugcg deletion in brain tissue under the control of the nestin promoter die within 3 wk after birth [15]. In 2003, it was shown that GM3 synthase-deficient mice are more sensitive to insulin, thereby protecting these mice from high-fat-diet-induced insulin resistance [17]. A different ganglioside species, GD1a, has been shown to exert activating effects on tyrosine kinase receptors [18]. To address the functional role of GCS in neuronal regulation of energy homeostasis, we have generated and characterized mice with inducible neuron-specific Ugcg deletion in adult mouse CNS (Ugcgf/f//CamKCreERT2 mice). Cre activity in this mouse model was restricted to distinct populations of forebrain neurons. Hypothalamic nuclei involved in the regulation of energy homeostasis were targeted by this approach. Explicitly, Cre activity was absent in the brain stem NTS, which also contributes to regulation of energy homeostasis.
The present study highlights GCS-derived gangliosides as mediators for ObR-dependent signal transduction at the hypothalamic neuronal membrane. GCS-depleted neurons failed to show ObR activation upon leptin stimulation. Major neuronal gangliosides GM1 and GD1a were recruited to ObR upon ligand stimulation and subsequent signal transduction depended on ganglioside expression in hypothalamic neurons. Ugcgf/f//CamKCreERT2 mice deficient in GSLs in hypothalamus developed progressive obesity and decreased sympathetically mediated thermogenesis. rAAV-mediated Ugcg delivery to the hypothalamic Arc with ensuing nucleus-specific GSL synthesis significantly ameliorated obesity.
Ugcgflox/flox (Ugcgf/f) mice were bred with mice expressing the inducible CreERT2 recombinase under the control of the Calcium/Calmodulin-dependent Kinase II-alpha (CamK) promoter, resulting in forebrain neuron-specific Ugcg deletion (Ugcgf/f//CamKCreERT2) followed by ganglioside depletion after tamoxifen injection (Figure 1A). Generation of Ugcgf/f mice and CamKCreERT2 mice has been described earlier [15],[19].
Beta-galactosidase (X-Gal) staining of brains from R26R/Ugcgf/+//CamKCreERT2 reporter mice indicated strong Cre activity in distinct hypothalamic nuclei, namely in the Arc (Figures 1B and S1B), in the paraventricular nucleus, and in median preoptic area (MnPO) (Figure S1A,B). Additional Cre activity was detected in the lateral hypothalamic area (LHA), in hippocampus, and in the cerebral cortex (Figure S1A,B). Notably, Cre activity was absent in the ventromedial hypothalamus and the NTS in the brain stem (Figure S1A,B). Ganglioside depletion was confirmed in Cre-targeted areas by GD1a immunofluorescence, whereas non-targeted areas retained GD1a expression (Figure 1B and Figure S1A).
Consistent with the expected Cre-activity pattern, in situ hybridization showed Ugcg mRNA depletion in hippocampus, cerebral cortex, amygdala, as well as hypothalamic nuclei (Figure S1C). Recombination events were absent in peripheral organs and peripheral nervous tissue (Figure S1D).
Neuron-dense total hippocampi showed significant and stable ganglioside reduction 3 wk postinduction (p.i.), as assessed by thin layer chromatography (TLC) (Figure S1E). Residual gangliosides in the dissected tissue were assumed to result from glial cells as well as from innervating nerve fibers emerging from nontargeted neurons [14]. Ceramide levels in Cre-targeted neuronal populations were unchanged (Figure 1C), and a slight increase in sphingomyelin could be detected (Figure S1F).
In order to investigate if ganglioside depletion abated general neuronal function and integrity in Ugcgf/f//CamKCreERT2 mice, both electron microscopy and electrophysiological slice recordings were done at late time points p.i. Electron microscopy from Arc neurons displayed normal ultrastructure of the neuronal nucleus, organelles, and an intact, regular plasma membrane of Ugcgf/f//CamKCreERT2 mice both 6 and 12 wk p.i. (Figure 1D). Basic biophysical parameters [spontaneous firing rate, action potential (AP) width, and AP rate of rise] from slice recordings of Arc neurons 12 wk p.i. were unaltered (Figure S2A). The resting membrane potential and the AP threshold were marginally increased in Ugcgf/f//CamKCreERT2 mice, however not to an extent that impairs neuronal function (Figure S2B).
In order to confirm these findings in vitro, immortalized mouse hypothalamic cells (N-41 cells) expressing GCS-derived gangliosides (Figure S3A,B) were treated with n-butyldeoxynojirimycin (NB-DNJ) specifically inhibiting GCS [20]. NB-DNJ treatment resulted in approximately 80%–90% ganglioside depletion (Figure S3C). Consistent with the findings in Ugcgf/f//CamKCreERT2 mice, membrane integrity and normal cellular ultrastructure of ganglioside-depleted N-41 cells was confirmed by electron microscopy (Figure S3D). Additionally, passive and active membrane properties of cultured primary GCS-deficient hypothalamic Ugcgf/f//NesCre neurons [15] were examined by whole-cell recordings. There were no differences toward control cells in membrane resistance, capacitance, and resting potential (Figure S3E). Spikes evoked by somatic current injection had unaltered threshold, amplitude, and duration (Figure S3F). These results indicate that basic neuronal integrity and general function are not affected by Ugcg deletion and subsequent lack of plasma membrane gangliosides.
Coinciding with neuronal ganglioside depletion 3 wk p.i. female and male Ugcgf/f//CamKCreERT2 mice displayed progressive body weight increase (Figure 2A,B). This phenotype was not detected in heterozygous mice (Figure S4A), as residual GCS activity accounted for maintenance of neuronal ganglioside biosynthesis [15]. Ugcgf/f//CamKCreERT2 mice were larger than control littermates 16 wk p.i. (Figure 2C).
Hematoxylin and eosin (HE) staining revealed enlarged adipocytes in epigonadal white adipose tissue (WAT) (Figure 2D). In line with this, epigonadal WAT pad weight was significantly elevated (Figure 2E). Whole body nuclear magnetic resonance (NMR) analysis revealed that body weight increase was due to progressive accumulation of body fat (Figure 2F); lean mass was only marginally elevated 4 wk p.i. (Figure S4B). Adjusted for body weight the initial increase of fat and lean mass was proportional, whereas at later stages fat mass overrode lean mass gain (Figure S4C). Liver steatosis and morphological changes in major peripheral organs of obese Ugcgf/f//CamKCreERT2 mice were not detected 9 wk p.i. (Figure S4D). Serum enzyme activities indicative for liver function (glutamate dehydrogenase, glutamic oxaloacetic transaminase, and glutamic pyruvic transaminase) were unaltered (Figure S4E). Likewise, serum cholesterol, urea, glucose, and creatinine did not show any biologically relevant abnormalities (Figure S4F). Coincident with obesity, Ugcgf/f//CamKCreERT2 mice were less glucose tolerant than Ugcgf/f mice 12 wk p.i. (Figure S4G) and insulin sensitivity was marginally impaired 10 wk p.i. (Figure S4H). These results demonstrate that Ugcgf/f//CamKCreERT2 mice develop progressive obesity that is evident in all adipose compartments with constant lean mass and a shift in body composition toward fat accumulation.
As tight regulation of energy homeostasis is crucial for body weight maintenance [1], a metabolic characterization was carried out in order to study the relation of energy intake to energy expenditure. Food intake and metabolizable energy (EMET) adjusted to body weight were slightly elevated in Ugcgf/f//CamKCreERT2 mice before the onset of obesity 3 wk p.i. (Figure 3A,B) when gangliosides were already depleted in Cre-targeted brain regions. Hyperphagia was no longer evident 6 and 11 wk p.i., as food intake and EMET were simply elevated due to higher body weight (Figure 3A,B). Fecal excretion of free fatty acids (FFAs) as well as energy content of feces and extraction efficiency from the food (Figure S5A) were unaltered. Thus, abnormalities in food intake do initially contribute to obesity development, but not for obesity maintenance.
Energy expenditure was monitored by indirect calorimetry for 21 h. Before onset of body weight gain, the metabolic rate was indistinguishable from Ugcgf/f mice 2 wk p.i. (Figure S5B). When adjusted for body weight, the average metabolic rate tended to be lower in Ugcgf/f//CamKCreERT2 mice at 5 and 9 wk p.i. (Figure 3C). Spontaneous locomotor activity is one contributor to daily energy expenditure and has been reported to be decreased in obese rodents [21]. However, both before the onset of weight gain and during progressive adiposity, spontaneous open field activity of Ugcgf/f//CamKCreERT2 mice was indistinguishable from control littermates (Figure S5C).
The respiratory exchange ratio (RER) provides information on metabolic fuel preferences [22]. Ugcgf/f//CamKCreERT2 mice displayed significantly elevated average daily RER values (Figure 3D). This finding suggests a shift from lipid oxidation toward lipid storage [22]. In line with this, fat mobilization in response to fasting as assessed by measuring plasma nonesterified free fatty acids (NEFAs) was impaired. Significantly decreased plasma NEFAs were detected in Ugcgf/f//CamKCreERT2 mice 11 wk p.i. (Figure S5D), suggesting a reduced capability to mobilize lipid stores when challenged by food withdrawal.
After the onset of weight gain, Ugcgf/f//CamKCreERT2 mice displayed a prominent drop in core body temperature, as exemplarily depicted 10 wk p.i. (Figure 3E). Adipocytes in intrascapular brown adipose tissue (iBAT) were enlarged (Figure S6A), suggesting reduced triglyceride turnover. Ultrastructural analysis of iBAT furthermore revealed mitochondrial disorganization as well as a lower average mitochondrial size (Figure S6B,C). Thermogenesis in iBAT is regulated by synergistic actions of thyroid hormones and sympathoadrenergic signaling [23]. Free triiodothyronine (fT3) and free thyroxine (fT4) levels were normal in Ugcgf/f//CamKCreERT2 mice (Figure S6D,E). Thus, thyroid dysfunction was unlikely to account for inappropriate thermoregulation. Decreased sympathetic outflow to adipose tissue is assumed to be associated with impaired lipid mobilization [24]. In fact, both iBAT sympathetic activity, as assessed by norepinephrine (NE) turnover rate (Figure 3F, Figure S6F), and NE content (Figure S6G) were decreased in Ugcgf/f//CamKCreERT2 mice.
These results demonstrate that Ugcgf/f//CamKCreERT2 mice develop progressive obesity and a shift in body composition toward fat accumulation initially supported by hyperphagia, but maintained due to hypometabolism and hypothermia.
Several distinct hypothalamic and nonhypothalamic brain regions were targeted by Cre activity in Ugcgf/f//CamKCreERT2 mice. Arc neurons in Ugcgf/f//CamKCreERT2 mice expressing the long form of the ObR were targeted by Cre activity, as demonstrated by co-immunofluorescence of PStat3 and beta-galactosidase (b-Gal) in R26R/Ugcgf/+//CamKCreERT2 reporter mice (Figure 4A). Other leptin-responsive neurons outside the Arc also targeted by Cre activity, such as the MnPO are likely in part contributing to the observed phenotype. However, ObR-expressing neurons in the LHA seem to be recessed by Cre activity (Figure S7A,B).
In order to furthermore clarify the role of the Arc in obesity development, we injected recombinant adeno-associated viruses encoding either Ugcg and lacZ (rAAV-Ugcg/LacZ) or only lacZ (rAAV-Empty/LacZ) bilaterally into the Arc of Ugcgf/f//CamKCreERT2 mice after ganglioside depletion before 4 wk p.i. Injection of rAAV-Ugcg/LacZ significantly ameliorated obesity, underlining the importance of Ugcg expression in the Arc for body weight maintenance (Figure 4B). Consistently, serum leptin levels tended to be lower in rAAV-Ugcg/LacZ-treated mice (Figure 4C). We verified correct targeting of the Arc by X-Gal staining of the brains injected with rAAV-Ugcg/LacZ and displayed targeted regions in a schematic drawing as well as a typical staining (Figures 4D–F and S7C). Animals that were not targeted by rAAV-Ugcg/LacZ in the Arc (rAAV-Ugcg/LacZ missed) did not improve their weight gain (Figure S7D). Restored ganglioside biosynthesis in the Arc of rAAV-Ugcg-treated animals compared to mice injected with viruses encoding empty plasmid was demonstrated by GD1a immunofluorescence (Figure 4G and Figure S7E).
Taken together, these results indicate that loss of GCS expression in the Arc is significantly involved in part of the metabolic deregulation seen in Ugcgf/f//CamKCreERT2 mice.
Since the number of neurons in the Arc did not differ between Ugcgf/f//CamKCreERT2 mice and controls (Figure S8A), a functional analysis of the Arc was performed. Leptin signaling in the hypothalamus is crucial for the maintenance of body weight and energy homeostasis. As adipocyte-secreted leptin is a major regulator of body weight in the CNS, we hypothesized that leptin signaling might be disturbed in GCS-deficient neurons of Ugcgf/f//CamKCreERT2 mice. In order to test this hypothesis, we investigated hypothalamic Stat3 phosphorylation (PStat3) in the Arc after peripheral leptin stimulation. Decreased PStat3 was detected by immunofluorescence in the Arc (Figure 5A) and by Western blot in mediobasal hypothalamus (Figure S8B). Interestingly, baseline Stat3 levels were elevated in Ugcgf/f//CamKCreERT2 mice (Figure S8C). The PStat3/Stat3 ratio was decreased both at baseline and upon leptin challenge (Figure S8D).
It has been shown that deficient ObR signaling due to leptin resistance of the Arc in mice with diet-induced obesity (DIO) is a consequence of long-term elevated leptin levels [25]–[27]. The suppressor of cytokine signaling 3 (SOCS-3) is a major negative regulator of the ObR that is elevated in rodent models of leptin resistance [25],[28]. In line with progressive obesity, Ugcgf/f//CamKCreERT2 mice show indeed elevated leptin levels 7 wk p.i. (Figure 5B). However, expression of hypothalamic Socs-3 did not rise with increasing obesity and leptin levels, as measured 2, 6, and 9 wk p.i. (Figure 5C). Moreover, hypothalamic ObR expression, usually elevated in leptin-resistant rodents [29],[30], was normal in Ugcgf/f//CamKCreERT2 mice 6 wk p.i. (Figure 5D).
To further investigate if GCS-derived gangliosides regulate proper leptin receptor signaling at the level of the plasma membrane in hypothalamic neurons, we first assured that loss of gangliosides would not interfere with ObR transport to the membrane, which would have impaired ObR signaling per se. ObR was labeled by an in situ proximity ligation assay (PLA) on non-detergent-perturbed cells by two ObR antibodies. The number of detected surface ObR PLA spots on cells treated with NB-DNJ was similar to control cells (Figure 5E), indicating that ObR at the plasma membrane of ganglioside-depleted hypothalamic cells is not significantly changed compared to control cells.
As GCS-derived gangliosides have previously been shown to modulate the activity of plasma-membrane-located receptors through close interactions in both adipocytes [17] and neurons [31], we investigated ObR interactions with major neuronal gangliosides. The PLA indicating close proximity events [32] indeed revealed proximity between ObR and gangliosides GM1 and GD1a. In demonstration of activity-dependent interaction between GSL and ObR, the number of GD1a/ObR and GM1/ObR PLA spots increased upon stimulation with leptin (Figures 5F,G and S8E). Complex formation between GD1a/GM1 with ObR was further corroborated by co-immunoprecipitation (Co-IP) of ObR and GD1a/GM1 in saline- and leptin-stimulated N-41 cells (Figures 5H and S8F). As N-41 cells do not express the complex neuronal gangliosides GD1b and GT1b, potential interactions with ObR had to be analyzed in hypothalamic tissue of Ugcgf/f mice. GD1b and GT1b could not be co-precipitated with ObR (Figure 5H).
Ganglioside-depleted cells were then assessed for leptin-dependent signal transduction. Ganglioside-depleted cells did not show the leptin-stimulated increased complex formation between ObR and Jak (Figures 5I and S8G). Time- and dose-dependent Jak phosphorylation could be induced by leptin treatment in N-41 cells and was decreased in NB-DNJ-treated GSL-depleted cells (Figures 5J and S8H). It has to be noted that NB-DNJ evokes ganglioside depletion by only approximately 80%–90% (Figure S3C). Thus, residual gangliosides in the plasma cell membrane may explain the appearance of a P-Jak signal at a late time point after stimulation of NB-DNJ-treated cells. Ganglioside-depleted N-41 cells showed decreased Jak phosphorylation 30 min after stimulation with 0.5 µg/ml leptin (Figure 5J,K).
These results have now shown that two major neuronal GCS-derived gangliosides, GD1a and GM1, form dynamically leptin-stimulated complexes with ObR on the plasma membrane and that loss of gangliosides decreases signal transduction in hypothalamic neurons.
It is known that mice with deficient leptin receptor (db/db mice) function develop obesity and lack hypothalamic responsiveness to leptin stimulation [33]. Regarding the finding that neuronal gangliosides enhance ObR signaling, we hypothesized that hypothalamic neuronal function may be altered in Ugcgf/f//CamKCreERT2 mice. In order to investigate this question, neuronal activity after intraperitoneal (i.p.) leptin injection was evaluated by c-Fos staining [34]. Leptin-induced c-Fos formation was normal in non-obese Ugcgf/f//CamKCreERT2 mice 1–2 wk p.i. (Figure 6A). Since ganglioside depletion coincides with the start of the obesity development, Ugcgf/f//CamKCreERT2 mice that were weight-matched to control littermates were analyzed 3–4 wk p.i. Decreased leptin responsiveness could already be observed in the Arc of these mice (Figure 6B) as well as in the Arc of obese mice 6 wk p.i. (Figure 6C).
Neurons in the nontargeted and non-ganglioside-depleted VMH retained responsiveness to leptin at all time points (Figure 6D–F). As expected, the nontargeted brain stem NTS of Ugcgf/f//CamKCreERT2 mice showed regular leptin-induced c-Fos staining 6 wk p.i. (Figure S9).
Altogether, these results indicate a primary deficiency of ganglioside-depleted hypothalamic neurons to respond adequately to peripheral leptin signals.
Antagonistic orexigenic NPY and anorexigenic POMC neurons in the hypothalamic Arc are first-order responsive neurons initiating metabolic adaptations to altered peripheral leptin levels [4]. In order to determine leptin-dependent NPY and POMC neuronal function, neuronal activity and ObR activation were assessed by semiquantitative analysis of c-Fos, PStat3, and PIP3 formation in response to peripheral leptin injections. Leptin engaged POMC neurons (α-MSH positive) in control mice, as indicated by increased c-Fos (Figure 7A). Significantly elevated PStat3 (Figure 7B) and PIP3 formation (Figure S10A) confirmed activation of their ObR. Before ganglioside depletion (1–2 wk p.i.), POMC neurons of Ugcgf/f//CamKCreERT2 mice responded normally to leptin. However, c-Fos, PStat3, and PIP3 formation were not elevated in response to leptin in obese GSL-deficient mice 6 wk p.i. (Figures 7A,B and S10A). No significant changes were found in mediobasal hypothalamus (MBH) baseline mRNA expression of Pomc and cocaine- and amphetamine-regulated transcript (Cart) mRNA 6 wk p.i. (Figure S10B).
While a slight decrease in c-Fos–positive NPY neurons was found in leptin-injected control mice, leptin did not show any such effect in Ugcgf/f//CamKCreERT2 mice 6 wk p.i. (Figure 7C). Similarly, leptin did not raise PStat3 levels in NPY neurons of Ugcgf/f//CamKCreERT2 mice 6 wk p.i. (Figure 7D) and did not have any direct effect on PIP3 formation (Figure S10C). Remarkably, basal mRNA expression of Agrp and Npy was markedly elevated in the MBH of Ugcgf/f//CamKCreERT2 mice 6 and 9 wk p.i., with Agrp already increasing 2 wk p.i. (Figure S10D).
In summary, this study has indicated that GCS expression and sufficient gangliosides in neurons of the adult CNS play a seminal role in the regulation of body weight and energy homeostasis. Analysis of the leptin receptor signaling pathway, being one of the most prominent regulators of CNS metabolic control [35],[36], revealed that GCS-derived gangliosides interact with ObR on the plasma cell membrane, thereby facilitating ObR-dependent signal transduction (Figure 8A). In Ugcgf/f//CamKCreERT2 mice, leptin responsiveness and neuronal function are impaired in hypothalamic neurons involved in the regulation of energy metabolism (Figure 8B). Consequently, defective ObR signaling contributes to the observed metabolic imbalance and obesity development of mice with ganglioside deficiency in the CNS.
Although the seminal role of CNS feedback responses to peripheral energy signals for the regulation of energy homeostasis has been extensively studied, the role of the lipid microenvironment for energy signal receptor function has not yet been addressed. The present study demonstrates that GCS-derived GSLs are critically involved in a to-date unknown mechanism of hypothalamic control of body weight. In line with the finding that neurons of the constitutive Ugcgf/f//NesCre mice do not show increased apoptosis [15], ganglioside-deficient hypothalamic neurons are viable and they show normal membrane and organelle appearance both in vivo and in vitro. Electrophysiological recordings from Arc neurons in slices of Ugcgf/f//CamKCreERT2 mice at 12 wk p.i. did not show a major disruption of membrane functions. However, resting membrane potential and action potential threshold were both shifted to slightly more depolarized values. The molecular mechanism underlying the altered membrane potential remains presently elusive. However, it is well feasible that the shift of threshold is secondary to the slight depolarization, which might inactivate a fraction of Na+ channels. In line with the largely normal properties of neurons from brain slices, biophysical parameters of primary hypothalamic neurons devoid of gangliosides were unaltered. Thus, failure of basic electrophysiological membrane functions is unlikely to cause the observed phenotype of mice with ganglioside deficiency.
Therefore, the present work focuses on interactions of leptin receptors with the ganglioside-containing lipid microevironment in which receptors are embedded. We show with independent methods that two major neuronal GCS-derived gangliosides, GD1a and GM1, closely interact with leptin receptors on the neuronal membrane. This interaction is dynamically enhanced by stimulation with leptin. Both Ugcgf/f//CamKCreERT2 mice and ganglioside-depleted hypothalamic cells display deficient ObR signal transduction upon leptin stimulation, as assessed by decreased leptin-induced Jak phosphorylation, Stat3 phosphorylation, and PIP3 formation. Corroborated in situ by deficient leptin responsiveness in Arc neurons of Ugcgf/f//CamKCreERT2 mice, these results indicate that GCS-derived GSLs, primarily gangliosides, are seminal regulators for neuronal leptin signal transduction. Consequently, Ugcgf/f//CamKCreERT2 mice with deficient leptin-induced hypothalamic neuronal responsiveness develop progressive obesity.
Numerous hypothalamic feedback systems involved in body weight maintenance are known [1],[4]. Admittedly, the robust phenotype of Ugcgf/f//CamKCreERT2 mice may be caused by several peripheral hormones and defective ensuing signaling events occurring in various Cre-targeted CNS regions of this mouse model. The brain stem NTS, though an important mediator of metabolic control [37], is not targeted by activity under the CamK II alpha-dependent Cre recombinase used in this study. Consequently, the NTS shows normal responsiveness to leptin in obese Ugcgf/f//CamKCreERT2 mice and can be excluded to contribute to the observed phenotype. Recent reviews also highlight the LHA as an important regulator of energy balance [38],[39]. In fact, compensating neurocircuits involving nontargeted CNS regions may be considered for the return of food intake from initial hyperphagia to normal levels in obese mice despite the striking increase in orexigenic neuropeptides. Even though X-Gal staining could be seen in parts of the LHA, we could not verify Cre targeting of a major part of ObR-expressing LHA neurons in Ugcgf/f//CamKCreERT2 mice (Figure S7B). In strong support to this line of reasoning, we demonstrate that partial Ugcg replenishment in the Arc mediated by stereotactic injection of rAAV significantly ameliorates obesity and hyperleptinemia in Ugcgf/f//CamKCreERT2 mice. Even though limited infection of closely attached tissue by rAAV injection could not be definitely excluded, mainly Arc neurons were targeted by this approach, as assessed by X-Gal stainings of brains co-injected with LacZ-expressing viruses. The present investigation has thus been restricted to GCS effects focused on the MBH harboring Arc neurons.
Deficient leptin signaling as a consequence of leptin resistance occurs predominantly in the Arc of DIO mice with severe long-term hyperleptimenia [25]–[27],[40]. Socs-3 is a major negative feedback pathway of ObR signaling [41]. Thus, elevated Socs-3 expression levels are found in the hypothalamus of leptin-resistant rodent models [25],[42]. In line with observations in obese db/db mice with nonfunctioning ObR [25],[42], Socs-3 expression in the Arc remains indistinguishable from control littermates in nonobese and obese Ugcgf/f//CamKCreERT2 mice 2, 6, and 12 wk p.i. Elevated hypothalamic ObR expression, as it occurs in DIO mice [29],[30], has also been proposed as a potential mechanism playing a role in the development of leptin resistance [28]. However, normal ObR expression in Ugcgf/f//CamKCreERT2 mice supports the hypothesis that the ObR signaling in their neurons must be deficient due to ganglioside loss and not merely due to secondary leptin resistance. Furthermore, the nontargeted hypothalamic VMH and brain stem NTS retain leptin responsiveness even in obese mice 6 wk p.i. These results in combination with the decreased ObR signal transduction in ganglioside-depleted and non-leptin-resistant N-41 cells strongly suggest that loss of GCS-derived GSLs including gangliosides GD1a and GM1 is the reason for failing ObR activation and subsequently inhibited intracellular signaling.
GCS-depleted Arc neurons display normal leptin sensitivity 1–2 wk p.i., a time point when gangliosides are still present. Furthermore, onset of body weight gain, deficient neuronal activity in the Arc, and abolished ObR signaling coincide with ganglioside depletion 3 wk p.i. This strongly suggests that the mentioned defects are due to ganglioside depletion in these cells rather than due to lack of the enzyme GCS itself.
Further evidence for the postulate that ganglioside deficiency-dependent inhibition of ObR signaling in hypothalamic neurons leads to impaired neuronal function is based on our in situ results in ganglioside-depleted Arc of both obese and nonobese Ugcgf/f//CamKCreERT2 mice. Whereas leptin injection increases c-Fos immunoreactivity and thus neuronal activity in the Arc neurons of fasted lean mice, this response did not occur in GCS-deficient neurons. Leptin specifically engages POMC neurons. Even though the effects of PI3k- and Stat3-dependent signaling in POMC neurons do not overlap [12],[43], both pathways are activated by leptin [11],[12],[44],[45] and contribute to maintenance of energy homeostasis [46]. In ganglioside-depleted POMC neurons, neither PStat3 nor PIP3 formation is increased by peripheral leptin injections, strongly suggesting that defects in both pathways may contribute to partial failure of obesity prevention. As peripheral leptin stimulates both pathways through ObR activation [11],[47], defective ObR function is very likely to be assumed.
In NPY neurons, it has been demonstrated that Jak-Stat3 signaling plays an important role in maintaining NPY/AgRP-mediated energy homeostasis [48]. Additional ObR-mediated PI3k activation seems to be required for inhibiting Npy and Agrp gene expression [8]. Npy and Agrp expression is markedly increased in the MBH of Ugcgf/f//CamKCreERT2 mice, which may be a consequence of absent leptin-induced PStat3 formation in NPY neurons. On the other hand, leptin-induced PIP3 formation does not differ in neither of the groups, which goes in line with the hypothesis that leptin-dependent PIP3 formation in AgRP/NPY neurons is stimulated by an indirect mechanism involving synaptic transmission [12]. Overactive NPY neurons in obese ObR-deficient Leprfa/fa rats were shown to inhibit sympathetic nervous outflow to BAT and cause hypothermia [7] as observed in Ugcgf/f//CamKCreERT2 mice. With regard to the fact that Npy and Agrp but not Pomc expression differ in Ugcgf/f//CamKCreERT2 mice, the role of GCS expression in regulating neuropeptide expression and secretion has to be elucidated. Especially the role of hypothalamic insulin receptor signaling, which also regulates the expression of Pomc and Npy/Agrp in part similar to ObR signaling [11] and is antagonized by GM3 in the periphery [17], constitutes a promising target for further clarifying the differential neuropeptide expression. Moreover, a potential contribution of Cre-targeted ObR-expressing neurons in the median preoptic area of Ugcgf/f//CamKCreERT2 mice to hypothermia may also be considered.
Dynamic membrane microdomains are widely accepted as critical components involved in membrane receptor functions [16],[49]. Since GCS-derived gangliosides are important constituents of these microdomains, they potentially interact with and regulate a variety of membrane components including receptors such as Trk receptors [31] and insulin receptors [17]. In contrast to mice with neuron-specific insulin receptor deletion, which only display a gender- and diet-dependent subtle increase in body weight [11],[50], the obesity and glucose intolerance observed in db/db mice can be rescued by neuron-specific re-expression of ObR [51]. Furthermore, deficient ObR signaling in POMC neurons of the Arc itself leads to the development of mild obesity [52]. In consideration of these findings—despite the existence of potential alternative pathways that might be impaired in neurons of Ugcgf/f//CamKCreERT2 mice—we ascribe ObR and its regulation of activity to a major function in our model pointing to a novel mechanism for CNS metabolic regulation.
We demonstrate that GCS-derived gangliosides GD1a and GM1 closely interact with ObR. The leptin-induced increase in GD1a/ObR and GM1/ObR interaction assumes recruitment of these gangliosides to the ObR upon leptin stimulation. These results in combination with the demonstrated deficient ObR signaling in ganglioside-depleted hypothalamic neurons both in vivo and in vitro leads us to surmise that the lipid microenvironment surrounding the ObR can significantly modulate leptin-dependent intracellular signal transduction in hypothalamic neurons. Altogether, these results provide evidence that GM1 and GD1a are actively involved in enhancing the effects of leptin in hypothalamic neurons.
As insulin receptors contain a lysine residue predicted for interaction with GM3 [53], loss of GM3 synthase showed already a prominent effect on peripheral insulin receptor signaling [17]. It is a widely accepted concept that in the state of insulin resistance in peripheral adipocytes, the IR segregates from caveolae into GM3-enriched microdomains [53], an endogenous inhibitory mechanism [17]. Indeed, elevated GM3 synthase expression could be detected in adipose tissue of obese Zucker fa/fa rats and ob/ob mice [54]. Pharmacologic GCS inhibition in the periphery has been shown to exert beneficial effects on peripheral insulin sensitivity and liver steatosis [55],[56]. With regard to the fact that different ganglioside species can exert either stimulatory [18],[57] or inhibitory [17] effects on membrane receptors, the mentioned studies including the present work support the concept that any perturbation, either loss or excess, of membrane GSLs can alter receptor function. Contributions of GCS-derived lipid raft components apart from gangliosides, namely neutral GSLs in the CNS, to leptin receptor function have yet to be elucidated and constitute a challenging target for future investigations. Besides gangliosides, lactosylceramide has been shown to contribute to formation of lipid microdomains [58]. We, however, propose in the present study that in line with the findings for the insulin receptor, hypothalamic leptin receptor signaling is to a significant extent regulated through interactions with the dominant gangliosides GD1a and GM1.
Recent studies have highlighted the central role of systemic ceramide biosynthesis and GCS in the regulation of energy homeostasis [59],[60]. In accordance with earlier findings [15],[61], we show that neuronal ceramide levels in Ugcgf/f//CamKCreERT2 mice are indistinguishable from control mice, virtually excluding any effects of ceramides.
In conclusion, our study highlights the expression of neuronal GCS-derived GSLs, foremost gangliosides, as a novel class of hypothalamic metabolic regulators. Gangliosides GM1 and GD1a enhance the action of leptin on intracellular signaling and neuronal activity, most likely through dynamic changes of the lipid microenvironment of the ObR. We demonstrate by independent methods that gangliosides GD1a and GM1 strongly interact with the ObR upon leptin stimulation. Loss of these gangliosides leads to impaired responsiveness. By this relevant influence on hormone signaling, Ugcg deletion in adult mouse CNS leads to development of progressive obesity, hyperleptinemia, and glucose intolerance. The obesity can be partially ameliorated by restoration of GCS activity and ganglioside expression in the hypothalamic Arc of Ugcgf/f//CamKCreERT2 mice. Neuronal GCS expression therefore constitutes a novel mechanism for hypothalamic regulation of body weight maintenance.
Animals were kept in specific-pathogen-free barrier facilities. Ugcgf/f mice [15] and inducible CamKCreERT2 mice were bred to generate Ugcgf/f//CamKCreERT2 mice and control littermates. Mice were induced with tamoxifen 6 wk after birth for 1 wk as described [19]. We performed experiments in female mice, unless stated otherwise.
Mice were fasted overnight (o/n). Blood glucose levels were analyzed prior to i.p. injection of glucose (2 g/kg body weight). Glucose levels were determined from tail vein blood (Glucometer Accu Check, Aviva, Roche). Food was withdrawn 4 h prior to the insulin sensitivity assay. Mice were injected i.p. with 0.75 U/kg human insulin (Eli Lilly), and glucose levels were determined as described above (see also Text S1).
Staining was carried out as described earlier [62]. Cryosections (male mice) were incubated with mouse-α-GD1a (1∶100, Millipore) followed by secondary donkey-α-mouse-Alexa-Fluor 488 (1∶200, Invitrogen). Analysis was performed by confocal microscopy (TCS-SL, Leica).
Mice were fasted o/n and injected with leptin (5 mg/kg, Peprotech) or saline between 8.00 a.m. and 10.00 a.m. Animals were sacrificed at indicated time points and transcardially perfused with 4% paraformaldehyde (PFA). We prepared 40 µm cryosections covering the Arc. Alternating sections were collected in series for subsequent free-floating section immunostainings. First antibodies used for immunostaining were rabbit-α-PStat3 (1∶100, Cell Signaling Technology), rabbit-α-c-Fos (1∶100, Santa Cruz), and FITC-conjugated α-PIP3 (1∶100, Echelon). Secondary antibody was donkey-α-rabbit-Alexa-Fluor 488 (1∶200, Invitrogen). Sections were subsequently incubated with either goat-α-NPY (1∶50, Santa Cruz) or sheep-α-MSH (1∶1000, Millipore) followed by secondary antibodies (donkey-α-goat-Alexa-Fluor 546, donkey-α-sheep-Alexa-Fluor 546, 1∶200, Invitrogen). Stainings were analyzed by confocal microscopy (TCS-SL, Leica). Neurons with nuclear (c-Fos, PStat3) or cytoplasmic (PIP3, NPY, and α-MSH) staining above background were considered positive. Immunofluorescence for beta galactosidase was performed as described earlier [63].
The AAV Helper-free System (Agilent Technologies Inc.) was used for preparation of rAAV. Full-length mouse Ugcg cDNA was cloned into the pAAV-MCS vector from the Helper-free System (pAAV-Ugcg). Viruses were generated according to the manufacturer's guidelines and purified as described earlier [64].
Bilateral stereotaxic injections were performed as described [65]. We injected 400 nl virus solution containing equal volumes of viruses carrying Ugcg and lacZ (∼1.8*1011 genome copies/ml) into the Arc of each hemisphere (caudal to bregma: 1.4 mm, 1.44 mm; lateral: 0.25 mm; ventral: 5.7 mm). After surgery, mice were maintained with ad libitum access to lab chow and body weight was monitored weekly.
Immortalized hypothalamic neurons were purchased from CELLutions Biosystems (mHypoE N-41, Cedarlane) and cultured according to the manufacturer's guidelines. GCS was inhibited with NB-DNJ treatment (100 µM, 7 d, Sigma; Tocris).
Eight thousand N-41 cells were seeded onto coverslips and incubated at 37° o/n. The 3 h serum-starved cells were stimulated with leptin (100 ng/ml, Peprotech) for 10 min, washed with PBS and fixed in 4% PFA for 15 min. Cells were blocked with 5% skim milk/PBS. PLA was performed with primary antibodies against ObR (1∶50, Santa Cruz), GD1a (1∶100, Millipore), and GM1 (1∶10, Matreya). PLA was performed according to the manufacturer's guidelines (Duolink Orange Detection System, Olink Biosciences). Formation of PLA spots was analyzed by fluorescence microscopy (Zeiss Cell Observer).
Mice were injected with leptin or saline as described above and sacrificed 30 min later. The MBH was dissected homogenized on ice in lysis buffer (20 mM HEPES, 25 mM KCl, 250 mM sucrose, 2 mM MgCl2, 0.5 mM DTT, 1% digitonin) containing proteinase inhibitor (Roche) and phosphatase inhibitor cocktail (Sigma). Immortalized hypothalamic cells were treated with either saline or 100 µM NB-DNJ for 7 d, serum starved for 4 h, and subsequently treated with either saline or leptin (1,000 ng/ml, Peprotech, 1 h). Cells were lysed on ice in lysis buffer. Protein concentrations were determined by Bradford assay (Sigma). Western blots were performed as described earlier [25]. Primary antibodies: rabbit-α-PStat3, rabbit-α-Stat3, rabbit-α-PJak, rabbit-α-Jak (1∶1,000, Cell Signaling Technology), mouse-α-tubulin (1∶5,000, Zymed Labs), and rabbit-α-actin (1∶1,000, Santa Cruz). Secondary antibodies: HRP-conjugated α-rabbit-IgG (1∶1,000, Dako) and HRP-conjugated α-mouse-IgG (1∶5,000, Santa Cruz). Bands were visualized by chemiluminescence (Amersham) and quantified (ImageJ, NIH).
Four hours serum-starved N-41 cells were treated with leptin (1,000 ng/ml, 25 min). Cells were lysed in IP buffer [50 mM HEPES, pH 7.0, 150 mM NaCl, 10% glycerol, 1% Triton-X, 1.5 mM MgCl2, 1 mM EDTA, proteinase inhibitor cocktail (Roche)]. Co-IP for ObR/Jak was performed as described earlier [36]. ObR/GD1a- and ObR/GM1-Co-IP and subsequent lipid extraction and analysis was performed as described earlier [31],[53]. Anti-ObR were incubated at 4°C o/n. Immunoprecipitated lipids were desalted on an RP-18 column, spotted on a TLC, and run in solvent (chloroform/methanol/0.2% CaCl2; 60∶35∶8, by vol.). GD1a was visualized with mouse α-GD1a (1∶1,000, 4°C overnight, Millipore) on the TLC by immune overlay staining as described earlier [66].
Serum leptin and NEFAs were determined by commercially available kits according to the manufacturer's guideline [Leptin-ELISA (Linco); NEFA-HR2 kit (WAKO Chemicals)]. NEFAs were measured in male mice.
NETO rate in iBAT was determined as described earlier [67]. Tissue NE was measured by reversed-phase HPLC with electrochemical detection (Chrome Systems, Germany) (see also Text S1).
Body weight was measured once a week. Metabolic measurements were carried out in an open circuit respiratory system (SM-MARS, Sable Systems, USA). VO2 and VCO2 per mouse were analyzed for 21 h to determine the RER = VCO2/VO2 and HP (mW). Whole body composition was determined by noninvasive NMR analysis (Mini-Spec, Bruker Optics). Core body temperature was measured with a rectal probe (ALMENO 2390-1, Ahlborn) (see also Text S1).
Total RNA of the MBH was extracted from nonfasted mice as described earlier [68]. RNA was reversely transcribed by Superscript II Reverse Transcriptase (Invitrogen) and cDNA was quantified using the LC FastStart DNA Master SYBR Green I kit (Roche) according to the manufacturer's guidelines and the Light Cycler (Roche) (see also Text S1).
R26R/Ugcgf/+//CamKCreERT2 mice and R26R/Ugcgf/+ mice were induced with tamoxifen i.p. 6 wk after birth as described. At 3 d p.i., animals were sacrificed, and brains were removed and frozen on dry ice. X-Gal staining was performed as described previously [69]. Similarly, β-galactosidase activity in brains of rAAV-Ugcg/LacZ-, rAAV-Empty/LacZ-, and rAAV-LacZ-injected mice was visualized 7 d after virus injection.
GSLs were extracted and separated into neutral and acidic fractions containing gangliosides as described earlier [15]. The amount of GSLs spotted onto a plate by a TLC applicator (Camag, USA) was normalized to tissue protein content determined by the Lowry method [70]. TLC running solvent for acidic GSL was chloroform/methanol/0.2% CaCl2 (45∶45∶10 by vol). GSLs were visualized with 0.2% orcinol in 10% sulphuric acid at 120°C for 10 min. Ceramide was extracted as described earlier [15] and spotted onto a TLC plate. Running solvent for ceramide was chloroform/methanol/acetic acid (190∶9∶1 by vol), and ceramide was visualized with 10% CuSO4 in 8% H3PO4 at 180°C for 10 min. Lipid content was quantified by densitometry (Shimadzu, Japan).
Unless stated elsewhere, results were analyzed by a two-tailed, unpaired Student's t test (Graph Pad Prism, Graph Pad Software, Inc.). To analyze main effects of genotype on metabolizable energy or energy expenditure, body weight was employed as a co-factor in a linear regression model to account for the confounding effect of body size on energy metabolism parameters [71]. p≤0.05 was considered statistically significant and marked *. p≤0.01 was marked **, and p≤0.001 was marked ***.
|
10.1371/journal.pcbi.1003191 | Visual Nonclassical Receptive Field Effects Emerge from Sparse Coding in a Dynamical System | Extensive electrophysiology studies have shown that many V1 simple cells have nonlinear response properties to stimuli within their classical receptive field (CRF) and receive contextual influence from stimuli outside the CRF modulating the cell's response. Models seeking to explain these non-classical receptive field (nCRF) effects in terms of circuit mechanisms, input-output descriptions, or individual visual tasks provide limited insight into the functional significance of these response properties, because they do not connect the full range of nCRF effects to optimal sensory coding strategies. The (population) sparse coding hypothesis conjectures an optimal sensory coding approach where a neural population uses as few active units as possible to represent a stimulus. We demonstrate that a wide variety of nCRF effects are emergent properties of a single sparse coding model implemented in a neurally plausible network structure (requiring no parameter tuning to produce different effects). Specifically, we replicate a wide variety of nCRF electrophysiology experiments (e.g., end-stopping, surround suppression, contrast invariance of orientation tuning, cross-orientation suppression, etc.) on a dynamical system implementing sparse coding, showing that this model produces individual units that reproduce the canonical nCRF effects. Furthermore, when the population diversity of an nCRF effect has also been reported in the literature, we show that this model produces many of the same population characteristics. These results show that the sparse coding hypothesis, when coupled with a biophysically plausible implementation, can provide a unified high-level functional interpretation to many response properties that have generally been viewed through distinct mechanistic or phenomenological models.
| Simple cells in the primary visual cortex (V1) demonstrate many response properties that are either nonlinear or involve response modulations (i.e., stimuli that do not cause a response in isolation alter the cell's response to other stimuli). These non-classical receptive field (nCRF) effects are generally modeled individually and their collective role in biological vision is not well understood. Previous work has shown that classical receptive field (CRF) properties of V1 cells (i.e., the spatial structure of the visual field responsive to stimuli) could be explained by the sparse coding hypothesis, which is an optimal coding model that conjectures a neural population should use the fewest number of cells simultaneously to represent each stimulus. In this paper, we have performed extensive simulated physiology experiments to show that many nCRF response properties are simply emergent effects of a dynamical system implementing this same sparse coding model. These results suggest that rather than representing disparate information processing operations themselves, these nCRF effects could be consequences of an optimal sensory coding strategy that attempts to represent each stimulus most efficiently. This interpretation provides a potentially unifying high-level functional interpretation to many response properties that have generally been viewed through distinct models.
| As we seek to understand how sensory nervous systems process information about their environment, one of the most common quantitative descriptors of neural coding has been the notion of a classical receptive field (CRF) [1]. In general, the CRF is a measurement of the portion of the stimulus space that causes a change in a neuron's response when a stimulus is presented (or removed). For example, beginning with the pioneering work of Hubel and Wiesel [2], simple cells in the primary visual cortex (V1) have been characterized as feature detectors with CRFs that are selective for location, orientation and spatial frequency.
Unfortunately, a simple linear-nonlinear model based on the measured CRF (e.g., linear filtering with the CRF followed by nonlinear thresholding or saturation) is insufficient to explain many response properties of V1 cells. For example, extensive electrophysiology studies have shown that many V1 simple cells also receive contextual influence where stimuli not part of the CRF can modulate the cell's response to CRF stimuli (reviewed in [3]). Furthermore, when driven by rich stimuli within the CRF, simple cells exhibit complex nonlinear response properties that cannot be captured by thresholding or saturation alone [4]. We use the term non-classical receptive field (nCRF) effects to collectively refer to these contextual modulations and nonlinear response properties.
Understanding nCRF effects is likely critical for understanding the coding of natural stimuli because they arise under stimulus conditions that are more complex and ecologically relevant than the stimuli often used in CRF mapping experiments (e.g., sinusoidal gratings, white noise, sparse dots). Indeed, recent electrophysiology experiments with natural video stimuli have shown contextual influence in V1 responses [5]–[8]. Furthermore, observed V1 nCRF effects have been related to perceptual contextual effects such as contour integration [9].
Given the wide range of different nCRF effects reported in the literature, it is still unclear how these effects are related or what collective role they play in sensory coding. Many individual nCRF effects have been successfully described in terms of potential underlying circuit mechanisms (i.e., mechanistic models, reviewed in [10]) or compact stimulus/response descriptions (i.e., phenomenological models, reviewed in [3]). While valuable, these approaches do not fully address the functional significance of nCRF effects or illuminate their role in sensory information processing. In another direction, individual nCRF effects have also been connected to potential benefits in specific tasks (e.g., curvature detection [11], contour integration as reviewed in [12], figure-ground segregation as reviewed in [13]). While these approaches are also valuable, these types of models have limited explanatory power because they only address narrow subsets of biological vision (i.e., individual tasks) and they do not show that the processing strategies represented by nCRF effects are optimal for the given tasks. In short, models constructed for individual effects do not connect this broad range of response properties to the optimal sensory coding strategies that can provide a parsimonious description in terms of fundamental system goals.
One central goal of theoretical and computational biology is to provide functional insight into biological phenomenon by using high-level models (often abstracting away specific experimental detail) to generalize and explain disparate observations. Regarding CRF properties in biological vision, one model that has had success in this regard is the sparse coding hypothesis. Sparse coding conjectures an optimal coding goal where a population of cells encodes a stimulus at a given time using as few active units as possible. Specifically, the model of interest optimizes population sparsity, which is distinct from lifetime sparsity (a single cell being active a small fraction of the time). In seminal results, the high-level sparse coding model (combined with unsupervised learning using the statistics of natural images) has been shown to be sufficient to explain the emergence of V1 CRF shapes both qualitatively [14] and quantitatively [15]. In addition to this success providing functional insight into CRF properties, distributed sparse neural codes have many potential benefits (e.g., explicit information representation and easy decodability at higher processing stages [16], metabolic efficiency [17], increased capacity of associative and sequence memory models [18], [19]) and are consistent with many recent electrophysiology experiments [20].
Despite the success accounting for the emergence of CRF properties, there has been little work showing that sparse coding can account for response properties observed in V1 cells. There have been several recent experimental results showing that stimuli in the CRF surround can cause individual cell responses with higher lifetime sparsity than expected (e.g. [5], [6], [8], reviewed in [21]). While this experimental observation provides encouraging support for the sparse coding hypothesis, it does not imply that a sensory coding model optimizing sparsity is sufficient to account for V1 response properties (including nCRF effects). Sparse coding is one interpretation of the efficient coding hypothesis [22] (conjecturing that neural coding should successively remove stimulus redundancy), and other models related to efficient coding have shown individual model cells that produce some nCRF effects (reviewed in detail in the Discussion section). However, few of these models have shown the broad spectrum of observed nCRF effects in single cells, and none have yet demonstrated the diversity of population response properties reported in the literature for a single effect. Taken together, the evidence of sparsity in experimental observations and the prior success of other related models gives motivation for investigating the potential role of sparse coding in producing nonclassical response properties.
In this paper we demonstrate that a wide variety of nCRF effects are emergent properties of a sparse coding model implemented in a neurally plausible network structure. Specifically, we use the experimental paradigms described in the literature for a wide variety of nCRF effects (e.g., end-stopping, surround suppression, contrast invariance of orientation tuning, cross-orientation suppression, etc.) to replicate these electrophysiology experiments on a dynamical system implementing optimal sparse coding. In the first contribution of this paper, we show that this model produces individual units that reproduce a wide variety of canonical nCRF effects. While another recent model [23] has also shown nearly all of these effects in a unified model along with some increased sparsity of the responses, the present work is the first to show that these effects can arise in a model that has only sparsity as the coding objective. In the second contribution of this paper, when the population diversity of an nCRF effect has been reported in the literature (either through population statistics or multiple individual cells with varying response properties), we also show that this simulated population demonstrates much of the same population heterogeneity reported in the literature. Notably, the results we report are produced with a single set of model parameters (i.e., parameters are not tuned to produce each different effect), despite the system only being designed to optimize sparsity and not constructed to produce nCRF effects. These results show that the sparse coding hypothesis, when coupled with a biophysically plausible implementation, can provide a unified high-level functional interpretation to many population response properties that have generally been viewed through distinct models.
The sparse coding model proposes that V1 encodes an image patch with pixels as approximately a linear superposition of () dictionary elements ,(1)where the coefficients represent the population activity (e.g., average firing rates) [14]. In this model, a neural population encoding the image would calculate activity levels that minimize an energy function that is a weighted combination of a data fidelity term (e.g., mean-squared error) and a sparsity penalty (e.g., the coefficient magnitudes),(2)Here is a system parameter that controls the trade-off between the fidelity of the representation and the sparsity of the coefficients.
The sparse coding model is a functional model that can be implemented through many different mechanisms, including using generic convex optimization algorithms designed for digital computers. In this study we use a dynamical system proposed in [24] that employs neurally plausible computational primitives. Specifically, we implemented the sparse coding model by simulating the dynamical system given by:(3)where is an internal state variable for each node (e.g., membrane potential), is the system time constant, and is an inner product over the spatial dimensions. In the system dynamics, captures the feedforward filtering while captures the recurrent interactions that implement competition between cells to represent the stimulus. Note that the recurrent interaction between those cells is inhibitory if and excitatory if (since in our model). is the soft thresholding function:The input stimulus can be changed dynamically (e.g., a drifting sinusoidal grating), in which case the time-varying coefficients will track approximate solutions, with the solution accuracy determined by the time scale of the input changes relative to the system dynamics. We note that recent theoretical work has demonstrated several network architectures that can efficiently implement other versions of sparse coding with various degrees of biological plausibility [15], [25], [26]. The network architecture being used in this study provably solves the optimization in Eq. (2) with strong convergence guarantees [27], can implement many variations of the sparse coding hypothesis (i.e., different sparsity-inducing cost functions) [28], and is implementable in neuromorphic analog circuits [29].
In our implementation, a dictionary optimized for sparse coding with natural scenes was determined via unsupervised learning under sparsity constraints using whitened natural scenes as the training set (whitening is a first-order approximation of retinal processing). The learned dictionary was overcomplete with effective dictionary elements for the pixel image patches used as stimuli. The training set, whitening and learning rule were all exactly as in [14], while the sparse codes during training (i.e., solutions to (2)) were calculated using a standard software package [30] (for computational efficiency) with . We interpret these dictionary elements as the classical spatial receptive fields (CRFs) of the simulated neurons. This interpretation is supported by our own simulated receptive field mapping experiment (results not shown) using sparse dot stimuli, similar to previous studies (e.g., see Fig. 4b in [14]). The results demonstrated in this study are based on the responses of 72 units in this dictionary that had CRFs well-localized within the available image patch (shown in Fig. 1).
The system parameters described above (i.e., membrane time constant, sparsity level ) are kept the same for every simulation in this paper (details given in Materials and Methods). In other words, no attempt was made to tailor the system to reproduce each effect individually (some interesting exceptions where parameter changes correspond to apparently conflicting results in the literature are shown in the Supporting Information). We interpret the sparse coefficients in Eq. (2) as the trial-averaged instantaneous spike rate of neurons in the model population. To do this, we also impose a positivity constraint and extend the dictionary matrix by including both the original dictionary elements and the negative of the dictionary elements (i.e., doubling the size of the matrix to use the same effective dictionary as if there were both positive and negative coefficients). This mirrored receptive field structure is reminiscent of the push-pull feedforward input structure in the visual simple cells [31].
In the following sections, we highlight several common nCRF effects from the literature and illustrate that this sparse coding model can largely reproduce both reported individual response properties and much of the reported response diversity across V1 neurons. For each nCRF effect the simulation was constructed to match as closely as possible the experimental protocol described in the experimental procedures section of the corresponding electrophysiology paper, including stimulus construction parameters and data analysis (details given in Materials and Methods). We classify the studied nCRF effects into three groups: suppressive effects that are evoked by the presence of stimuli outside the classical receptive field (CRF surround effects), effects where the response modulation depends on the orientation of the stimulus in the surround (CRF surround orientation effects) and effects that reflect the nonlinearity of the CRF center (nonlinear CRF effects).
Stimuli in the region surrounding the CRF can have a modulatory effect on a neuron's response despite not inducing significant response in isolation (by definition of the CRF). In perhaps the simplest form of this suppressive modulation, it has long been known that some V1 neurons exhibit end-stopping where the spike rate decreases for a cell responding to an optimally-oriented bar stimulus when the bar length is increased beyond the CRF boundaries. An example figure depicting the end-stopping effect as observed in cat electrophysiology recordings [32] is reproduced in Fig. 2A. When simulating this experiment [32] on the sparse coding model, some of the model cells (such as the target cell shown in Fig. 2B) exhibit the same characteristic suppression with increasing bar length. The end-stopping effect was previously shown in [33] to emerge in the sparse coding model. The end-stopping effect can be simply understood in terms of the goals of sparse coding. When the bar is short, the CRF of the target cell is the most efficient description of the stimulus and that cell has the strongest response. However, when the bar is long enough that it is better explained by the CRFs of other cells, the target cell becomes suppressed by these competitors so as to maintain a sparse representation. The Discussion section contains a detailed look at how the network interactions supporting the sparse coding model can produce this effect.
Similar to end-stopping, some V1 neurons also exhibit surround suppression where their response to a sinusoidal grating patch decreases as the patch size increases beyond the CRF. Additionally, the tuning curve for patch size often exhibits receptive field expansion at low contrast, meaning that the patch size achieving the maximum response increases at low contrast (Fig. 3A). As illustrated in the response of an example model cell shown in Fig. 3B, the sparse coding model can exhibit the same basic suppression and receptive field expansion properties observed in electrophysiology experiments. In addition, we note that the slight increase of response level (i.e., response rebound) at large stimulus size visible in Fig. 3B is also visible in Fig. 3A and discussed elsewhere [34].
The network interactions giving rise to surround suppression are presumably similar to that of end-stopping, but are more difficult to specify given the added dynamics of the drifting grating stimulus. In particular, due to the suboptimal match of the target CRF to the larger stimulus, competition from other cells (that better match the larger stimulus) can suppress the target cell's response. This competition can also be modulated by the stimulus contrast and may contribute to the receptive field expansion. Specifically, at low contrast the competing cells have lower response levels (resulting in a weaker suppressive effect on the target cell), enabling the response of the target cell to grow with the stimulus size.
Despite the evidence detailed above that some biological and model V1 neurons exhibit surround suppression, a single example cell is insufficient to quantify the prevalence of this effect in a population encoding sensory information. While many nCRF effects are reported as single cell response properties, some studies have attempted to quantify how strongly an effect is expressed across the population. In the case of surround suppression, two metrics have been used to quantify the degree of suppression and receptive field expansion demonstrated by a cell. One is the suppression index (SI), calculated as the ratio between the (suppressed) response value at large stimulus sizes and the peak response value (indicated by arrows in Fig. 3B). The second metric is the RF expansion ratio, calculated as the ratio of the size tuning peak location at high contrast against that at low contrast.
In many physiological studies (both in monkeys [35] and in cats [36]), a large proportion of cells actually show little suppression, with relatively few cells exhibiting strong suppression. An example SI distribution from cat V1 is shown in Fig. 3C, demonstrating a dominant peak at zero suppression and a relatively uniform distribution among more suppressive cells. A similar population distribution emerges from the sparse coding model cells, as illustrated in Fig. 3D. Another characteristic of the surround suppression index is that it is largely invariant to the stimulus contrast. In other words, the difference in SI at high and low contrast is close to zero (Fig. 3E) with a mean value of 0.06 [37]. We also observed this characteristic in the sparse coding model cells (Fig. 3F), with a mean SI difference of 0.02. We note here that some studies (e.g. [38]) recorded unusually high percentage of cells showing significant surround suppression, perhaps due to a different experimental preparation. Interestingly, the sparse coding model can qualitatively reproduce these apparently conflicting results by using a different set of parameters to encourage more sparsity (see Fig. S2 which is described in Supporting Information Text S1).
A scatterplot of RF expansion ratios for V1 cells in macaque [37] shows clearly that on average, the CRF size is larger at low contrast than at high contrast (Fig. 4A). A scatterplot of expansion ratios for the sparse coding model population shows the same qualitative trend of expanding CRF size at low contrast. We note that the mean expansion ratio in the sparse coding model cells (1.16) is lower than typically reported values in the electrophysiology literature (e.g., 2.3 in [37]). This quantitative difference may be due to variations in the RF expansion ratio definitions (e.g., the study in [37] uses a difference of Gaussians fit rather than tuning curve peaks), the lack of contrast saturation in the present model (see Discussions), or biased sampling of neurons in the electrophysiology literature [39]. The possibility that the true expansion ratio might be lower than previously reported is corroborated by a recent study reporting that as many as 40% of cat V1 neurons show length tuning peaks that are invariant to contrast changes [40].
The modulatory effects seen from surround stimulation can depend on a number of stimulus properties, including contrast, spatial extent (relative to the CRF), and stimulus orientation in the surround. In particular, modulation is often most suppressive when the surrounding stimuli are at orientations parallel to the preferred CRF orientation (iso-oriented), and less suppressive (or even facilitatory) when the stimuli are perpendicular to the preferred CRF orientation (ortho-oriented). For example, when stimulating a cell with an optimally oriented sinusoidal grating just covering the CRF (i.e., the orientation eliciting the strongest response), a grating in the annulus surrounding the CRF often suppresses the cell when it is iso-oriented and has little effect when it is ortho-oriented. An example of this surround orientation tuning in macaque V1 cells [41] is shown in Fig. 5A. The sparse coding model cells can also demonstrate the same type of surround orientation tuning, as illustrated by the model cell response shown in Fig. 5B. This tuning behavior in the model is likely due to the difference in the strength of competition with different stimulus surround orientations. In particular, the competing cells stimulated by iso-oriented surrounds are likely to have stronger CRF overlaps with the target cell and therefore induce more competition than the cells stimulated by ortho-oriented surround stimuli.
Orientation tuned surround effects can have substantial variations, even with minor changes in the stimulus. For example, the modulatory effect can be facilitatory at some surround orientations, causing a net increase in the response of the cell to CRF stimulation alone. This facilitatory effect is often seen when using a center stimulus slightly larger than the optimal size [41], as shown in Fig. 5C for the same cell as in Fig. 5A. Interestingly, increasing the size of the center stimulus for a model cell can likewise induce facilitation when the surround stimulus is close to ortho-oriented (shown in Fig. 5D for the same cell as in Fig. 5B).
As with surround suppression, a single example of facilitation in the surround orientation tuning does not characterize the prevalence of this effect in a population of V1 cells encoding a stimulus. The degree of facilitation expressed by a neuron can be characterized by measuring the ratio between the maximum of the surround orientation tuning (the maximum of the solid line in Fig. 5B) and the response to the center at the optimal orientation with no surrounding stimulus (the maximum of the dashed line in Fig. 5B). In macaque V1 [42], the median of the facilitation ratio across the measured population was found to be 1.44 at high contrast and 1.71 at low contrast. The sparse coding model cells show a similar dependency on contrast levels, with the median facilitation ratio ranging from 1.15 at high contrast and 1.31 at low contrast.
The surround orientation tuning properties described above can be substantially influenced by the contrast difference between the center and the surround. For example, if the center contrast is fixed and the surround contrast is varied, the most significant suppression in individual macaque neurons was observed with the iso-orientated stimuli at high surround contrast (see Fig. 6A) [35]. Similarly, when plotting the responses as a function of center contrast for various surround settings (e.g., no surround, iso-oriented, and ortho-oriented), the suppressive effects in macaque were most pronounced with the iso-oriented stimuli at high center contrast (see Fig. 6C) [43]. Both of these dependencies on contrast can also be observed in the sparse coding model cells, as shown in Fig. 6D and Fig. 6B. Again we note that in some physiological studies an apparently conflicting result is reported where cat V1 neurons show facilitation with iso-oriented surround stimuli at low CRF contrast [44] (Fig. S3A which is described in Supporting Information Text S1). Interestingly, the sparse coding model can also reproduce this behavior when using a different set of parameters (see Fig. S3B which is described in Supporting Information Text S1).
Even when the stimulation is confined to the CRF with no involvement of the surround, cells in V1 exhibit several nonlinear effects that cannot be explained by a canonical linear-nonlinear model [4]. One example of such an effect is the contrast invariance of orientation tuning for V1 cells. In a linear-nonlinear model based on CRFs, higher contrast stimuli evoke stronger responses that more readily exceed the spiking threshold, thus broadening the orientation tuning curve for higher contrast stimuli (the “iceberg effect” [45]). However, as reported in the cat physiology literature, the orientation tuning width is largely contrast invariant [46] as demonstrated in Fig. 7A. Cells from the sparse coding model can also display this contrast invariance in the width of their orientation tuning curves, as shown in Fig. 7B. This invariance can potentially be attributed to recurrent inhibition from competing cells at orientations where the target cell is not the most efficient description (e.g., ortho-oriented stimuli). Even though these competing cells may not have large overlap with the CRF of the target cell, as the contrast increases they will become more active and induce stronger inhibition, thereby narrowing the tuning width of the target cell compared to the low-contrast response. Indeed, compared to the predictions of a linear-nonlinear model (not shown), the tuning width from our model is much narrower.
The degree to which the width of the orientation tuning curve changes for a cell can be quantitatively measured by calculating the half-width at half-height of the Gaussian fit to the tuning curve for various contrast levels [47]. The population statistics can be plotted as a histogram tabulating the slope of the best linear fit to the width expansion with contrast for each cell. An example of this measure from ferret V1 demonstrating that the tuning curve width is almost constant with contrast is shown in Fig. 7C [47]. In this same measure, the sparse coding model also exhibits strong contrast invariance properties across the population, as shown in Fig. 7D. Both the ferret V1 population and the sparse coding model have a slope tightly concentrated around zero in these histograms, with mean values of 0.002 and 0.032 respectively. The mean values of the half-width at high contrast measured in physiology () [47] and the model () are also similar.
An example of a nonlinear CRF effect using a more complex stimulus is cross orientation suppression, where a plaid (i.e., an ortho-oriented mask grating superimposed on an iso-oriented test grating) suppresses the response of the cell to the test alone. Fig. 8A and Fig. 8B show examples of this suppressive tuning property from cat V1 [48], as well as from a single cell in the sparse coding model. This kind of facilitatory effect may be due to a number of factors, including excitatory connections between cells (i.e., other cells in the population encouraging the target cell to represent the stimulus when they are unable to do so) or dis-inhibition, where a distant cell inhibits an intermediate cell that subsequently releases an inhibitory effect on the target cell [49].
The degree of cross orientation suppression depends on other factors beyond the orientation of the mask stimulus, including the contrast levels of the test stimulus. This contrast dependency was observed in cat V1 (shown in Fig. 8C) [50], and is also visible in the sparse coding model neurons as shown in Fig. 8D. Note that while the qualitative trends in the contrast dependency are the same in the model and in physiology, the lack of contrast saturation in the present model is evident in this figure (see Discussions).
The degree of cross orientation suppression expressed in a population of cells can be characterized by comparing the response to the plaid with the response to the test alone. A scatter plot of the normalized spike rate of cat V1 cells shown in Fig. 8E for the test versus plaid stimuli demonstrates that most cells have a suppressive response to the plaid (as depicted in the single cell response in Fig. 8A) [51]. Furthermore, the scatter plot indicates that the suppression is more pronounced for lower test contrasts. As shown in Fig. 8F, the sparse coding model population exhibits the same qualitative properties, with most cells exhibiting plaid suppression that increases with lower test contrast. Quantitatively, the mean cross orientation suppression ratio between the test and plaid responses for cat V1 was measured at 0.11 for low test contrast and 0.71 for high test contrast [51]. The sparse coding model cells have mean cross orientation suppression ratios of 0.59 and 0.95 for low and high test contrasts (respectively). While the model shows the same qualitative trend and overlaps in range, the specific values for these ratios are slightly higher than the reported experimental values. This small quantitative discrepancy might be due to the presence of contrast saturation in the physiology (visible in Fig. 8C) and its absence in the sparse coding model (Fig. 8D; (see Discussions).
Electrophysiology research in V1 has revealed a wide variety of nCRF effects that may appear to be due to many different aspects of neural coding or cortical processing. The functional interpretation of these effects is especially complex given the heterogeneity of the responses exhibited across populations of cells reported in the literature. We have demonstrated for a wide variety of nCRF effects that both the canonical individual cell response properties and a substantial diversity of population response properties are emergent characteristics of a simple dynamical system implementing a sparse coding model. This model appears to produce a very good qualitative match to many measures of population response statistics, and in many cases produces quantitative measures of these statistics that are in a similar range to reports in the physiology literature. By demonstrating a coding model that can account for these response properties, these results provide a potential functional insight into the role of nCRF effects in optimal sensory coding. While not mutually exclusive of other functional models that may also play a role in neural coding, the sparse coding model is one of the few models (along with [52]) able to substantially reproduce some nCRF effects as well as account for the emergence of localized, oriented, and frequency-selective CRFs [14]. In particular, despite not being constructed to produce nCRF effects, the present model appears able to capture population properties of nCRF effects that have been difficult for other functional models to produce (e.g. the contrast invariance of surround suppression index in Fig. 3F, as discussed in [53]).
There are several existing results that share a similar goal of providing high-level functional interpretation of nCRF effects. Perhaps most closely related to the present study is the PC/BC model [23], [52], [54]–[56], which has also been able to reproduce most of the nCRF effects demonstrated in this paper [23]. It is interesting to note that although it has other functional goals, the PC/BC model does exhibit high sparsity [52] and has accounted for classic CRF tuning properties [52]. While there is significant overlap in the demonstrated nCRF effects, the present work is unique in exhibiting the sufficiency of a model derived from sparse coding to produce the observed effects and to reproduce the population diversity seen in physiology (which the PC/BC model has yet to demonstrate). Given the similar behavior of the PC/BC model and the present model, it is possible that there is a deeper underlying relationship between the PC/BC model and sparse coding than is presently understood. Other example related works include the basic predictive coding model [57], where a subpopulation of model neurons communicating prediction errors exhibits some of the single cell nCRF effects documented in the present study. Another example is the divisive normalization model [58], where contextual effects emerge from a population interaction that modulates the gain in an attempt to maximize the independence of neighboring units. While both of these models account for some individual effects, they are not currently known to reproduce the population diversity seen in physiology or to alone be sufficient to also account for the emergence of known CRF properties (without an added sparsity constraint). More recent models capture the center-surround homogeneity (e.g. orientation co-alignment) in the natural scenes through a generalized form of divisive normalization [59] or capture the covariance structure between pixels in natural scenes [60]. While each of these models demonstrates some individual nCRF effects, these models are also not currently known to reproduce the population diversity seen in physiology (in particular, [59] simulates responses using a single generic unit and not a diverse population) and neither model currently has a fully specified implementation in a biologically plausible circuit (although an approximate form of the model in [60] may enable such an implementation). Another related model was described in [61], which demonstrated that a spiking input targeted divisive inhibition mechanism gives rise to competition among sensory feature detectors and non-classical-like effects. While this model have some interesting features that the present model does not have (e.g., biologically realistic spiking behavior), the stimuli and CRF representations were 1D idealized functions and it's not clear how the results extend to 2D images.
An important feature of the present work is that the same model (with the same parameters) is used to produce all of the presented results (i.e., parameters were tuned once and fixed for all experiments in the main text). The qualitative and quantitative matches observed in this paper rely on these parameter settings combined with the dynamical system implementation of the sparse coding rule. For example, changes in the system that would actually encourage responses with higher sparsity (e.g., increasing , solving Eq. (2) using a conventional digital algorithm, running the dynamical system implementation with more integration time steps/faster non-biological time constants) would often generate similar single cell nCRF effects [62] as presented here (results not shown), but those effects would be too strong to be a quantitative match to the population properties (e.g., a far higher percentage of model cells would show strong surround suppression than is reported in physiology; see Fig. S1 which is described in Supporting Information Text S1). The Supporting Information demonstrates some instances where simple parameter changes in the model can actually account for apparently conflicting reports regarding nCRF effects in the experimental literature. We speculate that different settings of in the model may reflect differences in experimental preparations, such as different species and various levels of anesthesia. Indeed, anesthesia is known to influence the sparsity level in sensory systems [63], [64], and some perceptual contextual effects only occur in awake animals [65]. These observations about changes in the results with varying sparsity levels indicates that the sparse coding objective appears to be sufficient to produce the nCRF modulations, but the dynamical system implementation (with biophysical time constants) is required to produce the heterogeneity necessary to be a good quantitative fit. We also note that the role the dynamical system plays in the present work is similar to recent work [15] showing that learned dictionaries can be a much better quantitative match with measured macaque CRFs when the sparse coding model is implemented in a neurally-plausible network model. It is presently unclear if a different dynamical system minimizing the sparse coding objective would also result in the heterogeneity necessary to still be a good quantitative fit to physiology. Similar variations in the quantitative fits (especially to population data) are expected when using other sparsity penalties beyond the norm used here [28], or when using sparse coding implementations that encourage more “hard” sparsity (i.e., more elements that are exactly zero) [15]. In a similar vein, the present study uses a four-times overcomplete dictionary optimized for sparsity under natural scenes, and this model component is also likely important to the presented results. Though investigating the role of the dictionary would be an interesting avenue of further exploration, we expect that larger dictionaries may enable more sparse responses which also may demonstrate more suppression than what is seen in the current model.
The recurrent interactions between cells in the sparse coding model implement a rich nonlinear response where cells compete to represent stimulus features. While it has been noted that stimuli in the CRF surround can produce sparse responses [5], [6], [8], the surprising finding of this work is that the particular form of inhibition and excitation necessary to implement a sparse coding model is sufficient to explain so many individual and population nCRF properties. At a high level, these effects likely arise from the present model because the observed responses produce a more efficient representation of the stimulus than alternative population responses. While a detailed investigation of how the network interactions give rise to the response properties is an interesting open question for future investigation, in general this is difficult to determine due to the interactions between the network dynamics and the stimulus dynamics (i.e., the response properties arise from the average response over a drifting grating, in addition to being influenced by network dynamics). In the case of end-stopping, the stimuli is not drifting and we can see more explicitly how this effect arises from the principles of sparse coding. In response to a given fixed stimuli, the steady-state network response is composed of a combination of feedforward excitation, recurrent excitation and recurrent inhibition. When plotting these three components of the steady-state response as a function of the bar length (Fig. 9A), it is evident that the overall response is mostly driven by the feedforward component and the recurrent inhibition. The feedforward excitation saturates as a result of the stimulus growing out of the CRF, but the recurrent inhibition keeps growing with increased bar length. To see the spatial extent of the recurrent influence, Fig. 9B shows the CRF locations and orientations of the cells influencing the target cell. As expected, inhibition mostly comes from cells with overlapping and co-linear CRFs that represent a more efficient description of the stimulus as the bar length increases.
There has been a long history of debate over the mechanisms underlying various nCRF effects [3], with each effect generally having a substantial literature attempting to answer questions about the detailed aspects underlying the modulatory response properties (e.g., the relative role of intra-cortical connections versus feedforward projections from thalamus in contrast invariant orientation tuning [66], as well as the role of feedback connections [67]). The implementation used in this work (see Materials and Methods) would appear to suggest that these contextual effects can emerge from recurrent network structure in the absence of nonlinearities in the thalamic input or feedback from higher cortical areas. However, mechanistic interpretation of functional models must be cautious as there are often many possible mappings of the model to circuitry and biophysical mechanisms. For example, past work has shown that it is possible to have mappings of functional models onto circuitry that are very different from their original intuitive mappings (e.g., divisive normalization [68] and predictive coding [54]). The sparse coding dynamical system used in this study is open to the same variety of mechanistic interpretations. For example, the recurrent inhibitory influences could be implemented [69] via local inhibitory interneurons receiving convergent inputs from local excitatory neurons [70] and having dense (many-to-one) output connections with these excitatory neurons [71]. Alternately, it is possible that these inhibitory influences could be implemented via a mechanism based on long term depression of synaptic connections between excitatory cells in cortical layer 4 [72] and global inhibition [73]. For another example, as demonstrated in [68], it might be possible to achieve similar computational goals through nonlinearities in the feed-forward thalamocortical circuit, rather than a recurrent network. For yet another example, the recurrent competition could be implemented through subtraction as in our model, or through division as in [23]. It remains an open question to determine the most biophysically appropriate mapping of the present model onto a circuit implementation.
While the mechanisms underlying individual nCRF effects is an interesting area of investigation, another related question of interest is to determine which aspects of the model are responsible for the observed population variability. In the present model, the dictionary serves to define both the activity driving each cell through the CRF, as well as determining the synaptic weights that define the recurrent influences in the network dynamics. Because the present dictionary was learned from the sparse coding objective on natural images, it is optimal for this coding strategy and demonstrates significant variability as observed in biological CRFs. While a detailed investigation of how the model gives rise to the response diversity is also a challenging and interesting open question for future investigation, one interesting preliminary question is what role the variability in the dictionary plays in the observed nCRF response variability. As a specific example, we have found the surround suppression index to be significantly anti-correlated with the CRF size (Fig. 10; correlation coefficient ; ). While we are unaware of studies investigating this relation in the physiology literature, there are several studies that do suggest this type of anti-correlation. One piece of evidence [74] shows that cortical layers with larger CRFs also tend to have lower SIs and vice versa. Another corroborating study [75] shows that suppressive V1 cells have smaller CRFs compared to plateaued and facilitative cells. This anti-correlation may be present simply because there are fewer cells with larger CRF size in the model (visible in Fig. 10) and in V1 [76], making these cells more likely to be used in an efficient coding model whenever the stimulus grows past a certain size. It is also possible that the limited stimulus sizes used in the current model and many physiology studies (e.g. [77]) could be producing a boundary effect that contributes to some of these observations. It is presently unclear if the inherent variability in the dictionary is alone sufficient to produce the response variability observed in biology (i.e., if another coding model could produce this same variability when using CRFs from this same type of learned dictionary) or if significant response heterogeneity requires the interaction of a learned dictionary with a dynamical system implementing sparse coding.
Some contextual effects, especially ones that involve perception such as perceptual pop-out, figure ground segregation [13], and contour integration [12] operate over a larger range (e.g. over 8 times the CRF size in [78]) and are likely to be mediated by long-range lateral connections [79]. The present study did not test the emergence of these types of effects in the sparse coding model due to the limited size of the dictionary elements. The sparse coding model simulated here used a substantially overcomplete dictionary (see Materials and Methods), thus the size of the visual field we were able to simulate is limited by the current computational complexity of learning large scale dictionaries from the statistics of natural images. While it may seem unlikely that long-range effects could emerge from the present model when the only direct influences are between cells with overlapping receptive fields (see Materials and Methods), it is conceivable that second order effects (e.g., dis-inhibition, where a distant cell inhibits an intermediate cell that subsequently releases an inhibitory effect on the target cell) may play a central role that would only be discovered in a study using larger visual fields. An alternative is to incorporate long-range lateral connections explicitly into a sparse coding model [80].
Despite the wide variety of nonlinear properties observed in the sparse coding model, this model alone is unable to reproduce some nCRF effects because it lacks the stereotypical saturating contrast response function [81]. While this contrast saturation would be a simple addition to the model, the present study focuses on the basic sparse coding model to isolate the response properties due to the nonlinear interactions required to achieve sparsity. It is interesting to note that the model can still reproduce several contrast dependent contextual effects even without an explicit contrast saturation mechanism. Indeed, it has been previously suggested that some of these contrast dependent effects may be independent of the response saturation [42]. Nevertheless, we expect that including some type of contrast saturation in the model may improve the quantitative fit of the current model to some nCRF effects. For example, introducing contrast saturation in the surround suppression simulation (Fig. 3) may further restrict the size tuning curve peak at high contrast and lead to a closer match to the expansion ratios reported in the physiology literature. Contrast saturation could be included in this model through several mechanisms, including modifying the cost function to encourage saturating spike rates (although by itself this mechanism may not accurately capture saturating membrane potentials [82]), including LGN saturation [54], modifying the network implementation to include contrast-dependent shunting inhibition [4], or coupling the sparse coding model with a model such as the previously reported divisive normalization [58].
To implement sparse coding in a neurally plausible network architecture, we solve the dynamical system in equation (3) using a first order Euler method with an integration time step of ms, 25 integration time steps per stimulus (i.e., corresponding to a stimulus presentation of approximately 1/30 second per frame of a video), a sparsity level of and a membrane time constant of ms (within the range of physiological values between 10 ms and 100 ms [83]). In simulations using static stimuli we measured the response after 1000 integration time steps to assure full convergence.
Stimuli such as bars and sinusoid gratings were generated as pixel image patches, whitened (to mimic retinal processing), and overlaid on a gray background with the same mean as the gratings. Finally, for all stimuli we used a contrast (defined as the range of the intensity values of the sinusoid grating or bar) of 0.3 unless otherwise noted.
As in physiological experiments studying nCRF effects (e.g. [36]), we first picked an arbitrary “target” neuron from the population that we would “record” from, pinpointed the center of its CRF ON-region by hand (interpreting the dictionary element as approximating the CRF), and searched for an optimal circular sinusoidal static grating patch stimulus (i.e., having the size, orientation, spatial frequency, and phase that gave rise to the maximal response of the target neuron in the model). We performed this search by a two-step exhaustive search over the parameter space using the following ranges: size of the grating was between 1 pixel and 16 pixels in diameter using 0.5 pixel increments; orientation was between 0 and 175 degrees using 5 degree increments; spatial frequency was between 0.5 to 2 radians/pixel using 0.25 radians/pixel increments; phase was between 0 to 2 using radian increments. We used this approach to map the optimal stimuli for a total of 72 simulated cells (each with CRFs well-localized within the limited visual field used in the simulation).
In most experiments we used drifting sinusoid gratings as stimuli (as described in the experimental literature for each effect). We simulated a drifting grating in discrete time by a series of static gratings at progressive phases. We fixed the temporal frequency of the grating to be about 3 Hz, which is typical of the preferred frequency of cortical neurons [83]. To simulate the dynamic effect of the neural response, we simulated the dynamical system in equation (3) through the entire experiment with the driving input switched at the appropriate time to match the drift speed of the grating. We measured the response to a full cycle of the grating presentation by the mean or F1 (first harmonic) component, depending on the measure used in physiology literature for the particular effect under consideration.
In the end-stopping experiment we found an optimal static bar stimulus for the target neuron by fixing the bar width to 2 pixels, the orientation to be the same as the optimal sinusoid grating orientation, and the bar length to be the same as the optimal grating size. We then found the optimal bar location by translating the bar around a 5-pixel neighborhood of the grating center and searching for the maximal model response for that cell. After the optimal bar stimulus location was found, we increased its length from 1 to 16 pixels and recorded the steady-state response from the model.
In the surround suppression simulation, we varied the contrast of the sinusoid grating stimuli from 0.05 to 0.5 with increments of 0.1, and we varied the size from 1 to 16 pixels in diameter with an increment of 1 pixel (other parameters were fixed). We measured the spike rate in response to the drifting grating by the F1 component. We defined the surround suppression index as , where represents the peak response across all stimulus sizes at a certain contrast, and represents the minimum response at a radius larger than the peak. Response to high contrast was measured at 0.5 and low contrast at 0.05.
In all orientation tuning studies, we stepped the orientation of the stimulus from 0 to 180 degrees in increments of 5 degrees. We measured the mean spiking response to the drifting grating. When studying the contrast invariance property, we stepped the contrast from 0.1 to 0.5 in increments of 0.1. In the population study of the tuning width, we measured tuning curve half-width at half-height by 1.17 times the standard deviation of the Gaussian fit to the orientation tuning curves. When measuring the slope of half-width vs. contrast, we normalized the contrast to 100 [47]. Five neurons in the simulated population had small unipolar CRFs and therefore showed very little orientation tunings. We could not fit Gaussians successfully to the tuning curves for these neurons, and therefore did not include their orientation tuning properties in the population study.
In the center surround orientation tuning experiment, the surround annulus grating had a thickness of 2 pixels and the center and the surround were phase-locked. When measuring the surround orientation tuning, we fixed the center orientation at the optimal orientation and measured the response to the center alone as well as the center plus the surround. We measured the response measurement for two different center radii: the optimal and the optimal plus one pixel. In the experiment that studied the contrast's effect on the center surround orientation tuning, the center contrast took on values on a logarithmic scale (0, 0.03, 0.06, 0.12, 0.25, 0.5) and we kept the surround contrast constant at 0.5. Similar to the observation in physiology (Fig. 8E), there are many cells with weak response at low contrast in the simulation. Due to the present simulation having more cells than the study in [48], this clustering around zero made the low contrast responses difficult to read when plotted. To better visualize the suppression effect of the plaid for weakly responsive neurons, we plotted the low-contrast population responses with the maximum response normalized to 1 (effectively spreading the points out over the full range to better see their position above or below the diagonal line). High-contrast responses were similarly normalized to plot on the same scale.
|
10.1371/journal.pgen.1002328 | Role of Exonic Variation in Chemokine Receptor Genes on AIDS: CCRL2 F167Y Association with Pneumocystis Pneumonia | Chromosome 3p21–22 harbors two clusters of chemokine receptor genes, several of which serve as major or minor coreceptors of HIV-1. Although the genetic association of CCR5 and CCR2 variants with HIV-1 pathogenesis is well known, the role of variation in other nearby chemokine receptor genes remain unresolved. We genotyped exonic single nucleotide polymorphisms (SNPs) in chemokine receptor genes: CCR3, CCRL2, and CXCR6 (at 3p21) and CCR8 and CX3CR1 (at 3p22), the majority of which were non-synonymous. The individual SNPs were tested for their effects on disease progression and outcomes in five treatment-naïve HIV-1/AIDS natural history cohorts. In addition to the known CCR5 and CCR2 associations, significant associations were identified for CCR3, CCR8, and CCRL2 on progression to AIDS. A multivariate survival analysis pointed to a previously undetected association of a non-conservative amino acid change F167Y in CCRL2 with AIDS progression: 167F is associated with accelerated progression to AIDS (RH = 1.90, P = 0.002, corrected). Further analysis indicated that CCRL2-167F was specifically associated with more rapid development of pneumocystis pneumonia (PCP) (RH = 2.84, 95% CI 1.28–6.31) among four major AIDS–defining conditions. Considering the newly defined role of CCRL2 in lung dendritic cell trafficking, this atypical chemokine receptor may affect PCP through immune regulation and inducing inflammation.
| Human chemokine receptors are cell surface proteins that may be utilized by HIV-1 for entry into host cells. DNA variation in the HIV-1 major coreceptor CCR5 affects HIV-1 infection and progression. This study comprehensively assesses the role of genetic variation of multiple chemokine receptor genes clustered in the chromosome 3p21 and 3p22 on HIV-1 disease outcomes in HIV-1 natural history cohorts. The multivariate survival analyses identified functional variants that altered disease progression rate in CCRL2, CCR3, and CCR8. CCRL2-F167Y affects the rate to AIDS development through a specific protection against pneumocystis pneumonia (PCP), a common AIDS–defining condition. Our study identified this atypical chemokine receptor CCRL2 as a key factor involved in PCP, possibly through inducing inflammation in the lung.
| Functional variation in the human leukocyte antigen (HLA) class I genes and in chemokine receptors affects HIV susceptibility, viral load, and rates of disease progression [1]–[4]. Recent genome-wide association studies (GWAS) performed in HIV-1 cohorts have shown that the HLA region and the chemokine receptor CCR5 gene have major roles in control of HIV-1 replication and disease progression—together they explain approximately 20% of genetic variability [3], [5]–[8] (reviewed in [4], [5]). These findings from GWAS highlighted the leading role of chemokine receptors among non-HLA genes in HIV-1 pathogenesis and prompted us to assess the role of other chemokine receptor genes on HIV disease using a gene-centric approach to identify common or rare functional variants in the region.
The chemokine receptor cluster on chromosome 3 contains at least 12 genes including CCR5, the primary HIV-1 co-receptor [9]–[11]. Multiple genetic variants in chemokine receptors and chemokines have been identified as modifiers of HIV-1 infection or disease progression [12], [13], including CCR5-Δ32 (a 32-bp deletion introduces a premature stop codon) [14] and CCR5 promoter variants [15]–[17] and variants in the CCR5 ligand gene CCL5 [18], [19]. The homozygous CCR5 Δ32/Δ32 genotype and complex heterozygotes with other rare amino acid mutations confers near complete resistance to HIV infection [12], [14], [20]–[22]. Individuals homozygous for a haplotype known as CCR5-P1 [15] or haplogroup HHE [23], a multisite allele of the CCR5 promoter region, progress to AIDS more rapidly than those with other CCR5 promoter haplotypes [15]–[17], [23]–[25]. CCR2 and CXCR6 are minor HIV-1 coreceptors used by a limited number of HIV-1 strains as an entry coreceptor [26], [27]. CCR2-V64I has been associated with delayed progression [12], [25], [28], [29]. Variants in CXCR6 were also associated with disease modification [30], [31].
The chromosome 3 chemokine receptor cluster extends from 3p21 to 3p24, with eight receptors occurring in an 520 kb region of 3p21(Figure 1) [32]. The cluster contains genes for several receptors, CCR3, CCR8, CX3CR1, and CXCR6 that have been shown to bind HIV env or to support varying levels of in vitro replication of HIV-1, HIV-2 or simian immunodeficiency virus (SIV) [26], [33]–[36] (reviewed by [11], [37]). The role played by minor coreceptors in HIV-1 pathogenesis is not clear, but studies have suggested that a broad spectrum of coreceptor usage may be correlated with rapid CD4+ cell depletion and AIDS progression [11], [38], [39]. Primary isolates of HIV-1 have been shown to use a wide spectrum of various chemokine receptors as HIV coreceptors [40]. HIV-1 isolates from a CCR5-Δ32 heterozygous or homozygous individual can use various minor coreceptors such as CCR3, CCR2B, CCR8, CX3CR1 for cell entry [41], [42]; amino acid mutations in the V3 loop of HIV-1 are responsible for utilization of multiple coreceptors [40]. Considering the high mutation rate and sequence heterogeneity of HIV-1, particularly within the env gene, it is plausible that a spectrum of receptors is used in vivo during the course of HIV infection and that genetic variants in the coreceptors may affect usage or binding efficiency by HIV-1. Furthermore, as CCR5 and CXCR4 antagonists blocking these major co-receptors are used therapeutically [37], the potential of HIV-1 to evolve to use other minor coreceptors as alternative cell entry points is expected to increase. Therefore, determining whether HIV-1 minor coreceptor genes, in addition to CCR5 and CXCR4 play a role in HIV pathogenesis is a timely topic.
In this study, we evaluate the impact of chemokine coreceptors (CCR) on HIV/AIDS using a candidate-gene based population association analysis in five treatment-naive HIV-1 natural history cohorts. Genotypes of exonic polymorphisms in chemokine coreceptor genes CCR3, CCRL2 and CXCR6 on Chromosome 3p21 and in CCR8 and CX3CR1 on 3p22 were tested for their genetic influence on AIDS progression. CCR3, CCR8 and CXCR6 were chosen as they are HIV-1 minor coreceptors [26], [33]–[36] (reviewed by [11], [37]). CCRL2 was selected as a candidate gene because of its homology with CCR5 (45%)—the most of any of the chemokine receptors genes—because of its proximity to CCR5, and because it is an atypical receptor without signal transduction, similar to DARC. Our results suggest that genetic variation in CCR3, CCR8 and CCRL2 may contribute additional genetic regulation of HIV-1 disease in addition to that conferred by the major HIV-1 coreceptor gene CCR5.
We resequenced selected chemokine receptor genes in the chromosome 3p21–22 region in 72 African Americans and 72 European Americans with three extreme phenotypes (resistance to HIV infection, very rapid or slow progression to AIDS) to assess the extent of exonic variation and to identify rare variants. We observed a total of 6 exonic variants, 5 of which were nonsynonymous, in CCR3, CCR8, CXCR6, and CCRL2 (Figure 1, Table 1). We did not resequence CCR5 and CCR2 since these receptor genes had been previously sequenced in HIV patients [28], [43], nor did we resequence CX3CR1. We previously showed that CX3CR1-V249I and -T280M had no effect on AIDS progression in our group of seroconverter subjects [44], and therefore did not include them in this analysis. No noticeable differences in SNP frequency among three extreme phenotype groups were observed (data not shown). These SNPs were then genotyped in 5 HIV-1 natural cohorts comprising 2594 European Americans (EA). All SNPs were in Hardy-Weinberg Equilibrium (P>0.05). CXCR6-E3K and CCR3-P39L were rare (<1%) in EA and were excluded from the analysis; other SNPs were common (>5%) (Table 1).
The SNPs considered fell into two distinct haplotype blocks (Figure 1). The larger block in 3p21 comprises CCRL2-F167Y, CCRL2-I243V, CCR5-+/Δ32, the CCR5-promoter SNP CCR5-2459A (rs1799864, previously shown to affect CCR5 expression levels and modify AIDS progression [15], [16]), CCR2-V64I, CCR3-P39L and CCR3-255T/C (Y17Y), spanning 77 Kb, and CXCR6-E3K, 318 Kb teleomeric to CCR3. The second block in 3p22, ∼10 Mb from the first block, consists of CX3CR1-V249I and -T280M and CCR8-A27G (rs2853699), which are separated by ∼160 Kb. There is moderate to substantial LD within the closely spaced CCR3-CCR2-CCR5-CCRL2 group (with pairwise D′ 0.47–1, Figure S1). In the 3p22 block, low levels of LD were observed between CCR8 and CX3CR1 SNPs (D′ ∼0.40). The LD level between 3p21 and 3p22 blocks is minimal.
The identification of independent additional genetic factors in this region, particularly for 3p21, is complicated by a moderate to high level of LD (Figure S1). To detect genes or markers additional to a primary predisposing variant(s) in a genetic region of high LD, stratification analyses or using a restricted dataset are frequently employed [45]. Stepwise regression is also a choice for assessing the relative importance of different variants in the linked region [46]. These approaches are feasible as the SNPs in this region have a low to moderate correlation (for r2, see Figure 2).
SNPs with allele frequencies of at least 5% were analyzed by Kaplan-Meier survival curve analysis and Cox proportional hazards model. We present the relative hazards (RH) of the individual variants on survival to AIDS from analysis of 670 EA seroconverters using multivariable Cox regression models in Table 2. Four of the five SNPs in the 3p21 block revealed new significant associations with delayed time progressing to AIDS, after conditioning on HLA alleles and CCR5-Δ32 (Table 2). Significant associations with differential progression to clinical AIDS were observed for CCR3 -255C (RH = 0.62, P = 0.009, Figure 2A), for CCRL2-243V (RH = 0.66, P = 0.03, Figure 2B), and for CCRL2-167F (RH = 1.89, P = 0.0003, Figure 3A) and for CCR8-27G (RH = 1.44, P = 0.004) (Table 2). Adjusting for potential population stratification had little affect on the significance of the associations (Table 2). Correcting for 7 chemokine receptor variants tested using a Bonferroni step-down (Holm) correction [47], the associations for CCRL2-167Y, CCRL2-243V, and CCR3-255C remained significant (P = 0.002, 0.03, and 0.016, respectively), in addition to CCR5-Δ32 and CCR5-2459, while CCR8-27G became non-significant (Table 2).
Using stepwise selection determined by the Akaike information criteria (AIC), we built an optimal prediction model with the genetic variants that were associated with AIDS progression in this or previous studies (Table 3). From Table 3, we can see the minimum AIC is achieved at the ninth step with nine covariates. Therefore, the best predictive model accounting for the rate of progression to AIDS (1987 CDC definition) includes the newly identified variants CCRL2-167F and -243V, CCR8-27G, as well as previously known variants: the two locus CCR2-V64I-CCR5-Δ32 composite genotype, CCR5-2459, HLA class 1 homozygosity, HLA-B*35Px, HLA-B*57 and HLA B*27 (without covariates, AIC:1796.32, with 9 covariates, AIC:1737.28). The significance level and RH values in this model represent that of each variant after considering all other linked and unlinked variants (Table 3). The rough shrinkage estimate is 0.88 (104/140), indicating that the model is fairly reliable without further shrinkage or data reduction treatment [48]. We calculated that R2 = 0.11 for the final fitted model, indicating that 11% of the variability in AIDS progression can be explained with the variants in the model.
The strongest new association was for CCRL2-Y167F with progression to clinical AIDS. The common 167F allele (allele frequency 67% in EA) had a significant dominant detrimental effect on progression to AIDS-defining conditions (CDC 1987 case definition) (RH = 1.89, P = 0.0003, Wald test; P = 0.0001, Likelihood ratio test; n = 670 seroconverters) in the Cox proportional hazards model (Table 2 and Table 3, Figure 3A). When the analysis was restricted to individuals not carrying CCR5-Δ32 (n = 539), the association of CCRL2-167F with AIDS remained significant (RH = 1.69, 95% CI 1.20–2.38), indicating that the CCRL2-167F association is independent of CCR5-Δ32 and not due to its linkage disequilibrium with the latter.
In an explanatory analysis, we tested whether CCRL2- Y167F affects AIDS progression through a specific AIDS-defining illness. A Cox model analysis in the seroconverter group showed significantly faster progression to PCP for carriers of 167F (RH = 2.84, 95% CI 1.28–6.31, P = 0.007; Figure 3B), but no differential effect on Kaposi's sarcoma, microbacterial avium infection, cytomegalovirus, or lymphoma (P>0.05, data not shown).
To assess whether the CCRL2-167F association with PCP was due to LD (D′ = 0.76, r2 = 0.47) with CCR5-2459 that associated with AIDS progression [16], we restricted the analysis in a group of SC (n = 365) that do not carry the homozygous CCR5-2459 genotype; the CCRL2-167F association with PCP remained significant (OR = 2.42, 95% CI 1.09–5.36), in support of its independent role.
We investigated the impact of genetic variation of 3p21–22 chemokine receptor genes on HIV/AIDS in this population-based association study. We identified a total of 6 exonic variants in CCR3, CCR8, CXCR6 and CCRL2 in 144 samples with extreme phenotypes. Through Cox regression survival analysis of the variants in HIV-1 natural cohorts, we identified three polymorphisms (CCRL2-Y167F, CCR3-255C and CCR8-27G) as having previously unidentified correlation with AIDS progression that appear to confer additional effect beyond the well-studied AIDS-modifying polymorphisms CCR5-Δ32, CCR5-P1/CCR5-2459A, and CCR2-64I. CCRL2-167F was associated with strong accelerated progression to AIDS, resulted almost entirely from rapid development of the AIDS-defining disease PCP.
The coverage of SNPs in the 3p21–22 region is limited in the GWAS SNP chips. Several exonic SNPs (CCRL2-167F, -243V, CCR8-27G, CCR5-2459A, CCR5- Δ32) genotyped in this study were not included in the SNP chips that have been used in GWAS for HIV-1 [3], [5]–[8], highlighting the ongoing need of candidate gene analysis in the GWAS era. The newly observed effects of chemokine coreceptor genes thus await further replication.
It must be cautioned that the combination of the presence of multiple AIDS associations in this chemokine receptor complex, and the LD between the receptor genes, makes determining the true source of the associations difficult. Overall, however, the association analysis points to receptor gene variant associations beyond those that can be attributed to the known AIDS affecting receptor gene polymorphisms. The association of additional chemokine receptors, beyond the primary CCR5 and CXCR4, with AIDS progression is plausible as several of these have been shown to bind to HIV in varying degrees [26], [35], [36], [39]–[42], [53]. The complexity of associations in this region makes it essential to identify a functional effect of the genetic variants on disease, before concluding that the associations are real. It should be noted that we may have not detected all existing SNPs in the region and was also underpowered in detecting rare SNPs (1%). We found that another CCRL2 exonic SNP in the public domain (168M, rs6441977) was not associated with progression to AIDS-87 (RH = 0.84, 95% 0.50–1.39) in our seroconverters. We performed haplotype analysis for the three exonic SNPs in CCRL2 and found that only the haplotypes bearing rs3204849 A (CCRL2 Y167F) and rs3204850 (V243I) were associated with AIDS; the haplotype bearing rs6441977A (V168M) had no effect. No additional information was gained by performing a haplotype analysis compared to single SNP analysis.
With these caveats we argue that the association of the CCRL2-167F variant is worthy of interest. First, within the noted limits of the association analysis, the association is quite strong (RH = 1.9, P = 0.002, corrected), remains significant with CCR5-Δ32 taken into account, and in an automatic selection analysis with all known factors taken into account. Second, although there is no direct demonstration of function for this polymorphism, and the F to Y substitution is generally a conservative one, several lines of phylogenetic, chemical modeling, and indirect experimental data suggest that CCRL2-167Y significantly alters the properties of this receptor.
Protein structure modeling data suggest that the risk-associated F to Y substitution could change the boundary of the transmembrane domain, and introduce a hydrogen bond. Of particular interest is the alignment data of the conservation of this residue among CC chemokine receptors. Strikingly, phenylalanine, the ancestral alternate variant to the AIDS risk tyrosine variant, or another nonpolar amino acid, occurs at this position in all receptors known to be functional [54]; a polar amino acid only occurs in the two cases of the nonsignaling DARC receptor. Further, substitution of tyrosine for phenylalanine at this position in the functional receptor CCR3 reduced migration of HEK293 cells in response to eotaxin threefold. We emphasize that all of these tests are in silico or indirect, and direct test of the effect of the F to Y substitution on the function of CCRL2 remains to be done; new knowledge of the functions and ligands of CCRL2 should make this more straightforward.
The association of CCRL2 with AIDS and PCP is unique as CCRL2 has not been shown to serve as a coreceptor for HIV-1. Our chemotaxis assay experiments excluded 11 common chemokines as ligands of CCRL2. The association does not appear to be a general function of increased HIV susceptibility, but instead specifically attributable to an increase in PCP among individuals carrying this receptor variant. PCP, caused by Pneumocystis jirovecii, is the most common opportunistic infection in untreated HIV-1-infected immunosuppressed persons. PCP is mediated by marked inflammatory responses in lung involving macrophages and chemokines and cytokines [55]. The association of CCL2 with PCP might be attributable to two aspects of CCRL2: immune regulation or direct interaction with HIV. CCRL2 may affect PCP through its immune regulating role at local inflammation sites, possibly by concentrating and presenting chemokines [51], [52], [56]. CCRL2 was rapidly upregulated in murine lung macrophages following inflammation induction [57] and deficiency of CCRL2 impaired lung dendritic cell migration [58]. Alternatively, CCRL2 may influence HIV coreceptors entry through interacting or sequestering with CCL5 or anti-HIV chemokines [56], or may serve as a coreceptor for some strains of HIV-1. The mechanism of CCRL2-F167Y effect on PCP remains to be explored.
In summary, this comprehensive study of the chromosome 3 chemokine receptor cluster region identified multiple genetic variants that associated with HIV disease. The strongest new association appears to result from an increased susceptibility to PCP, rather than from a specific effect on HIV. Added to the existing knowledge of the effect of the chromosome 3 chemokine receptors on HIV disease, which has been already been exploited therapeutically, our results affirm this gene complex as a fertile ground for further research, both for HIV and potentially for a broad range of additional diseases.
Institutional review boards (IRB) at National Cancer Institute, National Institutes of Health and participating institutes approved the study protocols. Written informed consent was obtained from all study participants and/or their legal guardians.
The study group includes 674 HIV-1 seroconverter European Americans, 669 seronegatives enrolled in the following natural history HIV-1 cohort studies: Multicenter AIDS Cohort Study (MACS) [59], the San Francisco City Clinic Cohort Study (SFCCC) [60], AIDS Link to the Intravenous Experience (ALIVE) [61], Hemophilia Growth and Development Study (HGDS) [62], and the Multicenter Hemophilia Cohort Study (MHCS) [63]. Seroconversion date was estimated as the midpoint between the last seronegative and the first seropositive HIV-1 antibody test date (mean interval 0.79 years, range 0.07–3.0 years). The censoring date was the earliest of the date of the last recorded visit, or July 31, 1997 for the ALIVE cohort, or December 31, 1995 for all other cohorts, to avoid the confounding effect of highly active anti-retroviral therapy (HAART). A later censoring date was used for ALIVE cohort because few ALIVE participants received HAART prior to July 31, 1997 [64].
A panel of 72 EA and 72 AA samples representing extreme phenotypes for infection and progression (rapid progression, long term non-progression, and infection resistance) were used for SNP identification. The 5′ and 3′ untranslated (UTR) and coding regions of the CCR3, CCR8, CCRL2, and CXCR6 genes were PCR amplified by overlapping primer sets (Table S2). PCR products were resequenced by BigDye terminator (Applied Biosystems, Foster City, CA). We did not sequence or genotyping SNPs in CCR1, CCXCR1, CCR9 and CCR4 as they are not recognized as HIV-1 coreceptors [26], [33]–[36] (reviewed by [11], [37]).
Genotyping was done using PCR-restriction fragment length polymorphism (RFLP) or TaqMan assays. PCR primer sequences, TaqMan probes and primers, PCR conditions, and restriction enzymes used to genotype each variant are listed in Table S1. Briefly, PCR was carried out with 35 cycles of denaturing at 94°C for 30 s, annealing at 54–60°C for 30 s and extension at 72°C for 45 s. TaqMan assays were performed according to the manufacturer's manual (Applied Biosystems, Foster City, CA). The CX3CR1 variants V249I and T280M were typed as previously reported [65].
Kaplan-Meier survival statistics and the Cox proportional hazards (PH) model (Cox PH model) were used to assess the effects of genetic variants on the time of progression from HIV-1 infection to AIDS (1987 CDC definition) [66], using PROC PHREG and LIFETEST of SAS version 9.13 (SAS, Cary, North Carolina). For SNPs that showed significant association with AIDS development, explanatory analyses were performed for their specific impact on the AIDS-defining diseases Pneumocystis pneumonia (PCP), Kaposi's sarcoma, microbacterial avium infection, and cytomegalovirus. The relative hazard (RH) and significance of associations were determined using a Cox PH model without or with adjustment for confounding genetic factors not on chromosome 3: for EA HLA-B*27 and HLA-B*57, HLA-B*35Px, and HLA Class I homozygosity [1], [67], [68]; for AA CCRL5-In1.1, HLA-B*57 and HLA Class I homozygosity [1], [19], [67]. CCR2-64I, CCR5-Δ32 and CCR5-2459 were also included as covariates in the adjusted multivariable regression analysis. CCR5 promoter haplotypes (P1, P2, and P4) are tagged by SNPs CCR5-2459 (rs1799987) and rs2734648 [43]. A visual inspection of the data with Kaplan-Meier survival curves was performed to determine the genetic models to be used in the Cox PH regression model. A dominant genetic model was tested for all genetic factors in this study, except for CCR5-2459 (recessive) [15], [16]. Participants were stratified by sex and by age at seroconversion: 0–20, >20–40, and over 40 years.
To determine the best explanatory set of genetic variants, while minimizing the number of comparisons in model selection, We used StepAIC procedure to build Cox proportional hazards models for the AIDS-1987 phenotype based on stepwise regression, Akaike information criteria (AIC), and the best subset selection [69]. Here we used PROC PHREG in SAS software with SLENTRY = 0.99 and SLSTAY = 0.995 (values chosen close to 1 to generate a sequence of models from null models to full model ordered by AIC) [48]. Model uncertainty caused by including large number of variables can be estimated by shrinkage (LR-p)/LR, where LR denotes the likelihood ratio χ2 and p denotes the number of the predictors in the final model. A shrinkage below 0.85 raises concern of overfitting [48]. It is recommended that no more than m (the number of uncensored event)/10 predictor degree of freedom p (number of parameters) should be examined to fit a multiple regression model. As in our sample, there were 194 events without censoring, we expect that fitting with <19 variables would be appropriate [48].
To control for potential population stratification, we adjusted the regression analysis with eigenvector values [70]. Eigenvector values were obtained by performing principal component analysis of 700,022 SNPs from a previous GWAS study carried out in the same samples [71]. In the Cox regression analysis, eigenvector values for the top two principle components were included as covariates. The genomic inflation factor in this seroconverter population showed minimal systematic overall bias due to population structure in regards to disease progression phenotype as it was quite close to 1.0 (λ = 1.01) (expected under no population stratification) [71].
We further corrected for multiple comparisons by counting 7 chemokine receptor variants tested using a Bonferroni step-down (Holm) correction method [47], as implemented in the MULTITEST procedure in SAS.
The “generalized” R2 statistic in Cox model is based on the likelihood-ratio statistic (LRT) for testing the global null hypothesis [72]. The formula is given as: R2 = 1−e−(LRT/n), where LRT = −2logL(0)−[−2logL(p)], n is the total sample size, logL(0) is the log-likelihood for a null model with no covariates, and logL(p) is the log-likelihood for the fitted model with p covariates.
We quantified LD between all pairs of biallelic SNPs using the absolute unsigned value of Lewontin's D′ statistic [73]. P values represent significance of departure from the null hypothesis that the pair is in equilibrium. All P values are two-tailed. Haploview was used for the LD plots.
Three-dimensional models of the two CCRL2 proteins with 167F or 167Y were constructed using the PROTINFO structure prediction server (http://www.protinfo.compbio.washington.edu), using the comparative modeling protocol [74], [75]. The detailed modeling method is presented in Text S1. The CCRL2 TM topology was established using ConPred II (http://bioinfo.si.hirosaki-u.ac.jp/~ConPred2/), a predication program based on consensus results of several prediction methods including TMpred, TMAP,TMHMM, HMTOP and MEMSAT [50].
cDNAs coding for the full-length open reading frame (ORF) of human CCRL2 carrying 167Y or 167F were PCR amplified from two individuals with the respective homozygous genotype with proofreading DNA polymerase pfu (Stratage, La Jolla, CA). After confirmation of sequence accuracy, they were ligated into the pcDNA3 (Invitrogen, Gaithersburg, MD). Human embryonic epithelial cells line HEK293 was transfected with the constructs of CCRL2-167F or CCRL2-167Y. Stable transfectants were selected by culturing the cells in 800 µg/ml G418. Once the stable cell lines were established, they were examined for chemotactic responses to chemokines.
Migration of CCRL2-167F and CCRL2-167Y transfected HEK293 cells was assessed using a 48-well microchemotaxis chamber technique. They were examined for chemotactic responses to the following chemokines: CCL5, CCL3, CCL4, CCL2, CCL8, CCL7, CCL11, CXCL12, CCL21, CCL20, and CXCL10. The cells were also tested for migration in response to chemotactic peptides using formayl peptide receptors, including W peptide and MMK-1 (Text S1).
A cDNA coding for 169Y-CCR3 was created from the wild-type human 169F-CCR3 ligated into the pcDNA3, which were used to transfect HEK293 cells (Text S1).
|
10.1371/journal.pgen.1003343 | PRMT4 Is a Novel Coactivator of c-Myb-Dependent Transcription in Haematopoietic Cell Lines | Protein arginine methyltransferase 4 (PRMT4)–dependent methylation of arginine residues in histones and other chromatin-associated proteins plays an important role in the regulation of gene expression. However, the exact mechanism of how PRMT4 activates transcription remains elusive. Here, we identify the chromatin remodeller Mi2α as a novel interaction partner of PRMT4. PRMT4 binds Mi2α and its close relative Mi2β, but not the other components of the repressive Mi2-containing NuRD complex. In the search for the biological role of this interaction, we find that PRMT4 and Mi2α/β interact with the transcription factor c-Myb and cooperatively coactivate c-Myb target gene expression in haematopoietic cell lines. This coactivation requires the methyltransferase and ATPase activity of PRMT4 and Mi2, respectively. Chromatin immunoprecipitation analysis shows that c-Myb target genes are direct transcriptional targets of PRMT4 and Mi2. Knockdown of PRMT4 or Mi2α/β in haematopoietic cells of the erythroid lineage results in diminished transcriptional induction of c-Myb target genes, attenuated cell growth and survival, and deregulated differentiation resembling the effects caused by c-Myb depletion. These findings reveal an important and so far unknown connection between PRMT4 and the chromatin remodeller Mi2 in c-Myb signalling.
| Our manuscript deals with the Protein arginine methyltransferase 4 (PRMT4), which modifies arginine residues in histones and other chromatin-associated proteins and plays an important role in the regulation of gene expression. We addressed the question of how the transcriptional function of PRMT4 might contribute to cell lineage specification despite its ubiquitious expression pattern and how this could explain its involvement in tumorigenesis. As protein associations are likely to provide an answer to this question, we attempted to identify novel interaction partners of PRMT4 using a biochemical approach. By this means, we found that PRMT4 binds Mi2α and its close relative Mi2β. In the search for the biological role of this interaction, we found that PRMT4 and Mi2α/β interact with the transcription factor c-Myb and cooperatively coactivate c-Myb target gene expression in haematopoietic cell lines. Depletion of PRMT4 or Mi2α/β in human erythroleukemia cells resulted in deregulated cell proliferation and differentiation resembling the effects caused by c-Myb depletion. Our findings unravel an important and so far unknown connection between PRMT4 and the chromatin remodeller Mi2 in c-Myb signalling and gene activation and identify both coregulators as attractive targets for leukaemia research and therapy in the future.
| Protein arginine methyltransferases (PRMTs) constitute a family of nine members (PRMT1-9) in mammals, which are characterised by a conserved catalytic domain [1], [2]. They post-translationally mono- and dimethylate arginine residues in proteins using S-adenosylmethionine (SAM) as methyl group donor. Dimethylation can be either asymmetric or symmetric [3]. PRMTs regulate a plethora of cellular functions, including signal transduction, ribosome biogenesis, RNA processing, nucleo-cytoplasmic transport and chromatin-dependent processes, such as DNA repair, imprinting and transcriptional regulation, for which they usually require their catalytic activity. In agreement with their chromatin-related functions, a subgroup of PRMTs methylates histones as well as other chromatin-associated proteins and in this way contributes either to activation or repression of gene expression [4].
PRMT4, also named CARM1 (coactivator associated arginine methyltransferase 1), was the first member linked to transcriptional activation through asymmetric dimethylation of histone H3 at arginine 17 (H3R17me2a) [5]–[7]. Together with other coactivators, such as PRMT1 and the histone acetyltransferase (HAT) CBP/p300, PRMT4 is recruited to specific target genes through interaction with transcription factors, for example p53, NF-κB and nuclear hormone receptors such as the estrogen receptor (ER) [8]–[11]. The hierarchy of sequential coactivator recruitment in ER signalling has been studied in detail revealing that PRMT1-mediated dimethylation of histone H4 at arginine 3 (H4R3me2a) occurs as an early event following hormone treatment and is a prerequisite for promoter hyperacetylation [12], [13]. Subsequent histone acetylation by CBP/p300 facilitates promoter recognition by PRMT4 and methylation of H3R17 [14]. These various histone modifications at promoter-proximal nucleosomes of the target genes coincide with transcriptional activation.
PRMT1- and CBP/p300-mediated histone modifications are required for the subsequent recruitment and enhanced activity of coactivators explaining their direct support of active transcription [12], [14], [15]. Furthermore, histone acetylation is read by Bromo domain-containing proteins, such as the TAFII250 subunit of the TFIID complex, linking this modification directly to pre-initiation complex formation [16]. In the case of PRMT4 and H3R17 methylation, the mechanistic contribution is less clear, although the general relevance of PRMT4 and its catalytic activity in ER-dependent gene activation and embryonic development have been demonstrated in knockout and knockin mice [17], [18]. The recent identification of the Tudor domain-containing protein TDRD3 and the transcription elongation-associated PAF1 complex as readers of methylated H3R17 in the context of ER signalling provides a first hint to how this modification might directly promote transcriptional activation [19], [20]. Besides histone arginine methylation, modification of non-histone proteins plays a similarly important role for the transcriptional function of PRMT4. PRMT4 methylates HATs, such as CBP/p300 and SRC-3, thereby influencing their half-life and capability to interact with other proteins and thus modulating their coactivating function [21]–[24].
Similar to the majority of chromatin modifiers and transcriptional coregulators, PRMT4 seems to exert its functions not as an individual protein, but in close association with interaction partners or within multi-protein complexes. For example, PRMT4 was found in a complex of at least ten proteins, called the nucleosomal methylation coactivator complex (NUMAC), which includes components of the SWI/SNF remodeller complex and coactivates ER-dependent transcription in breast cancer cells [25]. As part of NUMAC PRMT4 acquires the ability to methylate nucleosomal histone H3, whereas recombinant PRMT4 preferentially methylates free H3. Such protein associations are likely to explain how PRMT4 contributes to cell type-specific functions and to cell lineage specification despite its ubiquitious expression pattern [5]. In the early embryo, PRMT4 regulates the development of the inner cell mass and activates expression of pluripotency markers [26], [27], whereas in differentiating skeletal muscle cells PRMT4 is required for the late myogenic transcription programme [28]. Deregulated expression of PRMT4 in certain tissues leads to aberrant transcription and is linked to tumorigenesis, such as high-grade breast tumors [29].
Further knowledge on interaction partners of PRMT4 would be necessary to understand its cell type-specific functions and its contribution to pathogenesis. In an attempt to identify such novel interaction partners of PRMT4 using a biochemical approach, we discovered here the ATP-dependent chromatin remodellers Mi2α and Mi2β, also called CHD3 and CHD4 (chromodomain-helicase-DNA binding protein 3 and 4) respectively, as such candidates. We found that PRMT4 coactivates c-Myb-dependent gene expression together with Mi2α as well as Mi2β in a cooperative manner. PRMT4 and Mi2 simultaneously occupy c-Myb target gene promoters in a c-Myb-dependent fashion and are regulators of cell survival and differentiation of the haematopoietic lineage resembling the function of c-Myb.
To explore the molecular function of PRMT4 in gene regulation we aimed to identify novel interaction partners of the enzyme. We performed gel filtration analysis of protein extracts from several cell lines and detected PRMT4 by Western Blot analysis. In HEK293 and Molt-4 cells endogenous PRMT4 protein formed higher molecular weight complexes than expected from its monomeric or dimeric molecular weight (monomeric MW of PRMT4 = 65 kDa) and peaked in elution fractions of approximately 500 kDa (Figure 1A). Similar results were obtained for overexpressed PRMT4 in HEK293 cells (data not shown). In MCF-7 cells PRMT4 did not peak in the 500 kDa fraction, but significantly eluted with proteins of 100 kDa molecular weight (Figure 1A). These results indicate that PRMT4 stably associates with other proteins in higher molecular weight complexes in a cell type-dependent manner.
In order to purify endogenous PRMT4 with associated proteins we next performed cation and anion exchange chromatography of HEK293 extracts (Figure 1B). The presence of PRMT4 was detected by Western Blot analysis and by methyltransferase assay towards histone H3 in each chromatographic fraction (data not shown). PRMT4 did not bind the cation exchanger phosphocellulose, but eluted from the following anion exchangers DEAE and MonoQ at defined salt concentrations (Figure 1B). Thereby, PRMT4 was separated for example from the PRMT1 enzyme, which eluted from the DEAE column at a higher salt concentration (data not shown). To confirm that the high molecular weight complexes of PRMT4 remained stably associated during the ion exchange chromatography we performed gel filtration analysis after each chromatographic step and detected the presence of PRMT4 by Western Blot analysis, as exemplarily shown for the elution fractions of the MonoQ column (Figure 1C). Using the PRMT4-containing MonoQ fractions we performed affinity purification of the endogenous PRMT4 protein by immunoprecipitation (IP). Both anti-PRMT4 and control IPs were analysed by SDS-PAGE and silver staining (Figure 1D). Silver-stained protein bands specifically detected in the anti-PRMT4 samples were excised and protein identity was determined by mass spectrometry analysis. Among other proteins (not shown), we identified PRMT4 itself and the ATP-dependent chromatin remodeller Mi2α, also called CHD3 (chromodomain-helicase-DNA binding protein 3) (Figure 1D). Components of the NUMAC complex were not identified [25]. This result suggests that Mi2α is a putative interaction partner of PRMT4.
We next analysed the putative interaction between PRMT4 and Mi2α by performing co-immunoprecipitations (co-IP) followed by Western Blot analysis. Immunoprecipitation (IP) of endogenous PRMT4 from HEK293 cells copurified overexpressed Flag-tagged Mi2α (Figure 2A) and reciprocally immunoprecipitates of Flag-Mi2α revealed the presence of endogenous PRMT4 (Figure 2B). IP of Mi2α from the PRMT4-enriched MonoQ fractions using a newly generated anti-Mi2α serum revealed also an interaction between PRMT4 and Mi2α on the endogenous level and supported the mass spectrometrical result (Figure S1). To address the question of whether other PRMTs, such as PRMT1, PRMT3 and PRMT6, are also able to interact with Mi2α we performed GST-pulldown assays. Flag-Mi2α preferentially bound to the GST-fusion of PRMT4 and to a lower extent to GST-PRMT1, but not to the other tested PRMT members (Figure 2C, Figure S2). These results confirm the specificity of this novel interaction and identify PRMT4 as the predominant PRMT to interact with Mi2α.
Given that Mi2α has a close relative, Mi2β/CHD4, we asked whether Mi2β could interact as well with PRMT4. Co-IP analysis revealed an interaction between HA-tagged Mi2β and PRMT4 (Figure 2D). Both Mi2 proteins harbour several conserved functional domains [30], including two PHD fingers, which possess individual histone-binding activities enabling bivalent recognition of two histone H3 tails within nucleosomes [31], [32], two Chromo domains that bind DNA [33] and a SNF2-type ATPase domain. To map the interaction domain of PRMT4 in Mi2 we expressed and radiolabeled Mi2α deletion mutants in an IVT system and performed pulldown experiments with GST-PRMT4. The deletion constructs contained either the N-terminus, the two PHD domains, the two Chromo domains, the helicase domain or the C-terminus (Figure S3). This assay revealed that PRMT4 interacts with the N-terminal region and the Chromo domains of Mi2α (Figure S3). Together, these results show that PRMT4 interacts with both Mi2 proteins and narrow down, as exemplified for Mi2α, the interaction surface of PRMT4 in Mi2.
Mi2α and Mi2β have been reported to be part of the NuRD (nucleosome remodelling and deacetylation) complex and accordingly to function in transcriptional repression, as NuRD provides a physical link between ATP-dependent chromatin remodelling and HDAC (histone deacetylase) activity [34]–[36]. Therefore we investigated whether PRMT4 associates with other subunits of the NuRD complex. In co-IP assays we confirmed the interaction of Mi2α with the NuRD components MBD3 or HDAC1 (Figure S4). However, specific interactions with MBD3 or HDAC1 were not detected in the PRMT4 immunopreciptates. These results indicate that PRMT4 selectively binds both Mi2 proteins, but no other components of the NuRD complex suggesting that the PRMT4-Mi2 interaction might not be linked to NuRD-mediated transcriptional repression.
Apart from their repressive function within the NuRD complex, both Mi2 proteins are also involved in transcriptional activation. For example, human Mi2β is required for T cell development and activation of the CD4 gene [37]. The Drosophila orthologue of Mi2β/CHD4 is localised to actively transcribed regions of polytene chromosomes [38]. Furthermore, Mi2α coactivates c-Myb-mediated transcription independently of its helicase activity [39]. The proto-oncogenic transcription factor c-Myb plays a central role in the proliferation and differentiation of different haematopoitic lineages, in particular of erythrocytes and thymocytes [40], [41], and similarly PRMT4 and Mi2 knockout studies revealed severe defects in early T-cell development [37], [42]. This led us to investigate whether PRMT4 as well as both Mi2 proteins are able to interact with c-Myb. HEK293 cells were transfected with untagged PRMT4 and HA-tagged c-Myb and co-IP assays were performed with antibodies against PRMT4, HA or IgG control. We detected PRMT4 in HA-c-Myb-immunoprecipitates and reciprocally HA-tagged c-Myb in PRMT4-immunoprecipitates (Figure 3A). Both proteins also interacted endogenously in the T lymphocyte cell line Jurkat (Figure 3B). Moreover, pulldown assays using GST-PRMTs and bacterially expressed His-tagged c-Myb revealed a preferential and direct interaction between c-Myb and GST-PRMT4, whereas GST-PRMT1 exhibited a weak interaction and GST-PRMT6 no interaction with c-Myb (Figure 3C). Furthermore, we showed that Flag-tagged Mi2α and HA-tagged c-Myb coimmunoprecipitate (Figure 3D, Figure S5), as previously reported [39]. Additionally, using the same approach we uncovered that also Mi2β was able to interact with c-Myb (Figure 3D). Using protein extracts from Jurkat cells, which reveal high expression levels of PRMT4, c-Myb and Mi2 (data not shown), we validated that Mi2α interacts with c-Myb and PRMT4 also endogenously (Figure 3E). These results identify PRMT4 and both Mi2 proteins as novel interaction partners of the c-Myb transcription factor.
To address whether PRMT4 together with Mi2α and Mi2β regulates the transcriptional activity of c-Myb, we employed the chicken myelomonocytic cell line HD11 that does not endogenously express c-Myb, but is competent to induce endogenous c-Myb target genes, such as Mim-1 and Lysozyme, upon overexpression of c-Myb [43], [44]. Mim-1 is one of the best-characterised Myb target genes and its transcription is strongly upregulated by c-Myb designating the gene an excellent model to study the influence of transcriptional coregulators [45]. We transfected HD11 cells with c-Myb alone or in combination with PRMT4 and Mi2, which both did not affect the expression levels of c-Myb itself (Figure S6), and measured the levels of Mim-1 and Lysozyme transcripts by reverse transcription-quantitative PCR (RT-qPCR). We found that increasing amounts of PRMT4 enhanced the transcript levels of Mim-1 and Lysozyme in a c-Myb-dependent and concentration-dependent manner (Figure 4A, Figure S7). In contrast, overexpression of PRMT1 and PRMT6 did not augment the transcriptional activity of c-Myb, as exemplified for the Mim-1 and Lysozyme gene (Figure 4B, Figure S8), rather PRMT6 repressed the c-Myb-mediated activation in line with its corepressor function [46], [47]. Noticeably, coexpression of PRMT4 and Mi2α further enhanced the transcriptional activity of c-Myb (Figure 4A, Figure S7). The same result was obtained for coexpression of PRMT4 and Mi2β (Figure 4C) suggesting that PRMT4 cooperates with Mi2α and Mi2β in coactivating c-Myb target gene transcription. This effect required the catalytic activity of PRMT4, as overexpression of a methyltransferase-deficient mutant of PRMT4 (VLD) resulted in the loss of coactivation of c-Myb (Figure 4D, Figure S9). Similarly, overexpression of an ATPase-deficient mutant of Mi2α (KA) led to a reduced coactivation in case of Mim-1 and to a loss of coactivation in case of Lysozyme gene expression (Figure 4D, Figure S9). Moreover, for both target genes the cooperativity between PRMT4 and Mi2 was impaired upon transfection of both catalytic mutants. Together, these data indicate that PRMT4 is a novel coactivator of the c-Myb transcription factor and synergises with both Mi2 proteins in a methyltransferase- and helicase-dependent manner to coregulate c-Myb activity.
To analyse whether the effect of PRMT4 and Mi2 on Mim-1 gene activation correlates with their concomitant recruitment to the Mim-1 regulatory regions, in which c-Myb binding sites have been identified [43]–[45], [48], we performed chromatin immunoprecipitation (ChIP). We used HD11 cells stably expressing a doxycycline-inducible c-Myb construct (HD11-C3). In response to doxycycline, the levels of c-Myb protein (Figure 5A) and consistently of Mim-1 transcript (Figure 5B) were increased. ChIP analysis revealed that upon doxycycline treatment c-Myb binds the promoter and to a lower extent the enhancer of the Mim-1 gene, whereas an upstream control region was not occupied by c-Myb (Figure 5C, 5D). Recruitment of both PRMT4 and Mi2 was detected at the Mim-1 promoter as well as enhancer in a c-Myb dependent manner, but not at the upstream control region (Figure 5E, 5F). Therefore the recruitment of the two coactivators did not reflect the binding preference of c-Myb for the promoter. The occurrence of H3R17 methylation (H3R17me2a) exclusively correlated with the binding of PRMT4 at the promoter of the Mim-1 gene (Figure 5G), suggesting that histone H3 is the substrate of PRMT4 preferentially at the promoter but not at the enhancer. Concomitantly with transcriptional induction of Mim-1, a reduction of histone H3 occupancy was detectable at the promoter (Figure 5H). When this decrease in total H3 levels was taken into account for the calculation of H3R17 methylation, its promoter-specific increase was even enhanced (Figure 5I). These findings indicate that the c-Myb-dependent Mim-1 gene is a direct target of PRMT4 and Mi2 and that the two coactivators are concomitantly recruited with c-Myb.
We next investigated whether PRMT4 and Mi2 were also relevant transcriptional coactivators of c-Myb in mammalian cells. Given that c-Myb is a key regulator of haematopoiesis, we decided to use the CML-derived erythroleukemic cell line K562, in which numerous c-Myb target genes have been identified [49], and performed siRNA-mediated depletion of c-Myb, PRMT4, Mi2α or Mi2β. For each knockdown condition, we then analysed the mRNA levels of published c-Myb targets. Depletion of c-Myb, as documented by Western Blot analysis (Figure 6A), led to a decrease in transcript levels of Cdc7, c-Myc, Gata3 and CycB1 (Figure 6B) as previously reported [49], [50]. Noticeably, the transcript levels of these genes were reduced to the same extent after PRMT4 depletion (Figure 6A, 6B). These results indicate that an overlapping set of target genes is regulated by c-Myb and PRMT4 in human haematopoietic cells.
Next we asked whether expression of the same c-Myb target genes is influenced by Mi2α and Mi2β. K562 cells efficiently depleted of either Mi2α or Mi2β subsequent to siRNA transfection (Figure 6C) exhibited in most cases reduced transcript levels of the above identified PRMT4-regulated c-Myb targets (Figure 6D). The mRNA levels of Cdc7, c-Myc and Gata3 were downregulated, most strongly upon Mi2β depletion, whereas knockdown of Mi2α showed a weaker (Cdc7, Gata3) or no effect (c-Myc, Figure 6D). In contrast, transcript levels of CycB1 were up-regulated upon Mi2 depletion, again most strongly upon Mi2β depletion, suggesting that both proteins exert a repressive function in this case. Comparison of the mRNA levels of both Mi2 in wild type K562 cells revealed that Mi2β is predominantly expressed (data not shown), which might explain its stronger effects on gene expression in these cells. Together, these results reveal that PRMT4 and Mi2 influence an overlapping set of c-Myb target genes in human haematopoietic cells. All tested c-Myb targets were activated by PRMT4, whereas Mi2 operated in a gene-specific manner either as an activator or repressor. In case of Cdc7, c-Myc and Gata3 genes, PRMT4 as well as Mi2 enhanced transcription corroborating our findings on a synergism of the two coregulators in c-Myb signalling.
In order to study whether PRMT4 and Mi2 proteins bind the regulatory regions of these c-Myb target genes, we performed ChIP analysis in K562 cells. We found that the promoter regions of Cdc7, c-Myc and CycB1 genes, that contain c-Myb binding sites, were enriched in the immunoprecipitates of c-Myb, PRMT4 as well as Mi2 compared to the IgG control (Figure 6E–6G). In contrast, control regions of the three genes and the β-Tubulin gene promoter, which are free of c-Myb binding sites, were not bound by c-Myb or the two coregulators (Figure 6E–6G). These results indicate that PRMT4 and Mi2 are concomitantly recruited together with c-Myb to target gene promoters and are directly involved in coactivating a subset of c-Myb-dependent genes in human haematopoietic cells.
The c-Myb target genes coregulated by PRMT4 and Mi2 in K562 cells fulfil well-established functions in cell cycle and proliferation control. Given that c-Myb is essential for the self-renewal and proliferative capacity of haematopoietic progenitor cells and suppresses differentiation [49], we next explored the biological significance of PRMT4 and Mi2 for the regulation of c-Myb activity in haematopoiesis. For this purpose we used K562 cells as a model for haematopoietic cell proliferation and differentiation.
First we intended to clarify the role of these novel coactivators in c-Myb-dependent proliferation. In agreement with earlier reports showing that inactivation of c-Myb in K562 cells leads to G2/M arrest [51], we found an increased cell number in the G2/M phase upon siRNA-mediated knockdown of c-Myb compared to control siRNA-transfected K562 cells, as analysed by propidium iodide (PI) FACS (Figure 7A, Figure S10). Furthermore, the number of cells in G1 phase was reduced, while the number of apoptotic cells (sub-G1 peak) was increased in c-Myb-depleted cells, corroborating the pro-proliferative and anti-apoptotic capacity of c-Myb in these cells. Next, we investigated whether depletion of PRMT4 and Mi2α/Mi2β, respectively, affects cell cycle distribution of K562 cells. Similarly, knockdown of the coactivators resulted in a G2/M arrest and in an increased apoptotic rate (Figure 7A, Figure S10). Depletion of Mi2β had a stronger effect on the cell cycle distribution than depletion of the other two coregulators or of c-Myb itself revealing that Mi2β might be implicated in additional c-Myb-independent pro-proliferative functions. To investigate the c-Myb-dependence of the PRMT4 as well as Mi2 function in cell cycle regulation, we depleted the three coactivators in U2OS cells, which clearly expressed lower levels of c-Myb compared to K562 cells, and monitored their cell cycle profile by PI-FACS (Figure S11). We found that PRMT4- or Mi2-depleted U2OS cells revealed no effect on apoptosis or at most a slightly decreased number of apoptotic cells in contrast to the enhanced apoptosis in K562 cells. However, depletion of the coactivators resulted in a decreased number of U2OS cells in G1 phase and in a G2/M arrest similar to our findings in PRMT4- and Mi2-depleted K562 cells suggesting that the apoptotic effects of PRMT4 and Mi2 might be c-Myb-dependent. These data show that PRMT4 and both Mi2 proteins regulate the cell cycle of K562 cells similar to c-Myb and might be functionally relevant coactivators of c-Myb with respect to its apoptotic function in haematopoietic cells.
Transformed cell lines are able to proliferate and form colonies in semi-solid medium. We then asked whether depletion of c-Myb or the coactivators influences this property of K562 cells. Depletion of either c-Myb, PRMT4, Mi2α or Mi2β resulted in reduced numbers of colonies in the methylcellulose colony formation assay compared to control transfected cells (Figure 7B, 7C). Again in this assay, the effect of Mi2β depletion exceeded the effect of c-Myb knockdown hinting at an additional c-Myb-independent role for Mi2β. These results independently confirm a pro-proliferative function of PRMT4 and Mi2 resembling the proliferation-promoting effect of c-Myb.
Finally, we aimed to investigate the potential function of the novel coactivators in the differentiation-suppressive activity of c-Myb. Expression levels of c-Myb are elevated in haematopoietic progenitors of different lineages including the myeloid lineage and decrease during differentiation into the various sub-lineages to allow cell cycle exit and terminal differentiation [49]. K562 cells maintain characteristics of multipotent haematopoietic progenitors and are able to differentiate along the erythroid lineage when treated with hemin [52]. Overexpression of c-Myb has been shown to specifically restrain the chemically induced differentiation of K562 along the erythroid lineage [53]. Therefore, we investigated whether depletion of c-Myb or its coregulators could enhance the hemin-induced differentiation of K562 cells. In the absence of hemin treatment, depletion of c-Myb as well as Mi2β enhanced the spontaneous differentiation into erythrocytes compared to the other siRNA-transfected cells, as quantified by staining of benzidine-positive cells (Figure 7D). Furthermore, the hemin-induced erythroid differentiation was enhanced upon depletion of c-Myb, PRMT4 and Mi2β. A weak increase was also obtained upon depletion of Mi2α (Figure 7D). Together with our findings that PRMT4 and Mi2 interact with c-Myb and coactivate c-Myb-dependent gene expression, these results suggest that PRMT4 and Mi2 are relevant coregulators of c-Myb and might contribute to its differentiation-blocking activity in human erythropoiesis.
In this study we searched for novel chromatin-associated interaction partners of the arginine methyltransferase PRMT4 to extend our understanding of its physiological and pathophysiological roles in transcriptional regulation. We found that PRMT4 forms stable complexes with other proteins as observed by gel filtration analysis, which revealed a cell type-specific elution profile of PRMT4. In HEK293 and Molt-4 extracts, PRMT4 is part of protein complexes that peak around 500 kDa, whereas MCF7 extracts predominantly showed PRMT4 in fractions of 100 kDa in size. Interestingly, the PRMT4-containing NUMAC complex was isolated from hormone-treated MCF7 cells using epitope-tagged PRMT4 [25] indicating that PRMT4 forms dynamic associations not only in a cell type- but also stimulus-dependent manner.
In search for novel interaction partners of PRMT4 we combined ion exchange chromatography of HEK293 protein extracts with endogenous co-IP of PRMT4 and mass spectrometrical analysis. The silver-stain analysis of the separated PRMT4 immunoprecipitates revealed that copurifying protein bands did not occur in stoichiometric ratio with PRMT4. Thus, the purification does not contain a predominant PRMT4-containing complex, rather PRMT4 temporarily and independently interacts with several proteins and these multiple interactions cause the molecular weight shift of PRMT4 in the gel filtration analysis. Components of the NUMAC complex were not identified in the mass spectrometrical analysis, which conforms to the fact that the PRMT4-containing complexes in HEK293 cells are around 500 kD in size, whereas the NUMAC complex possesses a size of approximately 1.5 MDa [25]. This finding additionally underlines that PRMT4 associates with other proteins in a cell type- and stimulus-dependent manner. The mass spectrometry identified Mi2α as a novel interaction partner of PRMT4 and subsequently we found that also its close homologue Mi2β is able to bind PRMT4. Both Mi2 proteins belong to the CHD family of chromatin remodellers containing a tandem Chromo domain and the SNF2-like ATPase domain as signature motifs [30]. They have predominantly been described in the literature as transcriptional repressors, since they are subunits of the NuRD repressor complex [34]–[36]. Several transcription factors, for example NF-κB and KRAB, have been shown to recruit the repressive activity of NuRD to target genes [54], [55]. Nevertheless, the two Mi2 proteins seem to have distinct function, since KRAB is specifically corepressed by NuRD containing Mi2α and not by Mi2β. We excluded the possibility that the PRMT4-Mi2 interaction takes place within the NuRD complex, since PRMT4 did not interact with two other subunits of NuRD and additionally the molecular weight of the PRMT4 complex separated via gel filtration was smaller than the expected size (∼1 MDa) of the NuRD complex [36].
PRMT4 acts as a coactivator of several transcription factors, particularly well-studied for ER, and requires for this function its methyltransferase activity, which marks active gene promoters by H3R17 methylation [7], [18]. We therefore envisaged a potential role of the PRMT4-Mi2 interaction in transcriptional activation, to which Mi2 proteins have also been functionally linked. It was found that Mi2β associates with essential transcription factors of lymphocyte development, such as Ikaros and HEB. Mi2β promotes CD4 gene expression together with HEB and the HAT p300 in developing thymocytes [37], whereas Mi2β and Ikaros antagonise each other's silencing activity and thereby allow transcriptional activation of lymphocyte-specific genes [56], [57]. Mi2α was reported to coactivate transcription mediated by the transcription factor c-Myb [39]. Given that PRMT4, Mi2β and c-Myb consistently revealed defects in the development of specific haematopoietic lineages in knockout models [37], [42], [49], we focussed on the potential functional link of the PRMT4-Mi2 interaction in the context of c-Myb signalling. We found that PRMT4 and c-Myb bind each other and additionally c-Myb associates with both Mi2 proteins. When we studied the impact of PRMT4 and Mi2 in HD11 macrophage cells, a well-established model for c-Myb function, we detected that PRMT4 and Mi2 are recruited to the endogenous target gene Mim-1 in c-Myb-dependent manner and cooperate in activation of c-Myb-mediated transcription. A recent report showed that among the C/EBP transcription factors, which are also known to activate Mim-1, C/EBPβ is antagonised by PRMT4-mediated arginine methylation [58]. However, in our experimental system the coactivating function of PRMT4 is predominant for the Mim-1 gene regulation.
Interestingly, recruitment of PRMT4 and Mi2 was detected at the Mim-1 promoter as well as enhancer in a c-Myb-dependent manner, whereas c-Myb preferentially bound to the promoter. Therefore, the recruitment of the two coactivators did not reflect the binding preference of c-Myb for the promoter, which might be due to c-Myb associations that differ in their composition at the enhancer and the promoter and lead to the masking of the antibody epitope within the c-Myb protein specifically when bound to the enhancer. Furthermore, other factors than c-Myb, with preferential binding to the enhancer, might promote binding of PRMT4 and Mi2 to the enhancer. Another interpretation of this observation is that after 30 hours of doxycycline treatment the chromatin structure has changed. Nucleosome repositioning, which has been described in the regulatory regions of the mim-1 gene [59], might lead to a better access of c-Myb to its binding sites selectively in the promoter region and therefore stronger recruitment of c-Myb to the promoter compared to the enhancer. Although PRMT4 and Mi2 recruitment to the promoter and enhancer depends on c-Myb, directly or indirectly, the coactivators do not necessarily have to follow c-Myb in their binding strength.
For both Mi2 proteins, the isolated N-terminus was shown to activate transcription in reporter gene assays [39], [60], which is interesting, as we mapped the N-terminus together with the Chromo domains in Mi2 to be responsible for the interaction with PRMT4. Given that the tandem Chromo domains were shown to stimulate the ATPase function of Mi2 [61], association of PRMT4 through these domains might influence the remodelling activity. Furthermore, we show here that PRMT4 and Mi2 need their catalytic activity for coactivation of c-Myb, which is contrary to recent findings that Mi2α coactivates c-Myb in a helicase-independent fashion and that solely the repressive function of Mi2α requires the helicase activity [39]. Interestingly, the Mim-1 gene undergoes intensive nucleosome remodelling at its enhancer region upon transcriptional activation in HD11 [59]. Whether nucleosome remodelling of c-Myb target genes is mediated by Mi2 will be subject of future studies. Our findings open up additional interesting questions e.g. of how PRMT4 and Mi2 mechanistically cooperate and whether Mi2 is a substrate of PRMT4.
c-Myb is predominantly expressed in immature haematopoietic cells and is involved in the regulation of proliferation and differentiation of stem cells and progenitor cells of the bone marrow, but also of colon and adult brain [49]. Several c-Myb target genes have been identified in the mammalian system in the past, the majority of which are activated by c-Myb and hint at its cell type-specific functions, i.e. cell cycle progression, differentiation and survival [49], [50]. Therefore we investigated whether PRMT4 and Mi2 are also relevant coactivators of c-Myb in the mammalian haematopoietic system. In these studies we used the CML-derived erythroleukemia cell line K562, since c-Myb has been shown to be functionally relevant in these cells [51]. We found that PRMT4 and Mi2 directly activated c-Myb target gene transcription. Depletion of PRMT4 or Mi2 resulted in deregulation of cell proliferation, apoptosis and erythrocyte differentiation resembling the effects caused by c-Myb depletion. These functional correlations are consistent with the physical interaction of the three proteins and their cooperation in gene expression and suggest for the first time a connection between PRMT4 and the Mi2α/β remodeller in c-Myb signalling and c-Myb-dependent erythropoiesis.
Elevated c-Myb levels due to overexpression or inappropriate activation by structural alterations of the protein sequence lead to a block in differentiation and contribute to the onset of certain human leukaemias, in particular AML, CML and T-ALL [49]. Recently, PRMT4 was identified in an shRNA screen among the group of genes required for disease maintenance in an AML mouse model, in which c-Myb is a critical driver of oncogenesis [62]. These findings together with our observations turn PRMT4 as well as Mi2 into attractive targets for cancer research and therapy in the future.
HEK293, HeLa, Molt-4, MCF-7 and U2OS cells were maintained in Dulbeccos Modified Eagle's Medium (Lonza), while K562 and Jurkat cells were cultured in RPMI 1640 (PAA). Growth medium was supplemented with 10% fetal calf serum (FCS, Invitrogen) and 1% Penicillin/Streptomycin (Lonza). HD11 and HD11-C3 (stably expressing a doxycycline-inducible chicken c-Myb) cells were cultured in Iscove's Medium (Biochrom AG) supplemented with 8% FCS and 2% chicken serum (Sigma). For induction of c-Myb expression, HD11-C3 cells were treated with 1 µg/ml doxycycline (Sigma) for 30 hours.
Transient transfections of HEK293 and HeLa cells with plasmids were performed following a standard CaPO4 protocol. HD11 cells were transiently transfected with plasmids using Fugene HD reagent (Roche). For siRNA transfection of K562 cells, 5×106 cells were electroporated in 300 µl growth medium together with 6 µg siRNA (Dharmacon) at 220 V and 950 µF (Electroporator BioRad) using 4 mm cuvettes and subsequently cultured in growth medium. U2OS cells (1.6×105 cells per 6-well) were transfected with 20 nM siRNA using Lipofectamin RNAiMax (Invitrogen).
Short interfering RNA (siRNA) oligonucleotide duplexes were obtained from Dharmacon or Eurogentec. The siRNA sequences, the plasmids and antibodies used are listed in the Text S1.
For whole-cell extracts, cells were lysed in IPH buffer (50 mM Tris at pH 8, 150 mM NaCl, 5 mM EDTA, 0.5% NP40, 1 mM DTT). To digest the cellular DNA, lysates were subjected to 62.5 units Benzonase (Invitrogen) per mg protein with addition of 7 mM MgCl2 for 40 min at 4°C. For co-immunoprecipitation (co-IP) assay, samples of 0.5–1 mg protein were adjusted to the same volume and to 10 µg/ml ethidium bromide. The IP procedure was performed according to [63].
Gel filtration chromatography for determination of the native size of PRMT4 and ion exchange chromatography for purification of PRMT4 with associated proteins followed by IP and mass spectrometry are described in the Text S1.
GST- and His-tagged proteins were purified from E. coli BL21 according to standard protocols. In vitro transcription and translation (IVT) of Mi2α deletion constructs in the presence of 35S-labelled methionine was performed with TnT T7 Coupled Reticulocyte Lysate System (Promega) according to the manufacturer's protocol. Between 0.1–10 µl of each IVT product and 1 µg GST-/His-tagged proteins were used per pulldown reaction. GST-pulldown experiments were performed as previously described [63]. The reactions were finally separated by SDS-PAGE and proteins were detected either by Western Blot analysis or autoradiography.
Total RNA was isolated using PeqGold total RNA Kit (PeqLab). First strand cDNA was synthesised from 0.5 µg of RNA by incubation with oligodT17 primer and 100 units M-MLV reverse transcriptase (Invitrogen) as described by the manufacturer. For chromatin immunoprecipitation (ChIP) analysis, a 145 cm2 dish of HD11 cells or 1×107 K562 cells were used per IP. The protocol was carried out according to [12] except that chromatin was fragmented by sonication 50×3 sec and 5 sec pause on ice at 30% amplitude (Branson Sonifier W-250-D). cDNA and eluted chromatin were subjected to qPCR analysis in triplicates with gene-specific primers listed in the Text S1. Quantitative PCR was performed using Absolute qPCR SYBR Green Mix (Thermo Scientific) and the Mx3000P real-time detection system (Agilent). Each qPCR reaction was performed in triplicates from the same experiment (technical replicates) and the standard deviation (indicated by error bars) was calculated accordingly. The presented data sets are representative of at least 3 independent experiments (biological replicates). For RT-qPCR, each mRNA expression was normalised to GAPDH mRNA expression. ChIP-qPCR results were expressed as % input.
For quantification of the cell cycle distribution, 1×106 K562 cells were harvested 3 days after siRNA transfection, washed in PBS and fixed in ice-cold ethanol for 30 min. Cells were washed again in PBS and DNA was stained with 54 µM propidium iodide (PI) in the presence of 38 mM sodium citrate and 10 µg DNase-free RNase A (Applichem) in the dark for 30 min at 37°C. Samples were then analysed in a Flow Cytometer FACS Calibur using ModfitLT Mac3 Software and for sub-G1 using CellQuest-Pro software (BD Biosciences). Reproducible and representative data sets are shown.
Three days after transfection of K562 cells with siRNA, 800 cells/100 µl growth medium were seeded in duplicates in 300 µl Methocult M2334 (Stem Cell Biotechnologies) supplemented with 20% RPMI and 1% Pen/Strep in 24-well plates. After 2 days of incubation, colonies were stained with 50 µl 1 mg/ml INT (Iodonitrotetrazolium chloride) for 5 days. Pictures of each well were captured with a binocular microscope (Leica MZ 125) using the Leica DC300 camera. Colonies larger than 0.1 mm were counted. Three independent experiments were analysed for quantification.
Three days after transfection with siRNA, K562 cells were seeded in triplicates at a density of 4×105 cells/ml and were treated for 3 days with 30 µM hemin [64]. Subsequently, 700 µl of the suspension cells were pelleted and washed twice with 0.9% NaCl solution. Then cells were resuspended in 100 µl 0.9% NaCl and 50 µl TMB solution (10 mg 3, 3′, 5, 5′-Tetramethyle-benzidine-dihydrochloride, 1.2 ml acetic acid, 8.8 ml ddH2O, 2% H2O2). After 30 min incubation, 200 µl 0.9% NaCl was added. The number of benzidine-positive cells was estimated by counting triplicates of 300 cells under a microscope using the Neubauer counting chamber.
|
10.1371/journal.ppat.1005654 | Type I Interferons Direct Gammaherpesvirus Host Colonization | Gamma-herpesviruses colonise lymphocytes. Murid Herpesvirus-4 (MuHV-4) infects B cells via epithelial to myeloid to lymphoid transfer. This indirect route entails exposure to host defences, and type I interferons (IFN-I) limit infection while viral evasion promotes it. To understand how IFN-I and its evasion both control infection outcomes, we used Mx1-cre mice to tag floxed viral genomes in IFN-I responding cells. Epithelial-derived MuHV-4 showed low IFN-I exposure, and neither disrupting viral evasion nor blocking IFN-I signalling markedly affected acute viral replication in the lungs. Maximising IFN-I induction with poly(I:C) increased virus tagging in lung macrophages, but the tagged virus spread poorly. Lymphoid-derived MuHV-4 showed contrastingly high IFN-I exposure. This occurred mainly in B cells. IFN-I induction increased tagging without reducing viral loads; disrupting viral evasion caused marked attenuation; and blocking IFN-I signalling opened up new lytic spread between macrophages. Thus, the impact of IFN-I on viral replication was strongly cell type-dependent: epithelial infection induced little response; IFN-I largely suppressed macrophage infection; and viral evasion allowed passage through B cells despite IFN-I responses. As a result, IFN-I and its evasion promoted a switch in infection from acutely lytic in myeloid cells to chronically latent in B cells. Murine cytomegalovirus also showed a capacity to pass through IFN-I-responding cells, arguing that this is a core feature of herpesvirus host colonization.
| Gamma-herpesviruses establish chronic infections and cause cancers. They achieve this by immune evasion. Immune responses nonetheless suppress infection to a degree. By understanding how immune responses and viral evasion come together we can potentially control infection and prevent disease. MuHV-4 provides an accessible model with which to define outcomes. It enters new hosts via epithelial cells, passes to macrophages, then persists in B cells. Type I interferons—a key anti-viral defence—controlled epithelial and B cell infections but poorly restricted the intervening macrophage infection. Therefore for maximal effect interferons must act before B cells are infected.
| The γ-herpesviruses persist in lymphocytes and cause lymphoid and epithelial cancers. MuHV-4, like Epstein-Barr virus (EBV) and the Kaposi's Sarcoma-associated Herpesvirus (KSHV), persists in B cells [1]. After epithelial entry, it reaches B cells in organized lymphoid tissue via dendritic cells (DC) [2]. It then spreads with remarkable precision from splenic marginal zone (MZ) macrophages to MZ B cells, follicular DC, then follicular B cells [3]. Glycoprotein conformation changes guide host colonization, with epithelial-derived virions infecting myeloid cells but not B cells, myeloid-derived virions infecting B cells, and B cell-derived virions infecting epithelial cells [4]. However epithelial-derived virions still infect epithelial cells better than myeloid cells, and myeloid-derived virions still infect epithelial and myeloid cells better than B cells. Therefore efficient B cell colonization must involve also a suppression of non-B cell infections.
Immune cell colonization makes host defences an important feature of the γ-herpesvirus infection landscape. Type 1 interferons (IFN-I) are a core vertebrate anti-viral defence [5]. Myriad stimuli induce IFN-I: classically double-stranded RNA (dsRNA), but also other nucleic acids in unusual forms or places, such as unmethylated and cytoplasmic DNA [6]. IFN-I secretion is triggered by phosphorylation of interferon regulatory factors (IRFs) 3 and 7. Signalling through the STAT-1/2-linked IFN-I receptor (IFNAR) then induces an anti-viral state in infected and surrounding cells via restricted protein synthesis, a reduced apoptosis threshold and immune effector recruitment. IFNα is produced mainly by myeloid cells, IFNβ by many cell types, and both in large amounts by plasmacytoid DC [7]. The multiplicity of induction pathways sensitive to different infection hallmarks ensures that essentially all viruses elicit an IFN-I response.
Most viruses also evade IFN-I [8]. MuHV-4 reduces IFN-I induction in infected cells, inhibiting IRF3 via ORF36 [9] and TANK binding kinase 1 via ORF11 [10]. It also reduces IFN-I signalling, down-regulating STAT-1 and STAT-2 via M2 [11], degrading IFNAR via ORF54 [12] and inhibiting responses downstream of IFNAR via ORF37 [13]. A third interaction is that IFN-I transcriptionally suppresses M2 to inhibit viral reactivation from latency, both directly [14] and by restoring STAT-1/2 expression to allow STAT-1-dependent transcriptional suppression of ORF50, the main viral lytic transactivator [15]. Such effects could explain why IFN-I retains an important role in preventing disease [16, 17]. However MuHV-4 still reactivates sufficiently to cause disease in IFN-I-competent mice lacking CD4+ T cells [18]; and M2 is still made sufficiently to promote acute lymphoid infection [19–21] and provide an important T cell target in long-term infection [22]. Thus, the quantitative relationships between IFN-I signalling and viral evasion remain unclear.
Qualitative questions also remain. ORF11, ORF36, ORF37, ORF54 and M2 all limit IFN-I signalling to infected cells, so ongoing viral gene expression should preclude a cellular IFN-I response; but in cells already exposed to IFN-I, viral signalling blocks should be ineffective and infection should be suppressed. In a uniform cell population such mutual inhibition would manifest as transient viral replication; but in vivo infection is more complicated, with MuHV-4 spreading between cell types and between anatomic sites [23]. Splenic B cell colonization involves at least 5 lytic cycles [3], so virus entry into IFN-I-exposed B cells must be common. For Herpes simplex virus, IFN-I does not prevent epithelial virus production but impaired responses promote pathological neuronal infection [24], suggesting that cell type is an important outcome determinant [25]. Most studies of MuHV-4 and IFN-I have averaged effects across whole organs, so how host response and viral evasion vary with cell type is unclear. To understand this better we tracked MuHV-4 replication in IFN-I responding cells. The results provide new insight into how a complex pathogen interacts with an ancient host defence, and how cell type-specific infection outcomes guide host colonization.
MuHV-4 replicates in a wide range of primary and transformed cells, suggesting that it either induces little IFN-I or evades its effects. Weak inhibition of fibroblast infection by MuHV-4-expressed IFNα [26] suggests effector evasion. However the expression kinetics of IFNα were unclear, and the cells infected (hamster fibroblasts or IFNAR-/- fibroblasts transfected with IFNAR encoding DNA) may have responded poorly to murine IFNα. Myeloid cells play a central role in MuHV-4 host colonization [27], and RAW-264 monocyte-macrophages are well-described IFN-I producers and responders [28]. Therefore we tested in RAW-264 cells whether MuHV-4 induced IFN-I, and whether IFN-I affected viral replication (Fig 1). ELISA of cell supernatants for IFNβ showed only limited induction in infected cultures (Fig 1A), so we tested viral susceptibility to IFN-I signalling more stringently by treating the RAW-264 cells with poly(I:C), a well-characterised TLR3 ligand. IFNβ was then readily detectable regardless of whether virus was present, and viral replication was reduced 10-fold (Fig 1B). However the infectivity of induced RAW-264 cell cultures still increased 100-fold between day(d)1 and d2 after MuHV-4 inoculation (Fig 1B), despite IFNβ being detected at both time points (Fig 1A), and continued to increase at d3. Therefore MuHV-4 both induced little IFNβ and partly resisted its effects.
We tested next how IFN-I affected MuHV-4 replication in the lungs, where it infects alveolar macrophages (AMs) and type 1 alveolar epithelial cells (AEC1s) [29]. Again we wanted to know whether IFN-I was induced, and if induced whether MuHV-4 resisted its effects. IFN-I induction is generally rapid, for example peaking 12-24h after experimental respiratory virus infection [30], but IFNAR-/- mice given MuHV-4 i.n. have normal virus titers in the lungs for at least 4d [16], implying poor IFN-I induction or poor efficacy. Bioassays have suggested poor induction [31], and consistent with this result, IFNβ was not detected by ELISA of lung washes at 1d post-infection (Fig 1C). To test MuHV-4 resistance to IFN-I therefore, we induced it with poly(I:C) 6h before virus inoculation. This gave detectable IFNβ in lung washes (Fig 1C), and increased Mx1 transcription in lungs by d1, whereas infection took until d3 (Fig 1D). Live imaging of luciferase+ MuHV-4 (Fig 1E) and plaque assays of infectious virus (Fig 1F) showed IFN-I induction reducing lung infection at d3 but not at other time points. Although poly(I:C) also induces inflammatory cytokines, these seem not to affect acute MuHV-4 replication [32, 33] and IFNAR blockade abrogated the effect of poly(I:C) (Fig 1G). This implied also that IFN-III induction by poly(I:C) [34] did not have a marked effect. Thus, as with RAW-264 cell infection in vitro, MuHV-4 in the lungs induced little IFN-I, and inducing IFN-I only modestly reduced viral replication.
To track how IFN-I responses and viral replication overlap in single cells, we infected Mx1-cre mice, in which cre is transcribed from an IFN-I-inducible Mx1 promoter [35], with MuHV-4 in which cre switches a fitness-neutral (S1 Fig) reporter construct from mCherry (red) to eGFP expression (green) (MHV-RG) [27] (Fig 2). By typing recovered viruses as mCherry+ or eGFP+ we could determine whether they had passed through a cre+ cell; and as Mx1 induction is essentially specific to IFN-I [36–39], this would tell us whether they had passed through an IFN-I-responding cell. MuHV-4 infected Mx1-cre mice equivalently to non-transgenic littermates (S2 Fig). Viruses recovered from the lungs at d6 after i.n. inoculation under anesthesia showed little fluorochrome switching (Fig 2A); those recovered from mediastinal lymph nodes (MLN) at d10 were significantly more switched; and those recovered from spleens at d14 were more switched again. Upper respiratory tract inoculation similarly gave little virus switching in noses at d5; significantly more in superficial cervical LN (SCLN) at d5; and significantly more again in SCLN at d14. Therefore exposure to IFN-I was associated with lymphoid rather than epithelial infection.
As lymphoid infection follows epithelial infection, greater fluorochrome switching in spleens than in lungs could also have reflected time-dependent IFN-I induction. To analyse spleen infection without a preceding lung infection, we gave MHV-RG to Mx1-cre mice i.p., when it reaches splenic MZ macrophages directly—presumably via the thoracic duct and blood [3]. At d3 after i.p. inoculation, splenic virus was significantly more switched than that recovered from lungs at d3 after i.n. inoculation (Fig 2C). This result suggested that viral IFN-I exposure is intrinsically greater in the spleen. Poly(I:C), which is used widely to activate the Mx1-cre transgene, increased virus switching in each site, with again significantly more switching in spleens. It inhibited viral replication only in lungs (Fig 2D). IFN-I responses should drive viral fluorochrome switching but inhibit replication; viral inhibition of IFN-I production or signalling should preserve replication but limit switching; high level switched virus recovery from spleens implied both IFN-I induction and resistance to its effects.
pDCs are important IFN-I producers in some settings [7]. Depleting them with mAb 120G8 (Fig 2E) had no significant effect on the switching or titer of i.p.-inoculated MHV-RG. The same treatment increased significantly the titer of footpad-inoculated MuHV-4 (S3 Fig), so it was functionally effective. Thus, the IFN-I response to spleen infection did not depend strongly on plasmacytoid DCs.
The M3 promoter used to drive MuHV-4 fluorochrome expression is active in the lytic cycle [40]. It is transcribed in infected lungs and acutely infected lymphoid tissue [41, 42], including flow cytometrically sorted lymphoid and myeloid sub-populations [43], and it has revealed AMs and AEC1 colonization in the lungs [29], and MZ macrophage and B cell colonization in the spleen [3]. Therefore it identifies not only any virus reactivating ex vivo (Fig 2), but also the cells known to be acutely infected in vivo (Fig 3). In infected lungs, >99% of fluorescent cells were unswitched (mCherry+eGFP-). Even when IFN-I was induced with poly(I:C), essentially all the cells infected at d4 were unswitched. Fig 3A shows examples of staining; Fig 3B shows pooled results. At d5, most fluorescent cells were unswitched AEC1s (mCherry+PDP+), but of fluorescent AMs (CD68+ or CD169+) 50% were eGFP+. By d6, all fluorescent cells were again mCherry+. Therefore MuHV-4 initiated lytic gene expression in IFN-I responding AMs, but the subsequent loss of eGFP+ AMs, lack of eGFP spread to AEC1s (Fig 3B) and low switching of recovered virus (Fig 2) argued that these cells poorly supported new virion production. The lack of eGFP+ AEC1s after poly(I:C) treatment suggested that these cells made little Mx1 response, although mCherry+ cells might still produce eGFP+ virions, as switching could occur after viral fluorochrome expression.
Virus tagging is most informative acutely, as the site of tagging is then clearer. Thus to track IFN-I exposure in spleens we gave mice MHV-RG i.p. for direct infection, as in Fig 2C. Fig 3C shows results pooled from multiple sections; Fig 3D shows examples of staining. Fluorescent cells were sparse, but across multiple sections there were significantly more examples of switched than unswitched B cells, and significantly fewer examples of switched than unswitched macrophages. Therefore the switching occurred in B cells, and this virus evidently remained reactivation-competent (Fig 2D).
To test more stringently the capacity of MuHV-4 to resist IFN-I in spleens, we induced it with poly(I:C) (Fig 3E and 3F). Now at d3 50% of fluorescent macrophages were eGFP+ (switched), and at d3-4 >90% of fluorescent B cells were eGFP+. >80% of these were in the MZ and stained for IgM (Fig 3G). As in lungs, the number of switched macrophages declined after d3, and the number of fluorescent B cells declined after d4, consistent with M3 transcription being silenced as splenic colonization shifts to latency in the white pulp (WP) [3]; but the extensive switching of viruses recovered from spleens at d14 after i.n. infection (Fig 2A and 2B) indicated that IFN-I-exposed genomes remained reactivation-competent.
The Mx1-cre transgene has been used extensively to delete floxed cellular genes via poly(I:C) injection. We reasoned that it could also reveal cellular exposure to IFN-I, via activation of a floxed cellular fluorochrome. To this end we crossed Mx1-cre with ROSA26-YFP mice, in which cre-dependent removal of a floxed translational stop activates YFP expression, and infected them with wild-type MuHV-4 (Fig 4). Infected lungs contained few YFP+ cells, even after IFN-I induction with poly(I:C) (Fig 4A). The YFP+ cells were all AMs: none showed the characteristic cytoplasmic extensions of AEC1s and YFP failed to co-localize with PDP, despite AEC1s being the main site of lung infection [29]. Thus, viral fluorochrome switching in the lungs was limited both by poor IFN-I induction, and—as IFNβ was readily detected in lung washes after poly(I:C) treatment (Fig 1C)—also by the lack of Mx1 response made by AEC1s. The Mx1 response of lung cells evident in Fig 1D was presumably made by AMs.
Infected spleens contained contrastingly large numbers of YFP+ macrophages (F4/80+, CD169+) and B cells (B220+) (Fig 4B). However YFP expression in macrophages was not specific to infection, as it was widespread also in naive (and specific pathogen-free) Mx1-cre x ROSA26-YFP mice (Fig 4C). The original description of Mx1-cre mice showed little spontaneous cre expression [35]. However the marker used—floxed DNA polymerase inactivation—might have been subject to negative selection. Our evidence of constitutive Mx1 transcription in macrophages was consistent with the significant role IFN-I plays in normal myeloid cell differentiation [44]. YFP+B220+ B cells were contrastingly uncommon in naive Mx1-cre x ROSA26-YFP mice and increased >5-fold by infection. IgM+YFP+ cell numbers also increased significantly after MuHV-4 infection or poly(I:C) injection (Fig 4D and 4E). Therefore while it was not possible to infer viral exposure to IFN-I in macrophages, cellular fluorochrome switching supported the idea that MuHV-4 is little exposed to IFN-I signalling in AEC1s and abundantly exposed in B cells.
B cell colonization despite viral exposure to IFN-I implied an important role for IFN-I evasion. The MuHV-4 ORF36 inhibits IRF3 signalling, and ORF36- MuHV-4 delivered i.n. shows an IFNAR-dependent infection defect in lungs and spleens [9]. Because spleens are colonized down-stream of lungs, direct and indirect effects on spleen infection are hard to distinguish by i.n. inoculation. Therefore we compared ORF36- MuHV-4 with wild-type also by i.p. inoculation, which reaches the spleen directly [3].
Our ORF36- MuHV-4 showed a relatively minor defect in direct lung infection when given i.n. (Fig 5A), and a marked defect in direct spleen infection when given i.p. (Fig 5B). IFN-I induction with poly(I:C) increased the lung infection defect (Fig 5A). Thus in lungs, limited IFN-I induction normally allows ORF36- MuHV-4 to achieve near wild-type titers; but in spleens, where IFN-I induction was greater, ORF36 was an important outcome determinant. Poor splenic infection by ORF36- MuHV-4 was associated with a shift in viral antigen from MZ (CD169+) to red pulp (RP, F4/80+) macrophages (Fig 5C and 5D). Impaired virus transfer to the WP provided an explanation for the reduction in titer, as lymphoproliferation in the WP normally amplifies the viral load. Virus deposition in the RP was consistent with virions being carried by blood flow from the MZ when transfer to the WP was impaired.
Although MuHV-4 colonizes IFN-I-responding mice, i.n. and i.p. inoculations are more pathogenic in IFNAR-/- mutants [16, 45], so IFN-I must normally exert some restriction on virus replication. Which cell types support the additional infection when IFN-I is lacking has been unclear. Blocking IFN-I signalling with an IFNAR-specific antibody increased acute lung infection <5-fold (Fig 6A), consistent with little IFN-I induction in this site (Fig 1); by contrast spleen increased infection >50-fold. Immunostaining revealed extensive viral lytic spread through the splenic MZ and RP, with increased viral antigen expression in cells morphologically typical of macrophages (Fig 6B). Splenic macrophages are diverse and lack a single unifying marker. Viral antigens were evident in CD169+ (MZ macrophage) (Fig 6C and 6D), CD206+ (tissue resident, non-MZ macrophage) and F4/80+ (RP macrophage) populations (Fig 6E), so IFNAR blockade made many macrophage subtypes more permissive for lytic infection. Viral staining in the WP remained low (Fig 6D). Thus in the absence of IFN-I, MZ to WP transfer became rate-limiting for virus spread, and most infection was diverted to the RP, again consistent with untransferred MZ virus following splenic blood flow.
We tracked latent infection in spleens by viral eGFP expression from a constitutive promoter (Fig 6F). Again blocking IFNAR increased infectious virus titers at d4. Total recoverable virus (infectious centre assay) increased similarly to infectious virus, consistent with most early spleen infection being lytic in MZ macrophages [46]. By d7 after i.p. challenge, when spleen infection is largely latent in B cells [46], virus titers remained elevated above controls in IFNAR-blocked mice; however the elevation was no greater than at d4 (Fig 6F), implying that IFN-I limited mainly MZ macrophage infection. Staining spleen sections at d7 (Fig 6G and 6H) showed more WP infection (eGFP+) in IFNAR-blocked mice than in controls, so more lytic infection in MZ macrophages eventually fed through to more latent infection in WP B cells. However IFN-I blockade did not increase viral antigen+/ viral eGFP+ cell ratios in the WP at d7—in fact they were reduced—so the proportion of lytic infection in WP B cells did not increase. We conclude that splenic macrophage infection was strongly restricted by IFN-I, but that acute viral reactivation in WP B cells was largely IFN-I-independent.
IFN-I principally targets viral lytic infection. Therefore it possibly had least effect on B cell infection because this is mostly latent. MuHV-4 that cannot shut down lytic infection due to an additional promoter element inserted upstream of its ORF50 lytic switch gene (M50) replicates normally in mice for 3d, but is then progressively attenuated [47]. This virus cannot drive B cell proliferation because it is constitutively lytic, but it is attenuated also in the lungs, where AEC1s and alveolar macrophages are infected. That IFNAR deficiency does not significantly increase MuHV-4 lung infection for at least 4d [16] suggested that M50 virus attenuation might be due to impaired IFN-I evasion. To test this hypothesis we gave mice IFNAR blocking antibody i.p. then wildtype or M50 MuHV-4 i.n. (105 p.f.u.). Lung virus titers after 7d showed restoration of the M50 replication defect by IFN-I blockade (S4 Fig). Thus, a capacity to establish latency was important for IFN-I evasion.
MuHV-4 passage through IFN-I responding cells was surprising: we expected that either viral evasion would prevent an IFN-I response in infected cells, or IFN-I would prevent viral replication. To test whether it was a unique to MuHV-4, we infected Mx1-cre mice with murine cytomegalovirus that cre switches from eGFP to tdTomato expression (MCMV-GR). We induced IFN-I or not with poly(I:C), gave MCMV-GR i.p., and recovered infectious virus from livers and spleens by plaque assay at d3 (Fig 7A). As with MHV-RG, IFN-I induction with poly(I:C) increased MCMV-GR switching (Fig 7B) without reducing titers. Thus MCMV, which like MuHV-4 inhibits IFN-I induction and signalling [48], could also pass through IFN-I-responding cells. Spleen sections (Fig 7C and 7D) revealed eGFP-tdTomato+ cells around lymphoid follicles, consistent with MCMV infecting IFN-I-responding MZ macrophages [49]. Liver sections of (Fig 7D and 7E) showed CD68-tdTomato+ cells with the morphology of hepatocytes. Infected cells were more switched than recovered virions, consistent with IFN-I exerting some restriction on virion production [50], but productive infection was clearly possible in IFN-I-responding cells.
MuHV-4 provides an experimental window onto the γ-herpesviruses, whose colonization of lymphoid tissue directly confronts host immune defences. IFN-I attenuated macrophage infection, whereas B cell infection was protected by viral evasion. This cell type-dependent outcome explained how IFN-I and its evasion both control infection, with each dominating in a different setting (S5 Fig). Together they played a significant role in shifting the focus of viral tropism from macrophages acutely to B cells chronically.
In the lungs, AMs provide MuHV-4 with a gateway to AEC1s, which then support lytic replication [29]. AEC1s made no detectable Mx1 response to infection or to poly(I:C). Thus, although IFN-I responding AMs poorly supported virus spread, inducing or blocking IFN-I or disrupting its evasion all had modest effects because AEC1s still allowed virus replication. Splenic MZ macrophages provide a gateway to MZ B cells. Again IFN-I restricted virus spread from macrophages. Unlike AEC1s, B cells made Mx1 responses, but their IFN-I responses were bypassed by viral evasion.
An important role for the ORF36 IFN-I evasion gene in splenic infection was surprising, as its inhibition of IRF3 should limit IFN-I induction, and the abundant exposure of wild-type splenic virus to IFN-I implied that MuHV-4 evades its effector functions rather than relying solely on limiting induction. A possible explanation is that ORF36, as a lytic gene, operates more in macrophages than in B cells, as macrophages support lytic infection [43] and are important IFN-I producers, whereas B cells support a more tightly latent infection and function more prominently as IFN-I responders. IFN-I production by macrophages depends on positive feedback through IFNAR [51]. Thus ORF36, by limiting IFN-I production, should limit IFN-I signalling by infected macrophages both to themselves and to B cells. This would promote macrophage to B cell virus transfer: first, by promoting new virion production in macrophages; and second, by reducing the effector evasion required for those virions to productively enter B cells. The importance for splenic infection of IFNAR degradation by ORF54 [12] suggests that it may play a similar role. Plasmacytoid DC produce IFN-I independently of feedback through IFNAR [7], so their production would not be affected by ORF36 or ORF54. They (B220+ DC) are not an acute infection target in spleens [3]. However they contributed relatively little to virus switching or control, so other evasion mechanisms may limit their scope.
The recovery of fluorochrome-switched virions implied that viral IFN-I evasion also operates down-stream of IFN-I induction or signalling, that is downstream of ORF36. Switching was evident early after i.p. virus inoculation, when splenic infection is mostly lytic [46]. Established lytic infection would inhibit IFN-I signalling, so we envisage that most switching occurred when virions entered cells already making IFN-I responses. In this context the disassembly of IFN-I induced ND10 domains [52] may be an important evasion mechanism. This is a function of the MuHV-4 ORF75c tegument component [53, 54] and is conserved in the homologous Kaposi's Sarcoma-associated Herpesvirus ORF75 [55] and Epstein-Barr virus BNRF1 [56]. Tagged viral genomes could then remain latent until IFN-I responses have subsided [14]. Therefore, pace cell differences in IFN-I response, macrophage infection may be more susceptible than B cell infection to inhibition by IFN-I because it is more lytic [43, 46].
Together the IFN-I response and its evasion promoted B cell over macrophage infection, making IFN-I an important determinant of in vivo viral tropism. A corollary is that IFN-I-based therapies are likely to have only a small window of efficacy—mainly reducing acute gamma-herpesvirus replication in myeloid cells. Downstream of this, virions entering IFN-I-exposed B cells would still establish a viable infection. IFN-I deficiency increases ex vivo MuHV-4 reactivation rates [14]. However the explant reactivation assay is complicated, relying on plaque formation in complex in vitro cultures that contain both virus-infected and immune cells. Reactivation rates are low, particularly from B cells [43], and may be affected by cell viability, in vitro antibody production and cytokines. The failure of IFNAR blockade to increase the viral antigen / eGFP staining ratio of d7 splenic WP B cells argued that here IFN-I is not a major regulator of in vivo lytic reactivation. The capacity of MCMV also to re-emerge acutely after Mx1-dependent tagging suggested that many herpesviruses can enter IFN-I responding cells and rapidly re-emerge. How far IFN-I can restrict MCMV was not explored, but clearly there are limits on what it alone is likely to achieve.
BALB/c, C57BL/6J and Mx1-cre mice [35] were maintained at University of Queensland animal units. Mx1-cre x ROSA26-YFP mice were bred at the Walter and Eliza Hall Institute. Mice were infected with MuHV-4 or MCMV when 6–8 weeks old, either i.n. (3x104 p.f.u.) under isofluorane anesthesia or i.p. (105 p.f.u.). Luciferase+ MuHV-4 infection (MHV-LUC) [42] was imaged by i.p. injection of D-luciferin (2mg, Pure Science) and charge-coupled device camera scanning (IVIS spectrum, Xenogen). IFNαβR signalling was blocked by i.p. injection of mAb MAR-5A3 (100μg/mouse every 2d); pDCs were depleted by i.p. injection of mAb 120G8 (200μg/mouse every 2d) (Bio X Cell). Poly-inosinic/cytidylic acid (poly(I:C), 50μg) was given i.p. (for i.p. infection) or i.n. plus i.p. (for i.n. infection) 6h before and at the time of virus inoculation. Statistical comparison was by Student's 2 tailed unpaired t test unless otherwise stated.
All animal experiments were approved by the University of Queensland and Walter and Eliza Hall Animal Ethics Committees in accordance with Australian National Health and Medical Research Council (NHMRC) guidelines. Project 301/13.
Bovine Hamster Kidney (BHK-21) fibroblasts (American Type Culture Collection CCL-10), RAW-264 monocytes (American Type Culture Collection TIB-71), NIH-3T3 cells (American Type Culture Collection CRL-1658), NIH-3T3-cre cells [57] and fibroblasts (from d13-14 mouse embryos) were grown in Dulbecco’s Modified Eagle’s Medium with 2 mM glutamine, 100 IU/ml penicillin, 100 μg/ml streptomycin, and 10% fetal calf serum (complete medium). All MuHV-4 variants were derived from a BAC-cloned viral genome [58]. ORF36-deficient MuHV-4 was made by shuttle mutagenesis, inserting into an XcmI site (nucleotide 53027 of Genbank sequence NC001826) of the ORF36 coding sequence (52848–54161) an oligonucleotide with multiple stop codons and an EcoRI restriction site. Correct mutagenesis was identified by EcoRI digestion of BAC DNA, and confirmed by sequencing of viral DNA across the insertion site. Infectious virus was recovered by BAC DNA transfection into BHK-21 cells, and the loxP-flanked BAC cassette removed by virus passage through NIH-3T3-cre cells. Luciferase+ [42], floxed reporter (MHV-RG) [27], and eGFP+ MuHV-4 [59] are described. MuHV-4 was grown and titered on BHK-21 cells. Floxed reporter MCMV (MCMV-GR) [60] was grown on NIH-3T3 cells. Virions were harvested from infected cell culture by ultracentrifugation (30,000 x g, 120min) and cell debris was removed by low speed centrifugation (500 x g, 10min).
To titer infectious virus, culture-grown stocks or freeze-thawed organ homogenates were plated on BHK-21 (MuHV-4) or embryonic fibroblast (MCMV) monolayers [61]. To titer total reactivatable MuHV-4, organs were disrupted into single cell suspensions then plated on BHK-21 cells. The cells were cultured in complete medium for 3h, overlaid with complete medium plus 0.3% carboxymethylcellulose, cultured for 4d, then fixed with 1% formaldehyde and stained with 0.1% toluidine blue. To measure viral fluorochrome switching, plaque assays were performed at limiting dilution, with 16 replicate wells per dilution. After 4d wells were scored for green (eGFP) and red (mCherry or tdTomato) fluorescence to derive virus titers for each colour., with % switching = 100 x switched plaque titer / (switched plaque titer + unswitched plaque titer).
Murine IFNβ was assayed by ELISA (PBL Verikine). Mx1 mRNA was quantitated in lung tissue (Aurum RNA isolation kit, Bio-Rad) by quantitative PCR (iTaq universal SYBR green kit, Bio-Rad) with Mx1-specific primers (qMmuCID0023356, Bio-Rad), and normalized by parallel amplification of Nidogen-1 (Rotor-Gene, Qiagen).
Organs were fixed in 1% formaldehyde / 10 mM sodium periodate / 75 mM L-lysine (18h, 4°C), equilibrated in 30% sucrose (24h, 4°C), then frozen in OCT. Sections (6μm) were air-dried (1h, 23°C), washed 3x in PBS, blocked with 0.3% Triton X-100 / 5% normal donkey serum (1h, 23°C), then incubated (18h, 4°C) with combinations of antibodies to eGFP (rabbit, chicken or goat pAb), CD68 (rat mAb, FA-11) (AbCam), B220 (rat mAb RA3-6B2), F4/80 (rat mAb CI:A3–1) (Santa Cruz Biotechnology), mCherry (rabbit pAb, Badrilla), CD206 (rat mAb MR5D3), CD169 (rat mAb 3D6.112) (Serotec), podoplanin (goat pAb, R&D Systems), and MuHV-4 (polyclonal rabbit sera raised by 2 subcutaneous virus inoculations). Sections were washed 3× in PBS, incubated (1h, 23°C) with combinations of Alexa568-donkey anti-rat IgG pAb, Alexa488 or Alexa647-donkey anti rabbit IgG pAb, Alexa647-donkey anti-mouse IgM pAb, Alexa488-donkey anti-chicken IgG pAb (Abcam), and Alexa488-donkey anti-goat pAb (Life Technologies), then washed 3× in PBS, stained with DAPI and mounted in Prolong Gold (Life Technologies). TdTomato fluoresence was visualized directly. Images were captured with a Zeiss LCM510 confocal microscope or a Nikon epifluorescence microscope and analyzed with Zen imaging software or ImageJ.
|
10.1371/journal.pcbi.1006679 | Change, stability, and instability in the Pavlovian guidance of behaviour from adolescence to young adulthood | Pavlovian influences are important in guiding decision-making across health and psychopathology. There is an increasing interest in using concise computational tasks to parametrise such influences in large populations, and especially to track their evolution during development and changes in mental health. However, the developmental course of Pavlovian influences is uncertain, a problem compounded by the unclear psychometric properties of the relevant measurements. We assessed Pavlovian influences in a longitudinal sample using a well characterised and widely used Go-NoGo task. We hypothesized that the strength of Pavlovian influences and other ‘psychomarkers’ guiding decision-making would behave like traits. As reliance on Pavlovian influence is not as profitable as precise instrumental decision-making in this Go-NoGo task, we expected this influence to decrease with higher IQ and age. Additionally, we hypothesized it would correlate with expressions of psychopathology. We found that Pavlovian effects had weak temporal stability, while model-fit was more stable. In terms of external validity, Pavlovian effects decreased with increasing IQ and experience within the task, in line with normative expectations. However, Pavlovian effects were poorly correlated with age or psychopathology. Thus, although this computational construct did correlate with important aspects of development, it does not meet conventional requirements for tracking individual development. We suggest measures that might improve psychometric properties of task-derived Pavlovian measures for future studies.
| Choice behaviour is guided by Pavlovian influences, so that particular features of a situation, e.g. if one seeks to gain rewards to avoid losses, privilege certain decisions over others–here, to be active versus vs. inactive respectively. Such privileging may be useful but may also impair optimal instrumental behaviour. We examined the balance of Pavlovian and instrumental guidance of choice in healthy, 14-24-year-old participants and found that young people with higher IQ relied less on Pavlovian guidance. Experience with a task changed Pavlovian guidance in a rational manner, so that it was used less when unnecessary. On the other hand, the degree of Pavlovian influence was not a highly stable trait and did not depend on age or mood. The degree of unpredictability in choice emerged as a crucial individual characteristic, associated with stronger Pavlovian influences yet more stable than them as a disposition.
| A leitmotif in the nascent field of computational psychiatry [1–4] is that carefully curated cognitive tasks can be used to identify latent dimensions of decision-making. These parametrize process accounting for how the tasks are solved, and are identified according to the models that best fit behaviour. Individuals are characterized according to their coordinates in these dimensions and it is by this means that dysfunction is delineated. A number of such dimensions, quantifying features such as reward and punishment sensitivity [5,6], uncertainty [7,8], exploration [9], metacognition [10], interpersonal modelling [11] have been extensively investigated in laboratory tasks.
However, to characterize individuals in a psychometrically competent manner, it does not suffice to have external validity in terms of indices of development and pathology. Temporal stability is also crucial [12,13].Stability and related psychometric properties are increasingly important as computational psychiatry moves from describing differences between selected groups of individuals, for example well vs. ill groups, to describing individual change attributable to development, vulnerability to psychopathology, and recovery from psychiatric disorder. Stability is also crucial if computational parameters are to guide diagnosis and personalized psychiatry. Unstable measures may have predictive value [14,15], especially if their variability can be understood [16], but cannot easily characterise individual trajectories. It is unclear whether computational tasks that have been well validated in the laboratory, and which are starting to be used in epidemiological samples studies [17,18], have psychometric properties sufficient to pinpoint individual dispositions. In particular, while learning tasks are amongst the most popular in computational psychiatry, it is not clear if they bear repetition, for instance whether the identity of the best fitting process model remains the same when they are applied again. Furthermore, we often do not know if parameters inferred by using these best models are psychometrically reliable, covarying with traits, or change with the individual’s state and experience.
Here, we study the population distribution and psychometrics of a paradigmatic computational measure, namely the extent to which an individual’s decision-making is guided by Pavlovian influences [19]. This is the direct predisposition to prefer particular actions in response to features of a stimulus, such as the appetitive or aversive consequences that it predicts. This predisposition can help or hinder instrumental behaviour, which is defined in terms of the contingency between action and outcome. Pavlovian biases have often been studied because of their translational relevance for anxiety, post-traumatic stress, and other disorders [20–23]. The incidence of psychiatric symptoms where Pavlovian influences have been implicated rises in adolescence. For example, we recently described a peak in mood symptoms in a large non-clinical sample around the age of 16 in females [13]. It is thus important to examine how Pavlovian influences vary with age as well as characteristics such as mood, sex and IQ [1]. This in turn makes psychometric questions related to tasks assessing Pavlovian phenomena particularly pressing.
Pavlovian influences elicited by predictions of reward and punishment have been extensively studied through variants of a Go-NoGo task [19,24–26]. Here, subjects prefer to execute, rather than withhold, actions in proportion to their expectations of winning money. This Pavlovian ‘bias’ is quantified as a perturbation of a standard reinforcement learning model, using a form of bonus that is proportional to the predicted value associated with the stimuli concerned [25]. This rewards ‘Go’ actions in the face of ‘potential win’ stimuli and ‘No-Go’ actions in the face of ‘potential loss’ stimuli.
To assess Pavlovian bias and associated decision-making characteristics, we used an orthogonalised Go-NoGo task, wherein optimal decisions (Go vs. NoGo) are independent of the goal outcome (winning vs. avoiding loss). This has been extensively validated to assess Pavlovian bias, while much is known about neural function in this task [19,24,25,27]. We administered it on two occasions (termed ‘baseline’ and ‘long follow-up’) in a large, naturalistic, epidemiologically informed sample of 14 to 24 year olds [28]. We first validated the class of reinforcement-learning models developed in the laboratory in this population. In the process, we asked which model best described behaviour and ascertained that estimates of Pavlovian bias were robust with respect to secondary modelling details. We considered a model with differential sensitivity to wins and losses, which Guitart-Masip et al, 2014, found to fit behaviour best (‘valenced-sensitivity’ model). We also considered variants, particularly equal sensitivity to wins and losses, but differential learning to these outcomes (‘valenced-learning model’). We then compared the psychometric properties of the best models using real but also simulated data. We examined the external validity of the best models by assessing correlations between parameter values and the variables of age, IQ and mood. Calendar age is a key variable in development, albeit not the same as developmental time [29–31]. IQ is also an important yardstick, as theoretical [32] and experimental [33] findings motivate further examination of its relationship with Pavlovian tendencies. Specifically, we argued that in the orthogonalized Go-NoGo task used here, efficient instrumental learning rather than reliance on non-instrumental Pavlovian biases is most profitable. Therefore, participants with higher IQ might be expected to rely less on Pavlovian guidance. Finally, we used the ‘Mood and Feelings Questionnaire’, or MFQ for external validation. This was motivated by two considerations: first, research has suggested links between Pavlovian bias and so-called internalizing disorders [20,21,26]. Second, we have found that MFQ is a good simple proxy for the ‘general psychopathology factor’ (see S1 Fig and [13]). It may thus shed light into how Pavlovian bias contributes to psychiatric vulnerability or resilience in general, although further research should address associations with other, specific dimensions of psychopathology.
We examined longitudinal changes and correlations, particularly concerning the best fitting model and the trajectory of Pavlovian parameters. First, we performed a ‘short follow-up’ study over an interval of 6 months. This is short in developmental but not in test-retest terms. It helped us to subsequently interpret the results of our main, ‘long follow-up’ study, about 18 months post-baseline. Over and above model parameters, model-fit was of over-arching importance, as it assessed how well a specific cognitive model captures individual behaviour. We then explored the dependence of model-fit on age, task repetition and IQ.
We tested three hypotheses for the trajectory of the Pavlovian parameter. One is that the bias characterising an individual was a trait that remained stable over the time of our study. Alternatively, it might reflect a slowly changing developmental disposition, specifically one where younger participants were more strongly guided by Pavlovian biases. A third possibility is that Pavlovian biases reflected prior beliefs not only dependent on the context of ‘opportunity’ (appetitive trials) or ‘threat’ (aversive trials), but also dependent on other features of the context, such as ‘task taking place in a particular laboratory’. In the latter case, participants could update their prior beliefs about a link between appetitiveness and the appropriate action across test sessions.
Key descriptive characteristics of participants at baseline are shown in Table 1.
We tested 61 participants at baseline, and then after an interval of 6 months (Fig 1), using the task briefly described above (and fully in Methods; very similar to [26]). We thus first explored temporal individual stability and group-level change over a time scale which was short in developmental terms. We report uncorrected Pearson r for approximately Gaussian quantities and Spearman ρ for non-Gaussian ones, using parameters inferred from the preferred model variant that emerged from model-comparisons. This was the ‘valenced learning’ variant, quite similar but not identical to the established one [19,24]. Please see the Methods section for details.
For the overall propensity to choose action over inaction, parametrized by the ‘Go Bias’, baseline estimates were significantly correlated with short follow-up estimates, as hypothesized (r = 0.30, p = 0.018; Fig 2). However, this was not true for the other parameters. The Pavlovian bias, parametrizing the propensity to action in the context of opportunity and inaction in a context of loss, had p = 0.54. The motivational exchange rate, which measures how strongly likelihood of a choice depends on its value, had p = 0.55. For the learning rates for the appetitive and aversive contexts, p was 0.13, and 0.52 respectively. The irreducible noise parameter, quantifying decision variability that could not be reduced by learning motivating actions, had r = 0.24, p = 0.052. The extent to which the model accounted for behaviour, the integrated likelihood measure, was the most inter-correlated variable between baseline and short follow-up, r = 0.43, p = 0.00047. As we shall see below, similar results obtained in the larger, long follow-up study, suggesting that developmentally, individuals largely maintain their rank within the cohort with respect to this measure.
Next, we tested whether each of the parameters increased or decreased with task repetition. As we had no a priori hypotheses, we applied a Bonferroni correction for 6 tests, so that a corrected threshold of p = 0.05 corresponded to uncorrected p = 0.0083. We found that outcome sensitivity clearly increased (uncorrected Wilcoxon p = 5.5e-7) over the 6-month interval. There was weak evidence that Pavlovian bias decreased (from a median on 0.20 to 0.12; uncorr. t-test p = 0.024).
There was no significant change in the other parameters over the short follow-up, but there was good evidence that the median integrated likelihood increased (from -66.4 to -53.7, uncor. Wilc. p = 0.0038). 54 of the short-interval participants were also included in the long follow-up sample (Fig 1). They are included in the long follow-up analyses below, but their exclusion results in minimal change. For example, the baseline vs. long follow-up correlation ρ of Pavlovian bias does not change, while its p-value would drop slightly from 0.017 to 0.020.
Key results from the short-follow up study thus were that model fit was longitudinally the most stable measure, while the group shifted its outcome sensitivity and Pavlovian bias in the direction of benefitting performance, and it improved its model-fit.
In the large naturalistic study, we first collected data from N1 = 817 participants (‘baseline’ sample). Of these, N2 = 556 (68%) also provided valid data at a follow-up session, on average 18 months later (‘long follow up’ sample). We first analysed performance simply in terms of the proportion of correct responses that participants achieved in each task condition, time during the task and session (Fig 3).
The characteristic ‘Pavlovian bias’ interaction pattern was seen, with Pavlovian-incongruent conditions showing worse performance than the corresponding congruent ones (Go to Win > NoGo to Win, NoGo to Avoid Loss > Go to Avoid Loss) at all stages. As shown in Fig 3A (‘early trials’ panel, second vs. third pair of boxes), G2AL showed a clearly better level starting level of performance than the other pavlovian-incongruent condition, NG2W (baseline difference: 18%, pcor < 1e-10; long follow-up difference: 16%, pcor < 1e-10). However, for the ‘late’ trials (Fig 3B) the improvement in median fraction of correct responses in G2AL was modest compared to those of NG2W. Hence, median performance in the latter now matched the former (baseline difference: -0.6%, pcor < 0.05; long follow-up difference: -0.6%, pcor = NS). This pattern suggests that not only Pavlovian congruency, but action and/or valence biases in learning and decision making need to be considered.
As descriptive statistics do not distinguish clearly the roles of Pavlovian bias and other processes of interest, we fitted a range of computational models capable of these distinctions to the data. We used the integrated likelihood (iL) and Bayesian Information Criterion (iBIC) to quantify complexity-corrected accuracy (Fig 4) and thereby compare models [24]. We assessed whether the identity of the best-fitting model remained the same over testing sessions, and whether estimates of Pavlovian bias were robust to secondary modelling considerations. We expected correlations for each task parameter across time to be positive. Accordingly, we report uncorrected p’s for Spearman correlations and Wilcoxon paired t-tests. We were also interested in the direction of any systematic change. Here, in the absence of a priori hypotheses, we report Wilcoxon tests, applying a Bonferroni correction for as many comparisons as there were parameters.
In the baseline sample, the valenced-learning model performed best (Fig 4, leftmost), with the valenced-sensitivity model 255.7 BIC units behind. Importantly, these two models produced very similar estimates for the parameters of interest here (r = 0.88 between models for the Pavlovian parameters, r = 0.93 for Go-bias, p <<1e-05). A variant of the valenced-senstivity model, the ‘sensitivity ratio’ model, also furnished highly similar Pavlovian bias estimates (e.g. r = 0.93, p <<1e-05 with valenced-learning). As a quality check, the ‘irreducible noise’ parameters, which quantify lack of attention and motivation-independent lapses, were reassuringly low (5–7%). In the long follow-up sample the valenced-learning model again obtained an advantage, here of 275.6 BIC units over the second-best, valenced-sensitivity model.
However, the mean advantage of the winning model per participant was only 0.15 BIC units at baseline, while the typical individual uncertainty in iBIC, estimated as the SD of BIC scores refitted to data generated using the exact mean population value of each parameter for the winning model, was 7.9 units. Thus, while evidence of >250 BIC units is considered overwhelming by conventional standards [34], we asked if in studies with N>500 it might arise by chance and, as importantly, what difference in predictive power it signifies. Using a paired Wilcoxon test to compare fits for the two models for the data obtained in the first (baseline) testing session gave p = 0.09, while for the long follow-up Wilcoxon p was 1.7e-4, overall providing evidence against a false positive finding. Similarly, using integrated likelihood estimates at the individual level yielded values of approximate protected exceedance probabilities of 0.569 for baseline and 0.974 for long follow-up in favour of valenced-learning [35]. However, when we asked if having a greater likelihood for the one model at baseline implied a similar ordering at long follow-up, a chi-square test showed no evidence (p = 0.38). Numerical studies later showed that even if individuals’ employment of a particular model [valenced learning or sensitivity] remained fixed, and so constituted a ‘type’, our relatively brief experiment would be under-powered to allocate individuals to their type reliably (see S1 Appendix, ‘In silico simulated agents’ reliability and biases’).
We then estimated predictive power. First, we expressed the difference in BIC in terms of the probability of the model better predicting a participant’s decision per trial. Even at long follow-up, this difference was very small, mean ΔPpt = 0.0011, compared to a grand mean prediction probability per trial, Ppt = 0.64. We also introduced a new out-of-sample, or ‘left out likelihood’ (LOL) comparison method, suitable for tasks involving learning which present challenges for predictive tests. This is important, as predictive tests do not rely on the approximations inherent in the BIC and become more and more powerful as datasets become larger (interested readers are referred to the Methods section for validation and details). LOL testing confirmed in an unbiased manner that the likely difference in predictive power between the two best models was very small, interquartile range of ΔPpt -0.012 to 0.0026. Here the mean predictability per trial was 0.73 (see Fig 5 and Methods).
To further assess model quality, we examined correlations between parameters within each of the best models. The ones remaining significant after correction for multiple comparisons for the valenced-learning model are shown in Table 2 (valenced-sensitivity is similar). Four within-model significant correlations were consistently found at baseline and long follow-up. The Pavlovian bias was anticorrelated with both learning rates, while the motivational exchange rate correlated with the Go-bias and anticorrelated with irreducible noise. The motivational exchange rate is also known as reward sensitivity or inverse decision temperature. It can be seen as the power that a unit of additional reward (or loss) has to shift behaviour off indifference between choices. The correlations were also significant in the long follow-up sample (Pav. bias vs. appetitive and aversive learning rates: ρ = -0.14, -0.19, p = 0.016, 0.00011; motiv. Exchange rate vs. irreducible noise and Go bias: ρ = -0.13,0.14, p = 0.043, 0.020). S2 Table shows similar results for the valenced sensitivity model.
We then examined stability in the quality and nature of the performance, which is the main focus of the study. We started with descriptive measures. As suggested in Fig 3, while there was no overall change in G2W performance between baseline and follow-up (uncorr. Wilcoxon p = 0.31) the other three conditions did improve (G2AL by a median of 5.5%, p< 1e-5 corrected for 4 comparisons; NG2W by 2.8%, p = 5.4e-4; No-Go to avoid loss (NG2AL) by 2.8%, p = 1.78e-4). Next, we used these changes to compare an estimate of Pavlovian bias in follow-up vs. baseline. This estimate showed modest stability across time (Spearman ρ = 0.146, p = 5.4e-4; Fig 6A) but there was a significant reduction in its mean (p = 0.0019), largely attributable to a closing of the gap between G2W-NG2W (Fig 6B). Cross-sectionally, at baseline the first three conditions showed no significant linear or quadratic dependence on age (uncorr. regression p: G2W, 0.22; G2AL, 0.16; NG2W, 0.65). For NG2AL, the linear regression explained 1.0% of the variance, p = 0.021 corrected for 8 comparisons, with positive linear dependency on age. At long-follow-up, again the first three conditions showed no significant dependency at the uncorrected level (regression p: 0.98, 0.096, 0.73). However, at this time there was trend evidence for a positive linear and negative quadratic dependence of NG2AL performance with age (regression p = 0.093 corr. for 8 comparisons, adj. r2 = 0.012; S2 Fig).
To explore the evolution of the component cognitive processes, we examined the stability of the parameter values extracted from the fits of the winning computational model. To put the results that follow in perspective, the Pearson correlation for IQ between sessions was 0.77, pcor << 1e-05 while for the mood measure it was 0.61, pcor << 1e-05. A modest correlation across time points was detected for the Pavlovian bias (r = 0.10, p = 0.017; Fig 6A). A more salient result was an overall reduction in Pavlovian influences with session (from a median of 0.205 at baseline to 0.142 at long follow-up, pcor < 1e-5). We found evidence that the short-follow-up participants changed their Pavlovian bias as much in the 6-month interval as the rest did in the mean-18-month interval. The scaled-information Bayes factor in favour of no difference was 4.01 (JZS scaled Bayes factor = 7.15, t = 0.76, p = 0.45). Decreases in Pavlovian bias between baseline and long follow-up were strongly correlated with improvements in performance in the Pavlovian-incongruent conditions, and anticorrelated with improvements in the Pavlovian-congruent conditions (S3 Table).
We then performed a set of latent change score (LCS;S3 Appendix) analyses, useful for describing longitudinal change [36]. The most complex multivariate-normal model that can be fitted to the data, known as the ‘just-identifiable model’, showed a significant dependence of change on baseline (regr. beta = -0.878, p<1e-3), with higher-bias individuals at baseline reducing their bias more at long follow up (visible in Fig 5A). This model was superior to one assuming that change only represented regression-to-mean (χ2 = 7.63, df = 1, p = 0.0057) and also to one assuming the same mean and variance at long follow-up vs baseline (χ2 = 6.37, df = 2, p = 0.041). See S3 Appendix for illustration and more details.
We found a substantial temporal correlation for the motivational exchange rate (ln(beta); r = 0.253, p = < 1e-5; Fig 7B) and especially for the model fit measure, the integrated likelihood, (r = 0.37, p < 1e-15; Fig 8A). The latter increased at long follow-up, from a median of -68.9 to -64.4, pcor = 0.023. There was evidence for a temporal correlation in learning rates, in both the appetitive and aversive domains (r = 0.09, p = 0.045 and r = 0.11, p = 0.011 respectively). Both increased significantly from baseline to long follow up (median differences of 0.028 and 0.027 respectively; both pcor< 1e-05). There was trend evidence of temporal correlation for the bias parameter favouring action over all trials (‘Go bias’: r = 0.082, p = 0.054) and the lapse rate parameter (r = 0.077, p = 0.068). Go-bias decreased (median 0.73 to 0.57, pcor = 0.014), but the most significant change was a decrease in lapse rate (median 0.069 to 0.055, pcor < 1e-05).
For none of the parameters, or their change, did we find evidence of correlation with participant age at baseline (correcting for seven comparisons). We also examined whether the baseline parameters depended on gender, mood (the ‘Mood and Feelings Questionnaire’) or IQ (WASI total IQ). Correcting for multiple comparisons, we found no significant dependency of baseline parameters on gender or mood and consequently did not analyse for such dependencies further.
IQ was significantly related to the parameters, decreasing with increasing Pavlovian bias at baseline (r = -0.13, pcor = 0.007) as hypothesized. Other parameters also related significantly to IQ, notably motivational exchange rate (r = 0.27, pcor< 1e-10), appetitive learning rate (r = 0.11, pcor = 0.044) and aversive learning rate (r = 0.19, pcor < 1e-05). IQ strongly correlated with overall model fit (r = 0.28, pcor < 1e-10). In the long follow-up sample the model fit (r = 0.31, pcor < 1e-10), motivational exchange rate (r = 0.24, pcor < 1e-05), and lapse rate (r = -0.14, pcor = 0.014), but no other parameter, correlated with IQ. Regressing IQ on the (log) appetitive and aversive sensitivities of the valenced-sensitivity model obtained a multiple r-squared of 6.1%, whereas the single beta of the valenced-learning model absorbed 7.6% of the variance in IQ.
We then examined the distributions of parameters and of model fit for evidence of sub-grouping. We used the baseline data for this screening, as it was epidemiologically most representative. There was no evidence for sub-grouping in the parameters, but there was for the fit measures, where a bimodal distribution was evident (Fig 8A). This motivated analysis of the joint distribution of baseline and long follow-up model-fit measures. This was best described by a mixture of two major, approximately equipopulous, Gaussian components and two somewhat less prominent ones (Fig 8B). This means that the behaviour of the high-likelihood clusters is much more predictable (less random) according to our models.
We performed a mixed-effects analysis (here, controlling for re-test) to draw out differences between the clusters in terms of Mood, IQ and task parameters. As expected, we found differences between sub-groups in integrated likelihood (by construction) and motivational exchange rates and irreducible noise, which are closely related. More interesting, we found a consistent pattern across parameters and IQ, where the cluster which fit worse at both time points (blue or 4 in Fig 8) did significantly worse than most in all performance-sensitive measures, and had increased Pavlovian bias compared to clusters 2 and 3 (2 and 3 had the better fit at follow-up, Fig 8B). Pavlovian bias did not differ amongst the other clusters, while cluster 4 had greater IQ than the others. See ‘S2 Appendix: Clustering analyses’, for statistical details and illustration.
Finally, we performed in silico analyses to determine whether the fitting procedure could reliably recover known parameters, and whether it might introduce spurious correlations between them (see S1 Appendix, ‘In silico simulated agents’ reliability and biases’ for details). Recoverability in silico was much better than stability in vivo, suggesting that the former is not a limiting factor in our study. Similarly, no spurious correlation was observed between the Pavlovian bias and learning rates, which means that the associations between them seen in the real data are likely to reflect a true feature of the study population rather than a modelling artefact.
We report the first longitudinal assessment of the psychometric properties of a key computational parameter, namely the Pavlovian bias. We did this in a large, epidemiologically-based study of young people. The Go-NoGo task that we used yielded informative results in terms of the cognitive process likely to operate, the evolution over time of the parameters of that process, the construct validity of the model parameters in question and also useful methodological considerations.
We found behaviour was well fit by two models, the winning one having two different learning rates for reward and loss but a single sensitivity to returns, and the other having a single learning rate, but two different sensitivities. This is consistent with, but also finesses, recent results stressing the dependence of learning rates on outcome valence, especially in young people [37,38]. We compared models not only on the basis of approximations to the statistical evidence for each, but also on their ability to predict left-out decisions. The dual-learning model fit better, but there was no evidence for a clear allocation of model type to individuals. Model-fit improved with practice and was greater for subjects with higher IQ. Estimates of the Pavlovian bias were robust to model type, but their test-retest stability was weak, limiting inferences about individual development. At the group level, Pavlovian bias decreased when participants were re-tested 1–2 years post-baseline.
Pavlovian bias changed over test sessions, but not over age, in a characteristic pattern. In terms of hypotheses we set out to test, we interpret this as strong evidence against this parameter, as measured by this task, being a fixed stable trait. Second, we interpret a longer follow-up resulting in the same reduction in Pavlovian bias as the short one as modest evidence against this bias being a disposition slowly changing with development. The pattern is most strongly supportive of a hypothesis that Pavlovian bias approximates an experience-dependent prior expectation. Of course, our models do not directly compute the ‘probability that in an appetitive context, the correct decision is to act’, which would formally be a belief. However, an agent using such a belief would prefer Pavlovian-congruent actions, and a weakening of such a belief with experience would lead to weakening of this preference, so the evolution of Pavlovian parameters approximates beliefs or expectancies about contingency. The fact that model-fit was good and improved with time, yet parameters changed, argues against a fixed disposition. The virtual absence of cross-sectional age dependency of the parameters, and shifts being as pronounced after 6 months as after 1–2 years, argues against spontaneous slow development and for an effect of practice. Lack of strong age dependency of performance has been observed in other reinforcement learning tasks (e.g. [37] in adolescents vs. adults), but both performance and cognitive parameter age dependencies very much depend on the specific details of the task at hand [39,40]. Practice effects may have affected the improvement in the extent to which people were described by our models. Selective attrition may also have affected our data, though a follow-up rate of 68% is reassuring here. However, attrition may have rendered the long follow-up sample epidemiologically less representative than the baseline one.
In this task, the Pavlovian bias aids performance in two conditions, and harms it in the other two. If one interprets the bias as a heuristic that is generally useful, the fact that it reduces with repeated testing is consistent with its approximating a prior belief which can be at least partially overwhelmed by evidence, rather than being rigidly hard-wired by genetics and the early-childhood environment. This may apply more generally to other cognitive biases, and may happen over multiple timescales. Our study only employed a limited number of follow-up occasions, making it difficult to discern longitudinal changes not due to practice effects. In this study, any age, gender or mood dependency of Pavlovian bias, if present, was too subtle to resolve.
Psychometrically, the orthogonalized Go-NoGo task had the capacity to provide reliable estimates of Pavlovian bias, in that parameter recovery in silico was greater than 80%. At the same time, in human participants stability was reduced down to ~ 10–15%. The return sensitivity parameter was more stable. However, the stability of task measures was considerably lower than those for IQ and even mood, and lower than that conventionally required for tracing developmental trajectories [41].
Model fit emerged as important, over and above individual parameters, being the most stable task measure. It classified participants into clusters with discernible longitudinal trajectories (Fig 8 and S2 Appendix). The most striking differences were observed between a cluster consisting of people who showed poor at both time points (4 in Fig 8B and S2 Appendix) and the rest. These participants were characterized by a higher Pavlovian bias and lower IQ than the rest, as well as by higher decision-noise parameters. Mood did not differ between any clusters. Poorly fitting participants may have followed a strategy less well captured by our models, or may have been irreducibly more stochastic. In either case, model-fitting causes decision-variability parameters to absorb this high variance, whether its cause is hidden cognitive variables that are as yet not represented in the models, or random noise. It would be interesting to use less constrained machine-learning models to estimate the upper limit on the amount of variance in the data that cognitive-mechanism models like ours hope to explain, and thus help interpret decision variability parameters [42]. Overall, model-fit emerged as a potentially important measure to classify developmental trajectories in future research.
We found that IQ correlated with motivational exchange rates, and indeed model fit. That is, the behaviour of those with higher IQ was more affected by a unit improvement in objective outcome, over and above differences in learning and bias. This is what one would expect if IQ test scores were themselves dependent on how motivating the participants considered finding the correct answers in the IQ tests. This is in turn consistent with evidence that IQ test scores can be increased by material incentives and that motivation in the absence of additional incentives predicts real-life outcomes [43]. Likewise, confidence is linked to both IQ and motivation [44]. Our motivational exchange rates may thus reflect ability-dependent confidence, important for development during youth. Tasks involving simple inference about counter-intuitive contingencies, building on our Go-NoGo task, may be useful in exploring these relations. Furthermore, the neural process underpinning reductions in Pavlovian bias would be interesting to elucidate, as it has been suggested that top-down processes actively suppress this bias in more able participants [45]. Alternatively, reductions in Pavlovian bias occurring over time or across IQ groups in young people may relate to differences in fixed, for the duration of the experiment, Pavlovian parameters integral to associative-learning systems (bpav in Eq 3) like those of our models.
Though not affecting our central hypotheses, model-comparison analyses raised important questions as to how young people learn differentially from loss and gain events. The fact that behaviour could be explained almost equally well either by recourse to valenced-learning or to valenced-sensitivity, merits further study. Learning rates and sensitivities can be distinguished [46], but the present task was not optimized to do so. A larger number of trials, for example, could help resolve this ambiguity. In terms of our models, subjects behaved either as if volatility were higher for loss compared to gain contexts (faster aversive learning), or were loss averse (higher aversive sensitivity). If appetitive and aversive sensitivities are valid concepts, they should be similar in comparable but learning-free settings, such as well-learnt versions of this task (improved from [26]). Based on the present results, we hypothesize that in a task capable of simultaneously resolving the valence-dependence of preference (valenced sensitivity) and of learning (valenced learning), the two sensitivities would be closer to each other than suggested by a pure valenced-sensitivity model, but would not be identical, as they would also be informed by loss aversion.
We studied the orthogonalized Go-NoGo task because it is well established, widely used, specialized to assess Pavlovian bias, and could provide insights (especially about decision noise and model fit) likely to be relevant to other computational tasks. However, the findings reported here should be used with caution in other contexts. Further insights about Pavlovian bias may be provided by existing paradigms [47], but testing for the putative experience-independent core of this bias with a brief task, suitable for developmental research, could be facilitated by an adaptive design. We suggest maximizing the number of plateau-performance trials by adaptively looking for the true indifference point between the effective values of the actions (q, not Q, in Eq 3). An experience-independent Pavlovian bias would mean that in the appetitive domain, the reward for the ‘No-Go’ action has to be (adaptively) boosted by an amount proportional to the mean reward of ‘Go’ and ‘No-Go’ in order to achieve indifference between these two choices. Similarly, in an aversive context the value of ‘No-Go’ has to be adaptively penalized by an amount proportional to the average of the loss returns for the two choices to achieve indifference. For symmetric average returns, the mean of the (absolute) adaptive boost and adaptive penalty components would be proportional to the Pavlovian bias, and the difference between these adaptive amounts would be the ‘Go bias’.
Methodologically, the present work extends the use of left-out-likelihood based model comparisons. These furnish outcomes that are directly intuitive and convenient for further statistical comparisons, as well as being free of approximations inherent in the BIC. On the other hand, they are much more computationally intensive and will require further refinement to render their use routine.
In conclusion, we show that Pavlovian influences characterizing young people are well described at the epidemiological level by established reinforcement-learning models. Practice and higher IQ correlated with weaker Pavlovian influences, while higher IQ scores were also associated with higher motivation to attain a given reward, suggesting important neurodevelopmental relationships. However, neither the Pavlovian bias parameter nor other key task measures met conventional psychometric standards for temporal stability or for external validity with respect to age and psychiatric symptoms, attributes useful in characterizing individual variability and individual change. It is a matter for further research whether similar problems affect other computational tasks, but our study does give urgency to the work of establishing the psychometric properties of such tasks and the parameters associated with computational models them. Studies aiming to characterize individual trajectories of decision-making will benefit from psychometrically improved computational tasks, which better exclude experience-dependent components, as well as designs that include more follow-up points.
We used an orthogonalized Go-NoGo task that contrasts a propensity to act, rather than not to do so, in context involving opportunity (‘win’) versus threat (‘avoid loss’). Participants were presented with four different abstract stimuli each of which had a constant, but unknown, association with a correct policy. The correct policy was either to emit (‘Go’) or to withhold (‘NoGo’) an action, here involving a button press. If the correct decision was made, the better of two outcomes was realised with probability 0.8. This better outcome was null (as opposed to a loss) for two stimuli and positive (as opposed to null) for the other two. The task closely followed a previously published paradigm [25], with some slight simplifications, unrelated to the core biases assessed by the task. These simplifications helped deliver it to a community sample, on a large scale and in the context of a multi-task battery. First, implementing the decision ‘Go’ was simpler, i.e. not dependent on any target features, unlike the original task in which the ‘Go’ action could be either ‘left’ or ‘right’ depending on the location of a target. This allowed trials to be shorter. However, time pressure from the remaining task battery (to be reported separately), meant that subjects performed a more restricted sample of 144 trials. Second, task clarity was improved by informing participants before performing the task that the outcome probabilities were 0.8 and 0.2. Third, motivation was made explicit by telling participants that they were playing for real money, that random performance would attract zero extra fee and excellent performance could be worth about five pounds sterling additional earnings. These changes were supported by piloting the whole battery in which the task was embedded, as we describe next.
We took precautions to ensure that the fact that the task was delivered as part of a battery did not affect the power for testing the hypotheses in question. The battery of which this task was part of consisted of 7 tasks and took over 2.5 hours to complete, whereas the task analysed here took about 23 minutes to complete, longer than the average in the battery. We first examined data from previous, longer versions of the task and performed a pilot of 15 participants. In addition to quantitative data, these participants were de-briefed in detail by trained research assistants (RAs) who interviewed them as to whether they found the tasks tiring, interesting or difficult. Although quantitative data were in line with the literature, qualitative data suggested that some participants might, subjectively, be affected by tiredness but most importantly some found the task hard to work out and felt discouraged by this. Therefore, first, the randomization of the order of tasks in the battery was constrained, so that this task took place within the first hour of testing. Second, research assistants were assertive in enforcing short breaks between tasks and emphasizing the importance of attending to the task. Third, they reminded participants that they were playing for real money and that all decisions counted approximately equally, in monetary terms, encouraging attention to each decision rather than assume that shorter tasks paid as much as longer ones. Fourth, participants were reassured that they should not be discouraged if the best answers were not clear to them as the task progressed, but on the contrary they should proceed by trial and error and the best answers were then likely to gradually ‘sink in’. This is consistent with the Rescorla-Wagner model that we used to analyse the data. After the first 50 participants were tested under careful RA supervision, an interim analysis of the whole battery and more limited feedback from RAs was reviewed. This gave no cause for concern with respect to the present task and the quantitative parameters extracted were reassuringly compatible with those from historical laboratory samples (as well as the Results reported here).
Participants were thoroughly informed about the task, including a veridical performance-related pay component.
Community dwelling participants were recruited from within the volunteer database of the Neuroscience in Psychiatry Network Study [28].
Volunteers were invited to be approximately equally distributed by gender and age, between the ages of 14 and 24 years old, from Cambridgeshire (60% of sample) and North London (40%). We excluded those with moderate or severe learning disability or serious neurological illness. Recruitment continued until a total sample of 820 young people agreed to participate at baseline. The Cambridge Central Research Ethics Committee approved the study (12/EE/0250). Participants gave informed consent themselves if they were at least 16 years of age, otherwise the participant was fully informed and agreed to the study, but their parent or legal guardian provided formal informed consent.
We explored variants of the core (here called valenced-sensitivity) model that [25] used to describe behaviour. First, the values of actions (‘Q values’) were calculated based on a learning rate λv. We use ‘v’ subscripts to indicate that, in different variants of the model, the parameter in question may be valenced, i.e. different for ‘win’ and ‘avoid-loss’ trials. In the valenced-sensitivity model all trial types shared the same learning rate but different motivational exchange rates ρv were used depending on trial valence:
Qt+1=Qt(at,st)+λv(ρvrt−Qt(at,st))
(1)
Only Q values pertaining to realized stimuli and actions were updated, with all others being carried forward from the previous trial. The model also kept track of the state values pertaining to each stimulus using the same parameters:
Vt+1=Vt(st)+λv(ρvrt−Vt(st))
(2)
Crucially, the conventional Q values were biased by two terms representing an overall tendency towards action (‘Go bias’) and a Pavlovian bias (towards action or inaction) that depended on valence (state value):
qt(at,st)=Qt(at,st)+bgo(at)+bpav(at)Vt(st)
(3)
where the two bias coefficients b are zero unless at=Go. In effect this means that the No-Go action was taken as a comparator, and the Go action was either penalized (in aversive contexts) or boosted (in appetitive ones) proportional to the value of the respective stimulus. Finally, the policy probability for choosing an action was given by the softmax function, modulated by a lapse rate parameter ξ:
p(at|st)=(1−ξ)exp(q(at,st))∑k(exp(q(ak,st))+ξ2
(4)
Thus the motivational exchange rates ρv acted as an inverse temperature parameter for the softmax by scaling the outcome values in Eq 1, which then fed into Eq 4 via Eq 3.
Several model variants were explored. First, separate learning rates were used depending on valence (‘valenced-learning model’) with a single outcome sensitivity.
Second, an additional memory/forgetting parameter was introduced such that Q values pertaining to unexperienced state-action pairs decayed by a constant fraction per trial, rather than being carried forward intact (‘forgetting model’), quantified by an additional ‘memory’ or ‘forgetting’ parameter [48]. In these ‘forgetting models’ models, we argued that during long periods when, by chance, a particular stimulus was not observed, the associated actions might drift back to zero if participants brought to the task a (strictly unjustified, but common-sensical) assumption that values might be subject to non-zero volatility. The other models assumed that the value of stimuli not seen in a particular trial would not change.
Third, the appetitive and aversive sensitivities of the valenced-sensitivity model were transformed into an overall sensitivity and an appetitive/aversive ratio (‘sensitivity ratio model’). This had two motivational exchange parameters, just like the valenced sensitivity model, but formulated them as an appetitive sensitivity and a sensitivity ratio. Therefore, at the level of the individual it was identical to the valenced sensitivity model. At the group level, the expectation-maximization fit used prior distributions for the parameters that were independent of each other. In the case of the valenced sensitivity model, this means that the distribution of appetitive sensitivity over the population was modelled as independent to the aversive one. For the sensitivity-ratio model, a positive mean for the population distribution of the eponymous parameter would encode a positive correlation between appetitive and aversive exchange rates. Model fits favoured the former model.
We used hierarchical type-2-ML model-fitting, assuming that each wave of data could be described by a set of independent prior distributions for the mean and spread of each parameter. Waves of testing were fitted independently. Individuals’ parameters were optimized given point estimates of the mean and spread of the group they belonged to, which were themselves re-estimated. Specifically, we used the expectation-maximization algorithm in [25], modifying MATLAB [49] code provided by Dr. Quentin Huys. We estimated the ‘integrated likelihood’ and ‘integrated BIC’ measures used in that work (Eq 5), using the same sampling technique.
iLpt=ln∫p(dpt|θ,Μ)p(θ|Θ)dθiBIC=−2∑ptiLpt+Nparln(ntrnpt)
(5)
Where iLpt is the integrated likelihood for participant pt, dpt the data provided by this participant, θ the ‘micro’ parameters of that participant according to model M, Θ the ‘macro’ parameters describing the population distribution of θ, Npar the length of Θ, ntr the length of d (same for each participant) and npt the number of participants.
For the purposes of model-fitting each parameter was transformed according to theoretical assumptions and the group distribution approximated by a Gaussian in the transformed space. For example, following [25] we assumed that bpav>0 so this parameter was fitted in log space, whereas parameters having both an upper and lower bound were logistic-transformed. We tested whether the model-fitting procedure was robust to assumptions about the distribution of parameter values in the population (S1 Appendix). We tested two assumptions that were particularly relevant to Pavlovian bias. First, we allowed the population distribution of Pavlovian bias to be normal rather than log-normal. Using mean and variance population estimates derived from the real data, this translated to the presence of a small tail of negative values (S1 Appendix). Second, we tested whether model-fitting was sensitive to the precise form of the population distributions used, e.g. gamma rather than log-normal. Recovery of Pavlovian bias parameters was reassuringly robust in the face of such twists (S1 Appendix).
In order to go beyond comparing models by using simple approximations like the BIC, we argued that better models should provide a higher likelihood for data on which they were not trained, compared to less good models. To do this, we fitted models as best as possible to everyone, leaving out certain test trials. Then we compared the sum-log-probability of the actual responses participants provided on these left-out trials, thus performing paired left-out-likelihood (LOL) comparisons. Confidence intervals around this difference of predictability-per-trial provide an intuitive measure of how much better one model is than another, especially when compared to the average predictability-per-participant-decision.
For the LOL comparison to be optimal, models must be given the best possible chance to describe the individuals whose parameters are used to derive the LOLs. In order to do this, we first divided the sample into a 300-strong ‘group training’ set and a test set. The ‘group training set’ was used to provide the best possible descriptive statistics of the entire population in terms of the means and variances of each (transformed) parameter. These were the group-level parameters fitted by the type-2 maximum-likelihood procedure (S3 Fig).
We then used these group-level parameters to provide priors for fitting the remaining ‘test set’, from which the ‘test trials’ were left out. Markov-chain monte-carlo (MCMC) fitting at the level of the individual participant was performed, using the sum-log-posterior over the included, non- left-out, trials only to derive posterior beliefs about parameters for each participant in the test sample. During fitting, MCMC efficiently provided sum-log-likelihood samples over the left-out-trials, thus forming the integrated LOL. Two trials were left-out, as described below. LOLs were not taken into account for parameter estimation, to avoid double-dipping. We repeated the procedure with different candidate models, thus obtaining (paired) model comparisons of their predictive power over the hidden trials only.
For optimal performance, we first decided how many trials to leave out (left-out trials, LOTs). When a limited amount of data per participant is available, greater numbers of LOTs result in noisier parameter estimates for each model, making it more difficult to detect differences between models. Furthermore, it is not a priori known how model fit may deteriorate as a function of the number of LOTs for different models. Hence, it makes sense to use the minimum number of LOTs and rely on our high number of participants to power model comparisons. In order to assess whether these considerations were important in practice, we generated synthetic data using the valenced-sensitivity model, the best in the literature. Using synthetic data we compared LOL using the procedure above with the true LOL according to the generative parameters. S4 Fig shows how increasing the number of LOTs significantly degraded the power of the true generative model to explain LOT data.
Next, we assessed the effect of learning on LOL estimation. Because learning occurs in every trial, learners follow different trajectories in the included trials depending on what happens in the left-out trials. Thus, even if the LOL is not used during model fitting, information from the LOTs may influence the fitted parameters, thus potentially biasing fitted parameters towards values most consistent with the participant’s choices (and hence high likelihood thereof) in the LOTs. Thankfully, we do not have to guard against every possible such influence, but only to make sure that using information from the LOTs (which stimuli where shown, which responses were performed and which returns were obtained) does not unduly bias the estimation of parameters based on the included trials towards values that make the LOTs appear more likely. In order to reassure ourselves about this, we performed a series of numerical experiments where we compared using the information above, to marginalizing over the above responses and rewards.
Consider a model M with a single parameter ε. Assume, furthermore, that we have a flat prior p0(ε)~1 over this parameter. If h is the left-out, or hidden, decision data and v is the included, or visible to the model, data, taking into account the flat prior gives:
p(h|v,M)=∑εp(h,ε|v,M)=∑εp(h|ε,M)p(v|ε,M)∑ε'p(v|ε',M)
(6)
The question is how p(v|ε,M) ought to be calculated in order not to bias estimation of how good model M is, i.e. not to bias p(h|v,M). As learning takes place from trial to trial, should the 'gaps' in v be filled in with the veritable choices of the participant, or be marginalized over? To investigate this matter, we first performed a numerical experiment with a simple Rescorla-Wagner model with learning parameter ε, making binary choices between alternatives via a softmax function of known parameter τ = 0.1 (i.e. a bare-bones version of our models). We generated 10000 x 28 trial epochs, for three levels of ε = 0.05, 0.15 and 0.25. Returns were deterministic returns (action1 → 1, action2 → 5), starting values for each run: Q0(action1) = Q0(action2) = 0. The first 8 trials were hidden. We looked for bias by examining how our estimate of p(h|v,M) depended on whether p(v|ε,M) is estimated using ‘informed’ visible trials, p(i)(v|ε,M), or ‘agnostic’ ones (i.e., marginalizing over hidden trials), p(a)(v|ε,M). A typical example of how p(i)(v|ε,M) may differ from p(a)(v|ε,M) is shown in S5 Fig. Although they gave different results for each individual subject, there was no difference (and no bias) with respect to the estimates over the hidden trials for any level of ε examined. This is shown in S6A Fig. It is interesting to note that some simple measures were biased in the expected way; for example, the maximum-likelihood estimate of the learning rate based on the ‘informed’ method was closer to the maximum-likelihood estimate over the hidden trials compared to the equivalent ‘agnostic’ estimate. In the case of ε = 0.15 this was by 0.03 log units, Wilcoxon p < 1e-08.
We then examined two further models, an η-greedy learner and an ‘observation-violating η-greedy’ learner. The latter was similar to the former, but, importantly, only updated action values for exploratory actions if they furnished a better-than-expected prediction error. We did not detect any bias in the simple η-greedy but we found a very small bias in the expected direction for the observation-violating model. The bias corresponded to 0.45% of the grand mean prediction probability. Given that this sequence of models was designed to showcase a difference between the more rigorous agnostic and the more practical informed approach, we concluded that any bias introduced by using the informed approach on our real data would be negligible.
We first compared longitudinal change with paired nonparametric tests. We also examined change using latent-change-score (LCS) models [36]. To do this we transformed the distributions of each (already transformed as above) parameter at each timepoint to normal as described below. LCS models formulate the change between baseline and follow-up as a latent variable, and estimate its mean, variance but also its possible dependence on baseline values. Pure regression towards the mean endows this change parameter with a value of -1. We used BIC and the likelihood ratio test to compare this model with nested, simpler models of change in the population distribution of parameters. Before we applied the latent-change-score formulation (S3 Appendix), we forced the marginal distributions of the transformed parameters into a Gaussian form. This further transformation-to-Gaussian to was achieved by first, estimating the mean and SD of the parameter distribution in question. Second, estimating an empirical (stepped) cumulative distribution function for this parameter with the R function edcf [50]. Third, applying an inverse-gaussian-cdf with the same mean and SD as the original. We then applied the just-identified univariate latent change score model.
This formulation does not extract from the data more than the statistics we might otherwise estimate–i.e. the means, variances and covariance of the baseline and long follow up measures, if we assume a bivariate normal distribution. It is however convenient in order to focus on change and phrase different hypotheses in terms of model comparison (BIC, likelihood ratio etc.).
We used the ‘Mood and Feelings Questionnaire’ (MFQ) as measure of mood and the Revised Children’s Manifest Anxiety Scale—2 (RCMAS) as a measure of anxiety [28,51,52]. General intelligence was measured by the full-scale IQ of the Wechsler Abbreviated Scale of Intelligence [53]. Measurements of IQ were performed on the same day as the task for the naturalistic longitudinal study. We used MFQ measurements taken near to the baseline testing session.
|
10.1371/journal.ppat.1003822 | Characterisation of a Multi-ligand Binding Chemoreceptor CcmL (Tlp3) of Campylobacter jejuni | Campylobacter jejuni is the leading cause of human gastroenteritis worldwide with over 500 million cases annually. Chemotaxis and motility have been identified as important virulence factors associated with C. jejuni colonisation. Group A transducer-like proteins (Tlps) are responsible for sensing the external environment for bacterial movement to or away from a chemical gradient or stimulus. In this study, we have demonstrated Cj1564 (Tlp3) to be a multi-ligand binding chemoreceptor and report direct evidence supporting the involvement of Cj1564 (Tlp3) in the chemotaxis signalling pathway via small molecule arrays, surface plasmon and nuclear magnetic resonance (SPR and NMR) as well as chemotaxis assays of wild type and isogenic mutant strains. A modified nutrient depleted chemotaxis assay was further used to determine positive or negative chemotaxis with specific ligands. Here we demonstrate the ability of Cj1564 to interact with the chemoattractants isoleucine, purine, malic acid and fumaric acid and chemorepellents lysine, glucosamine, succinic acid, arginine and thiamine. An isogenic mutant of cj1564 was shown to have altered phenotypic characteristics of C. jejuni, including loss of curvature in bacterial cell shape, reduced chemotactic motility and an increase in both autoagglutination and biofilm formation. We demonstrate Cj1564 to have a role in invasion as in in vitro assays the tlp3 isogenic mutant has a reduced ability to adhere and invade a cultured epithelial cell line; interestingly however, colonisation ability of avian caeca appears to be unaltered. Additionally, protein-protein interaction studies revealed signal transduction initiation through the scaffolding proteins CheV and CheW in the chemotaxis sensory pathway. This is the first report characterising Cj1564 as a multi-ligand receptor for C. jejuni, we therefore, propose to name this receptor CcmL, Campylobacter chemoreceptor for multiple ligands. In conclusion, this study identifies a novel multifunctional role for the C. jejuni CcmL chemoreceptor and illustrates its involvement in the chemotaxis pathway and subsequent survival of this organism in the host.
| Bacterial chemotaxis is an important part in initiation of colonisation and subsequent pathogenicity. In this study, we report direct evidence supporting the involvement of C. jejuni transducer-like protein Cj1564 (Tlp3) in the chemotaxis signalling pathway via small molecule arrays, surface plasmon and nuclear magnetic resonance (SPR and NMR) as well as chemotaxis assays of wild type and isogenic mutants. We further demonstrate its ability to interact with chemoattractants isoleucine, purine, malic acid and fumaric acid and chemorepellents lysine, glucosamine, succinic acid, arginine and thiamine. This is the first report identifying Cj1564 as a multi-ligand receptor for Campylobacter jejuni and its signal transduction initiation through the CheV and CheW proteins. Finally, our characterisation of C. jejuni Cj1564 provides additional basis for elucidating the roles of other group A chemoreceptors and their importance in the chemotaxis signalling pathway.
| Campylobacter jejuni is the one of the prevalent causes of acute human bacterial gastroenteritis worldwide [1]–[3]. C. jejuni is commonly found in the gastrointestinal tract of birds and poultry as commensal microbial flora [1], with infections in humans usually occurring from consumption of undercooked poultry, unpasteurised milk or untreated water. Symptoms include development of abdominal pains, fever and diarrhoea which can contain blood and leukocytes [4], [5]. Additionally Campylobacter enteritis is associated with post infectious complications ranging from reactive arthritis or reactive myositis to the more severe Guillain-Barré syndrome [1], [6]–[8].
The development of disease depends on the ability of bacteria to adapt to the environment of the human gut [9]. To date, it is known that factors associated with virulence and pathogenicity of C. jejuni include iron acquisition, chemotaxis, adherence and lipooligosaccharides (LOS) [5], [10], [11]. C. jejuni is known to be highly motile in viscous environments, with motility and swimming velocity increasing with increasing viscosity [12]. Chemotaxis, the ability of bacterial cells to detect temporal changes in the chemical concentration of their surrounding environment, and flagella-mediated motility, have been reported to play an important role in the intestinal colonisation of avian and mammalian hosts, as well as the invasion of intestinal epithelial cells [10], [12]–[17]. Furthermore, Hugdahl et al. (1988) have identified chemoattractants that C. jejuni is preferentially motile towards, which include amino acids found in the gastrointestinal tract, organic acid intermediates of the TCA cycle and components of mucous such as mucin. [18]. The importance of C. jejuni chemotaxis and motility in colonisation and pathogenicity has previously been shown with non-motile mutants defective in chemotaxis, unable to colonise and cause disease in the gastrointestinal tract of mice [13], [19]–[21]. Furthermore, strains with mutations in chemotactic genes and flagella associated genes were not capable of colonising chicken caeca [14] and lost the ability to colonise rabbits [22] and ferrets [20]. Another important observation was that mutants defective in chemotactic motility also lost the ability to autoagglutinate, adhere and invade mammalian cells [12], [15], [16], [23], [24].
The fundamental components of the chemotaxis signalling pathway, which are conserved in all motile prokaryotes, consist of the chemoreceptors, a cytoplasmic histidine kinase, CheA, a coupling or scaffolding protein, CheW and/or CheV and a response regulator, CheY. In the well-characterised chemotaxis system of E. coli, CheA is regulated by the chemoreceptors through association with CheW and uses ATP to autophosphorylate a specific histidine residue. The phosphoryl group is subsequently transferred to the response regulator CheY [25]. Phospho-CheY interacts with the flagella motor to induce clockwise (CW) rotation resulting in a tumbling motion, where the cell momentarily stops and randomly reorientates [25]–[28]. Attractant-bound chemoreceptors inhibit CheA kinase activity, with a resulting decrease in the levels of phospho-CheY. In contrast, repellent-bound or empty chemoreceptors stimulate CheA kinase activity, thereby increasing phospho-CheY levels and tumbling events, orientating the cell in a new direction [25], [26], [28], [29]. It is important to note that in C. jejuni, although the basic chemotaxis pathway backbone, consisting of Receptor-CheA-CheW(V)-CheY, is conserved, there are a number of differences to the paradigm E. coli model. C. jejuni encodes a two-domain CheA protein that includes a CheY-like response regulator domain in addition to the traditional histidine kinase domain, and a two domain CheV protein consisting of a CheW-like scaffolding domain and a CheY-like response regulator domain, as well as the paradigm CheW protein. C. jejuni also has a unique CheB protein that lacks a CheY-like response regulator domain found in all other bacterial chemotaxis pathways characterised to date [30].
Ten chemoreceptors have been identified in C. jejuni with homology to the methyl-accepting chemotaxis proteins (MCPs) in E. coli and have been designated Transducer-like proteins, Tlps. Group A Tlp receptors include CcaA (Tlp1), Tlps 2, 3, 4, 7 and 10 which have similarities to E. coli MCP structures and family A transducers of Halobacterium salinarium [30], [31]. Group A Tlps are thought to sense extracellular ligands and consist of a periplasmic sensory domain which is variable between different receptors, two transmembrane domains, and the highly conserved C-terminal cytoplasmic signalling domain [30], [32]. The sensory domain of each Tlp appears to be unique from that of non-Epsilon proteobacterial chemoreceptors; therefore sequence homology alone is not sufficient to determine specific ligand or ligands for each of the receptors. To date two of the group A Tlp receptors, Tlp1 and Tlp7, have been characterised [33]. CcaA was identified as the only receptor conserved in all sequenced strains of C. jejuni [34] and determined to be the receptor for aspartate [35] whereas Tlp7 binds to formic acid [36].
In this study we describe the characterisation of the periplasmic sensory domain of Cj1564 (Tlp3) chemoreceptor of C. jejuni strain NCTC 11168-O and characterise the ligand binding specificities of this chemoreceptor protein. We demonstrate the multifunctional role of Cj1564 in recognition of chemoattractants as well as chemorepellents, its association with the scaffolding proteins CheV and CheW and demonstrate the role of Cj1564 in cell to cell adhesion (autoagglutination) and biofilm formation, cell shape and host colonisation.
To gain a greater understanding of the C. jejuni chemotactic pathway, characterisation of ligand binding potential for Cj1564 (Tlp3) was performed using recombinant Tlp3 periplasmic sensory domain peptide with all 20 amino acids and salts of organic acid arrays as well as 96-well binding assays (Table 1). The array assay presents the compounds covalently bound to an epoxide group resulting in a high probability of a single molecular orientation. The plate assay relies on non-covalent charge and hydrophobic/hydrophilic interactions for presentation of the molecules offering a different presentation to the array. A total of 12 interacting ligands were identified through the plate and array assays. The binding of the Tlp3 sensory domain to amino acids and salts was then confirmed by STD-NMR and SPR (Biacore, Table 1). SPR analysis found the highest affinity interactions of Tlp3 with lysine and glucosamine (KD<10 µM; Table 1), additionally, biologically significant interactions (KD<50 µM; Table 1, Figure S1) were observed for isoleucine, succinic acid, arginine, purine, malic acid and thiamine.
Cj1564 (Tlp3) has previously been identified as a group A chemoreceptor for C. jejuni with homology in the signalling domain to other bacterial chemoreceptors such as that of E. coli and H. pylori [37]. The periplasmic domain of C. jejuni group A chemoreceptors are predicted to be involved in sensing and binding of extracellular ligands [37]. In order to identify biologically significant ligand binding specificity and function of Tlp3 in the chemotactic pathway, an insertionally inactivated isogenic tlp3 mutant and its complement were created. The mutant was constructed by deleting 52 central bp of the periplasmic domain of tlp3 and inserting a non-polar kanamycin resistance cassette which has the transcription terminator removed to minimise effects on genes downstream that may be transcribed in the same orientation as tlp3 (described in materials and methods: mutagenesis and complementation of tlp3). Microscopic analysis of Δtlp3 confirmed the presence of flagella and altered (loss of spiral shape) cellular morphology of all observed 11168-O Tlp3 mutant bacterial cells, with spiral morphology restored after complementation in approximately 50% of the complemented cells (Figure 1A, B & C). Comparison of C. jejuni 11168–O and Δtlp3 motility shows that there was a 5-fold decrease in non-directed swimming motility of the mutant with motility partially restored to wild type levels in the complemented strain, Δtlp3c (Figure 1D). Live imaging of fluorescently labelled C. jejuni isogenic strains allowed capture of bacterial cell motility and demonstrated a defect in motility illustrated by the inability of Δtlp3 cells to effectively ‘swim’ in a directional movement in absence of specific stimuli. Instead the cells were observed to ‘twitch’ and continuously ‘tumble’ without movement in any particular direction (Movie S1), thus demonstrating an altered “random walk” motility phenotype of the Δtlp3 mutant. The expression of the tlp3 gene in the isogenic mutant strain was reduced, as compared with the wild type and complemented strains, which is likely to be due to the interruption of the gene with the strongly promoted KmR cassette. Expression of tlp3 was analysed using quantitative real time PCR in 11168-O, Δtlp3 and Δtlp3c using primers that amplify the entire periplasmic domain gene region of tlp3. There was a 6.5±0.78 fold reduction observed in the expression of tlp3 in the Δtlp3 mutant strain compared to the wild type strain, conversely an 8.75±0.98 fold increase was observed when screening for tlp3 expression in the complemented mutant compared to wild type (data not shown). This indicates that the inclusion of the resistance cassette into the tlp3 gene has negatively affected the expression of tlp3 from its own promoter. This indicated that the Kanamycin resistance cassette encodes the dominant promoter signal that inhibits production of RNA molecules containing the 5′ end of the tlp3 gene region, upstream from the kanamycin resistance cassette. While the insertion of complementing tlp3 and chloramphenicol resistance cassette into the pseudogene (cj0046) results in over expression of tlp3 compared to the wild type strain. Further expression analysis was performed on the genes immediately upstream (cj1563c) and downstream (pflA) of the mutated tlp3 gene to confirm no polar effects were introduced as a result of mutagenesis. Expression of pgp1 and flaA were assessed to establish that the observed defect in motility is not due to changes in expression of genes involved in flagella development or function. No significant difference in expression (p>0.1) was observed for cj1563c, pflA, pgp1 and flaA (data not shown).
Biofilm formation and agglutination play an important role in the survival of bacterial cells. Consequently, we compared these characteristics between the wild type (11168-O), mutant (Δtlp3) and complemented (Δtlp3c) strains. Autoagglutination assays revealed that the Δtlp3 autoagglutinated while 11168-O and Δtlp3c had little to no autoagglutination (Figure 2A & B). It is also interesting to note that autoagglutination of the 11168-O Δtlp3 mutant was similar to that of C. jejuni 81–176 (Figure S2), which carries a natural mutation in its tlp3 gene, revealed by the published genome sequence (Sanger, 2006). Furthermore, autoagglutination was also shown to be independent of growth temperature at 25°C, 37°C and 42°C for all strains tested (data not shown). Additionally it appears that Δtlp3 autoagglutinates at a rate that is faster than normal gravitational pull. Biofilm formation, as assessed by crystal violet assay, indicated an approximate 1.5-fold increase in biofilm formation of Δtlp3 mutant compared to wild type, with the original levels in biofilm formation restored in Δtlp3c (Figure 2C).
In order to determine the biological relevance of Tlp3 in chemotactic motility, nutrient depleted chemotaxis assays were performed as previously described [35]. Chemotaxis towards a range of amino acids, glycans and other small molecules was investigated in 11168-O, Δtlp3 and the complemented mutant, Δtlp3c. A positive chemotaxis response was identified for 5 amino acids, suggesting these ligands are attractants (Figure 3A; Table 1). A 4-log reduction in bacterial numbers was observed for the Δtlp3 (3.6 × 102 cfu/ml) compared to the wild type (6 × 106 cfu/ml) in migration towards isoleucine, additionally this was also observed for fumaric acid (9 × 102 and 4.14 × 106 cfu/ml respectively). For migration towards purine, a 1-log reduction for the Δtlp3 (1.08 × 105 cfu/ml) compared to the wild type (1.4 × 106 cfu/ml) was observed, with a 2-log reduction in migration towards malic acid (1.13 × 103 and 9 × 105 cfu/ml respectively) and a 3-log reduction in migration towards aspartate (3.7 × 103 and 7.8 × 106 cfu/ml respectively). The migration of wild type C. jejuni 11168-O towards aspartate was comparable to that previously published for nutrient depleted assay by Hartley-Tassell et al., 2010 [38]. Tlp3 was also found to mediate a repellent response to 5 amino acids (Figure 3B; Table 1). Lysine mediated repellence was reduced for the Δtlp3 isogenic strain as a 2-log increase in viable bacterial numbers was detected around the lysine plug (2.4 × 105 cfu/ml) when compared to the wild type (2.2 × 103 cfu/ml). For migration of Δtlp3 towards glucosamine, a 3-log increase (7.8 × 106 cfu/ml) was detected, compared to that of the wild type (1.68 × 103 cfu/ml). In addition, for succinic acid, arginine and thiamine, a 1-log increase of bacterial numbers was observed in the Δtlp3 (2 × 107, 6.8 × 107 and 9 × 106 cfu/ml, respectively) compared to the wild type strain (1.5 × 106, 1.3 × 106 and 4 × 105 cfu/ml, respectively). Chemotaxis assay for alpha-ketoglutarate was not definitive (data not shown), possibly due to the low affinity of this receptor for alpha-ketoglutarate.
1H STD-NMR analysis was used to investigate the epitope binding preferences of the chemoattractants (isoleucine, purine, malic acid) and chemorepellents (lysine, arginine, glucosamine) to the recombinant Cj1564 (Tlp3) periplasmic sensory domain peptide. Binding was observed with chemoattractants isoleucine and purine along with the chemorepellents lysine and arginine in accordance with the SPR analysis showing varying binding affinities. For the chemoattractant isoleucine (KD∼17 µM), the binding epitope appeared to be the methyl group, as strong methyl resonances were seen in the STD spectrum and only very weak signals for other side chain resonances (Figure 4A). For purine (KD∼38 µM), the H6 and to a lesser extent the H2 protons (∼5-fold) of the pyrimidine ring showed an STD effect compared to a negligible STD effect for the H8 proton of the imidazole ring. For the chemorepellent lysine (KD∼2.8 µM) an STD effect was clearly seen for all proton resonances along the side chain (Figure 4B). A weaker STD effect (∼4 fold) was seen for all proton resonances along the arginine (KD∼38 µM) side chain. No signals were detected for malic acid or glucosamine using STD-NMR. SPR analysis demonstrated that glucosamine and malic acid have slow disassociation rates explaining the absence of STD signals due to these amino acids as saturation transfer is a result of the spin diffusion process which requires fast exchange for magnetisation to be spread from the protein to the ligand in order to observe an STD effect.
Competition STD experiments were also performed to probe the binding preferences of Tlp3 and to unravel the nature of the binding interaction. Competition experiments testing the effect on binding to Tlp3 of attractant/attractant (isoleucine/purine), repellent/repellent (lysine/arginine) and attractant/repellent (isoleucine/arginine; isoleucine/lysine; purine/lysine; purine/arginine) were performed. For attractant/attractant (isoleucine KD∼17 µM/purine KD∼38 µM) there seemed to be a minor change in the STD effect seen for the ligands independently. For the repellent/repellent (lysine KD∼2.8 µM/arginine KD∼38 µM) however, there appeared to be a significant reduction in the STD effect of lysine in the presence of arginine whose weak STD effect appeared unchanged, even though the binding affinity of lysine is stronger (∼14 fold) than that of arginine. These results demonstrate that Cj1564 (Tlp3) is able to bind to both chemoattractants and chemorepellents. No binding preference to Cj1564 (Tlp3) was observed in the presence of the chemoattractants isoleucine/purine, compared with the ligands alone, however for the chemorepellent competition experiment even though lysine has a greater binding affinity (KD∼2.8 µM) compared with isoleucine (KD∼17 µM) a significant reduction in the STD effect of lysine was observed in the presence of arginine suggesting preferential binding of arginine over lysine.
In the case of the attractant/repellent spectra: Isoleucine/arginine - isoleucine (chemoattractant, KD∼17 µM) binds preferentially over arginine (chemorepellent, KD∼38 µM). For isoleucine/lysine - isoleucine (chemoattractant) binds preferentially over lysine (chemorepellent) with a significant reduction in the STD effect of lysine observed in the presence of isoleucine compared to the STD spectra of the ligands individually, even though the binding affinity of lysine (KD∼2.8 µM) is stronger than that of isoleucine (KD∼17 µM) (Figure 4C & D). Purine/lysine - purine (chemoattractant) binds preferentially over lysine (chemorepellent) with a significant reduction in the STD effect of lysine in the presence of purine observed compared to the STD spectra of the ligands individually, even though the binding affinity of lysine (KD∼2.8 µM) is stronger than that of purine (KD∼38 µM). Purine/arginine - purine (chemoattractant) binds preferentially over arginine (chemorepellent), a negligible STD effect of arginine in the presence of purine was observed with these ligands determined to have comparable binding affinities (KD∼38 µM). These results demonstrate that there is preferential binding of chemoattractants over the chemorepellents regardless of the individual binding affinity of the ligands to Cj1564 (Tlp3).
To further investigate the role of Tlp3 in the chemotaxis signal transduction pathway in C. jejuni, the predicted cytoplasmic signalling domain of Tlp3 was analysed for protein-protein interactions with C. jejuni NCTC11168-O chemotaxis proteins using the yeast two-hybrid and three-hybrid systems as previously described [35]. The yeast two-hybrid system was used because of a lack of similar genetic manipulation systems available for C. jejuni. We have previously demonstrated biological validity of protein interactions identified by the yeast two-hybrid system via pull-down assays for the components of the C. jejuni aspartate receptor, CcaA [35] signal transduction. The yeast system allows detection of interacting proteins in vivo by utilising the two separable domains of the GAL-4 transcription factor, the DNA-binding domain (DNA-BD) and the transcription activation domain (AD). This system relies on the reconstitution of the GAL-4 transcription factor when two proteins of interest, the ‘bait’ protein fused to the DNA-BD and the ‘target’ protein fused to the AD, interact, thus allowing activation of reporter gene expression [40], Residues 517–662 of Tlp3, encompassing the region homologous to the highly conserved MCP signalling domain, were selected for analysis. The signalling domain is an independent structural motif, which interacts with the dimeric CheA and the scaffolding protein CheW to form MCP-CheW-CheA ternary signalling complexes that regulate CheA histidine kinase activity [41]–[43]. It is necessary to note that this region of Tlp3 is identical to residues 513–659 of Tlp2 (Cj0144) and 520–665 of Tlp4 (Cj0262) [37] and thus is referred to as Tlpsig2.3.4 in this study (Figure S5). The Tlpsig2.3.4 was found to interact with a number of the C. jejuni NCTC11168-O chemotaxis proteins and protein domains, as shown in Table 2. A medium strength interaction was detected between Tlpsig2.3.4 and CheV (AD-Tlpsig2.3.4 with BD-CheV). An identical interaction was observed in the reciprocal comparison of fusion proteins (AD-CheV and BD-Tlpsig2.3.4). The Tlpsig2.3.4 was found to bind only to the CheW-like domain of CheV (CheVdW). A medium strength interaction was detected between these domains in reciprocal comparisons (AD-Tlpsig2.3.4 with BD-CheVdW and AD-CheVdW and BD-Tlpsig2.3.4) while no interaction was detected with the response regulator domain of CheV (CheVdRR and Tlpsig2.3.4). Two-hybrid analysis also revealed that the Tlpsig2.3.4 signalling domain was capable of dimerisation (AD-Tlpsig2.3.4 and BD-Tlpsig2.3.4). An interaction was also observed between Tlpsig2.3.4 and CheW. This interaction was of a medium strength (AD-CheW and BD-Tlpsig2.3.4); however this interaction was not detected with the reciprocal combination of fusion proteins.
As it has previously been shown that CcaA (Tlp1) of C. jejuni preferentially binds to CheV over CheW [35], we utilised the yeast three-hybrid system to determine if the signalling domain of Tlp3 (which is identical to the signalling domains of Tlp2 and Tlp4) displayed similar binding preferences for the scaffolding proteins. In the three-hybrid system, a third protein can be conditionally expressed and its role in the interaction between the AD and DNA-BD fusion proteins can be examined [44]. We specifically looked at the effect of presence of both CheW and CheV proteins on their interactions with Tlpsig2.3.4. It was found that the co-expression of native CheW from pBrTlp234IWII (Table 3) had no effect on the medium strength interaction observed between Tlpsig2.3.4 and CheV (BD-Tlpsig2.3.4 and AD-CheV). Analysis of interactions between the reciprocal combinations of fusion proteins using pBrVIWII (Table 3) showed that the co-expression of native CheW slightly enhanced the medium strength interaction of CheV with Tlpsig2.3.4 (BD-CheV and AD-Tlpsig2.3.4). When protein interactions of Tlpsig2.3.4 with CheW were investigated with native CheV co-expressed (pBrTlp234IVII), it was found that the presence of CheV strengthened the weak interaction observed between the Tlpsig2.3.4 signalling domain and CheW (BD-Tlpsig2.3.4 and AD-CheW). This was confirmed in the analysis of the interactions between the reciprocal combination of fusion proteins (AD-Tlpsig2.3.4 and BD-CheW) using pBrWIVII.
In this study we have shown the role of Tlp3 in C. jejuni 11168-O chemotaxis as a multiple ligand-binding protein, capable of detecting numerous chemoattractants and chemorepellents using a range of confirmatory methodologies such as small molecule arrays, SPR and STD-NMR. The substrates identified as having a strong binding affinity include lysine and glucosamine with biologically significant interactions observed for isoleucine, aspartate, succinic acid, arginine, purine, malic acid, and thiamine. Therefore we propose to name this receptor CcmL; Campylobacter chemoreceptor for multiple ligands.
Similar observations have previously been made for H. pylori isolates, where positive chemotaxis was identified for the sugars and amino acids phenylalanine, aspartic acid, glutamic acid, isoleucine and a negative chemotaxis for leucine and tyrosine [45]. Isoleucine has previously been demonstrated to be a chemoattractant for other organisms including B. subtilis [46], H. salinarum [47], P. aeruginosa [48] and H. pylori [45]. In B. subtilis, the McpC chemoreceptor binds isoleucine weakly, yet with sufficient affinity to suggest direct binding. It is interesting to note that in our study, glucosamine has been identified as a repellent, while in other bacteria such as E. coli [49] and B. burgdorferi it has been determined to be an attractant, and has been classified as a non-essential nutrient [48], [50]. Lysine has been demonstrated as an attractant for P. aeruginosa [48] which is opposite to our findings which show lysine as a chemorepellent for C. jejuni. Furthermore, in B. subtilis, the McpC chemoreceptor did not show binding to lysine, however, it was suggested that McpC binds lysine by an indirect method most likely involving ancillary proteins, further suggesting McpC may be a universal chemoreceptor able to respond to numerous amino acids [51]. It had been shown in E. coli that even though chemoreceptors are sensitive to a particular ligand, they can also detect a large number of structurally related amino acids and their analogues [52].
C. jejuni is highly adapted to the environment of the avian gut and as a result, uses efficient chemotactic motility to colonise the mucous-filled crypts of the lower gastrointestinal tract [53]. It is reasonable to hypothesise that Tlp3 may be involved in interacting with ligands to sense the external environment in order to navigate inside the host and may be involved directly or indirectly in C. jejuni metabolic and catabolic pathways. It is interesting to note that aspartate, glutamate, proline and serine are the most abundant amino acids found in chicken excreta [54], and serine catabolism has been reported to be essential for colonisation of the avian gut by C. jejuni [55]. However, C. jejuni lacks the key glycolytic enzyme 6-phosphofructokinase as well as alternative pathways for sugar catabolism [56] so it utilises amino acids and Krebs cycle intermediates for energy production and encodes all of the enzymes required for a complete oxidative TCA cycle [37], [57], [58]. C. jejuni needs to be able to sense and move towards amino acids and small organic acids such as aspartate, asparagine and serine in order to catabolise these compounds to use as the sole source of reduced carbon and energy due to its inability to utilise glucose [59]. Additionally, a recent study has indicated that energy taxis may also be one of the driving forces behind movement to optimal conditions for energy generation and subsequent colonisation [60], [61].
We have shown that a mutation in the C. jejuni ccmL chemoreceptor gene lead to alteration of phenotypic characteristics of the bacteria, such as cellular morphology, autoagglutination behaviour and biofilm formation, highlighting its role in the C. jejuni life cycle. Furthermore, the signalling domain of this chemoreceptor also interacts and binds with both CheV and CheW scaffolding proteins.
Additionally, we used STD-NMR analysis to confirm the binding of amino acid and salts of organic acid arrays, plate binding assays and SPR analysis. STD-NMR indicated that the Tlp3 receptor binds chemoattractants preferentially over chemorepellents irrespective of binding affinity, however, is still able to recognise both attractants and repellents in isolation. This may indicate that upon binding an attractant, the repellent binding site of CcmL through allostery prevents the binding of a repellent ligand to the protein. Furthermore, it appears that the binding site may be able to accommodate more than one chemoattractant at a time as competition tests between two chemoattractants results in positive binding for both attractants present; indicating that one chemoattractant is not preferred over another regardless of the measured affinity. These results indicate a much more complex interaction between chemotaxis receptor proteins and ligands than previously thought, with a complex milieu of attractants acting in concert rather than the receptor binding preferentially to a specific ligand based on the hierarchy of affinities. Furthermore, the Tlp3 predicted structure indicates the presence of a single Cache_1 (PFO2743) (calcium channels and chemotaxis receptors) domain, representing a single ligand binding pocket that accommodates multiple ligands with varying affinity. This is in agreement with our STD-NMR data that shows interchange of the ligands binding to purified periplasmic domain of CcmL in competitive assays. CcmL (Cj1564) shares complete homology in the cytoplasmic domain with Tlp2 (Cj0144) and Tlp4 (Cj0262), but has less than 50% homology with the cytoplasmic domains of Tlp1 (Cj1506), Tlp7 (Cj0951/52) and Tlp10 (Cj0019). Tlp3 has the greatest homology to Tlp2 with 72% identity across the entire protein and 38% identify/59% similarity across the periplasmic domain. The greatest stretch of similarity between Tlp3 and Tlp2 in the periplasmic binding region is within the cache domain that is present in both proteins. Homology between the periplasmic domain of Tlp3 and the periplasmic domains of Tlp1 and 4 is below 33%. No homology was detected between Tlp3 periplasmic domain and the periplasmic domains of Tlp7 and 10.
To further characterise interactions of Tlp3 with components of the C. jejuni chemotaxis signalling pathway, a well-established yeast-two-hybrid and three-hybrid protein-protein interaction system was used. The two- and three-hybrid system data suggests that both the scaffolding proteins CheW and CheV are capable of binding to the Tlps2, 3 and 4 signalling domains with no obvious preference for either protein observed. CheW and CheV may form mixed multi-protein complexes with these receptors, as the expression of CheV was found to strengthen the interaction between the signalling domain of Tlpsig2.3.4 and CheW. This appears to vary from the observations for the CcaA (Tlp1) signalling domain, which has previously been shown to preferentially bind CheV [35]. While the signalling domains of Tlps 2, 3 and 4 and of Tlp1 are very similar, some amino acid differences do exist which may account for the differences in the binding preferences of these Tlps for the CheW and CheV scaffolding proteins. These amino acid differences are within the region homologous to residues 350–471 of the E. coli serine chemoreceptor, Tsr, a fragment which has been shown to be capable of mediating CW-biased signals and is therefore predicted to contain residues involved in the binding of CheW and CheA in this species [62] (Figure S5). This data lead us to speculate that in C. jejuni, the group A Tlp2, 3 and 4 signal via alternative or mixed scaffolding proteins and that different binding affinities of the chemoreceptors with CheV and CheW may control the composition of receptor clusters.
The mutation in the tlp3 (cj1564) gene resulted in an altered cellular morphology of the bacterial cell and inability of Δtlp3 cells to ‘run’. Increased formation of biofilm and the high rate of autoagglutination may indicate a possible increased response to stress by the tlp3 mutant. A previous study by Vegge et al. (2009) reported that mutation of the tlp3 gene along with other genes does not affect the motility of C. jejuni in a rich medium [63] however; our results indicate mutation of tlp3 significantly reduces motility when compared to the wild type and complemented strains. This is in agreement with a study carried out by Golden et al. (2002) [64] where mutation of tlp3 reduced motility and complementation restored the wild type phenotype [23]. Furthermore, reduced motility is also likely due to an inability of the bacteria to engage in efficient chemotaxis signalling. Further inspection of the isogenic Δtlp3 mutant cells by microscopy indicated that the bacteria retained the ability to twitch and tumble in a stationary position suggesting a bias toward clockwise rotation of the flagella, subsequently leading to a higher frequency of tumbles and inhibited smooth swimming. It appears however, that the bacterial cell was still capable of directional movement when a stimulus was supplied in vivo. Additionally, scanning electron microscopy and quantitative PCR analysis confirmed the presence of flagella indicating that the lack of motility was not due to a defect in the flagella development; however, the helical shape of the bacteria was altered. The helical shape of C. jejuni has long been associated with pathogenesis but the genetic components involved in modulating C. jejuni morphology have only recently been identified, where the protein termed Pgp1 (peptidoglycan peptidase 1) in C. jejuni 81–176 was identified to be involved in maintenance of the helical shape [65]. To date no correlation has been found between chemoreceptors and cellular morphology.
In the in vitro model of infection, it appears that Tlp3 may be a critical factor for invasion as Δtlp3 displayed a markedly reduced level of adherence and invasion compared to the wild type strain. However, in vivo assessment of the Δtlp3 mutant and wild type strain showed no differences in colonisation ability of the avian caeca based on cell counts, even though ability to adhere and invade caco-2 cells in vitro was significantly reduced. This finding agrees with the previously published data on avian colonisation for the Tlp3 mutant of C. jejuni 81–176 [66]. However, it is important to consider that subsequent publication of the genome sequence of 81–176 (Sanger, 2006) revealed a natural mutation of the Tlp3 homologue in 81–176 which exists as 2 separate reading frames: CJJ81176_1548 and CJJ81176_1549 encoding the majority of the periplasmic and cytoplasmic domains of Tlp3, respectively. There is a nucleotide deletion in the sequence of CJJ81176_1548 at position 1467624 (A) in the genome subsequently altering the amino acid sequence, consequently changing the reading frame of the transmembrane and cytoplasmic domains (CJJ81176_1549) resulting in incorrect translation and production of a non-functional Tlp3 protein in 81–176. There are an additional number of mutations in both the remainder of the periplasmic and cytoplasmic domains with several intervening stop codons. The absence of a functional Tlp3 in 81–176 has previously been reported [67].
It is interesting to note that C. jejuni 81–176 has the same complement of Tlp genes as 11168 and 81116, with the exception of Tlp3, which has a naturally occurring mutation in 81–176. When phenotypic characteristics pertaining to biofilm formation and autoagglutination of 81–176, and 11168-O are compared, a wild type 81–176 shows higher levels of aggregation, biofilm formation (data not shown) and autoagglutination, similar to that observed for Tlp3 mutant of 11168 (Figure S2).
Characterising the function of C. jejuni chemosensory proteins, as described in this study, will contribute to understanding chemotaxis signalling pathways which are involved in colonisation and further identify chemoreceptor ligand specificity of individual group A Tlp receptors and their involvement in the chemotaxis pathway and its importance in the survival of this organism. The findings in this study also provide insight into the complexity of chemotaxis receptor protein-ligand interactions with implications not just for C. jejuni chemotaxis but for all bacterial chemotaxis.
Animal experiments were carried out in strict accordance with the Griffith University Animal Ethics Committee guidelines and assigned approval number BDD/02/11. All procedures involving animals were reviewed and approved by National Health and Medical Research Council Australian code of practice for the care and use of animals for scientific purposes 7th edition 2004.
Bacterial strains, yeast strains and plasmids used in this study are described in Table S1. C. jejuni strain 11168-O from Skirrow collection was kindly provided by DG Newell, Central Veterinary Laboratories, UK. C. jejuni strains were grown at 37°C or 42°C on Columbia agar supplemented with 5% defibrinated horse blood (HBA) with vancomycin (10 µg/ml), trimethoprim (2.5 µg/ml) and polymyxin B (2.5 IU/ml) under microaerobic conditions (5% O2, 10% CO2, 85% N2) for 18–24 h. Strains with plasmids for mutation or complementation studies were grown with 50 µg/ml kanamycin (Km) or 30 µg/ml chloramphenicol (Cm). Host E. coli BL21 DE3 (Novagen, USA) and E. coli DH5α (Novagen, USA) strains were grown in Luria-Bertani (LB) medium (Oxoid) supplemented with ampicillin (100 µg/ml), kanamycin (50 µg/ml) and chloramphenicol (30 µg/ml) when required. Host yeast strains were grown and prepared according to manufacturers' instructions (Clontech). Labelling of bacterial cells with CFDA-SE was performed as previously described [35], [68].
The DNA fragment encoding the periplasmic sensory domain (amino acids 43–290) of Tlp3 was amplified using primers incorporating start and stop codons (Table S2) and ligated into pGEM-TEasy to form pGU0815 (Table S1). The insert in pGU0815 was restricted at primer-specific NdeI and XhoI sites and subcloned into pET-19b to form pGU0816 (Table S1). For recombinant protein expression, competent E. coli BL21(DE3) cells were transformed with pGU0816. An overnight culture of BL21(DE3)/pGU0816 was used to inoculate LB containing ampicillin (100 µg/ml) that was incubated at 37°C with aeration. Protein expression was induced using 1 mM IPTG when OD600 nm reached 0.4–0.6. Expression of the Tlp3peri -His fusion protein was verified by SDS-PAGE and Western Blot analysis using Anti-His mouse IgG (Cell Signalling) as shown in Figure S6. The cell pellet was resuspended in PBS containing 6 M urea, lysozyme (0.2 mg/ml) and protease inhibitor cocktail mix (50 µl/ml) and incubated at room temperature for 1 h using a rotational mixer. The cells were sonicated and an additional freeze/thaw step performed to aid in cell lysis. The insoluble cell debris was removed by centrifugation at 38 000 rpm for 90 min. The clarified supernatant was added to 1.5 ml of TALON Metal Affinity Resin (Clontech) and rotated overnight at 4°C using a rotational mixer. The slurry mix was then packed by gravity into a 10 ml Bio-Rad chromatography column. The column was washed twice with PBS, then washed with PBS containing 20 mM imidazole, washed three times with PBS and the bound His-tagged protein was eluted with PBS containing 150 mM imidazole. Residual imidazole was removed from the sample using Econo-Pac 10DG Desalting Columns (Bio-Rad) according to manufacturer's specification. Purity was confirmed by analysis of samples by SDS-PAGE and Western Blot using anti-His antibodies (Bio-Rad). To ensure that the recombinant protein was properly folded, its properties were compared to denatured protein. Denatured protein recombinant Tlp3 was not soluble in PBS or TBS buffers, unlike correctly folded protein. Moreover, the binding affinities of recombinant protein following purification without urea were also tested. No change in binding affinities could be observed (data not shown).
Amino acid and small molecule arrays and plate based binding assays were performed as previously described. [35]. The ligands investigated by Surface Plasmon Resonance (SPR) included alpha-ketoglutarate, glucosamine, lysine, purine (basic ring), malic acid and L-isoleucine (Sigma). SPR experiments were performed using a Biacore T100 biosensor system (GE- healthcare) at 25°C in 1× PBS pH 7.2 at a flow rate of 30 µl/min. Purified His-Tlp3 was diluted to 0.15 µM in 1× PBS pH 7.2 and loaded on flow cell 2 (FC2) of a Ni2+ NTA sensor chip with 5 min contact time. FC1 had no protein loaded and was used as reference. Amino acids were prepared in 1 × PBS pH 7.2 and serially diluted from 0.0125 to 0.2 mM. The amino acids were loaded to the sensor chip using single-cycle kinetics, i.e. after the injection of five amino acid dilutions the chip was regenerated with EDTA. Subsequently, the chip was re-loaded with Ni2+ and His-Tlp3 before the injection of the next amino acid to be tested. A 10 min dissociation time was allowed after the addition of each analyte. SPR signals were analyzed using the Biacore Evaluation software to determine KD.
Recombinant Tlp3 (1 mg/mL, 100 µL) dissolved in D2O (99.99% D Cambridge Isotopes) containing 50 mM NaCl and 50 mM KH2PO4 and added to various ligands (1 mg/mL, 150 µL ∼150 mole equivalents) also dissolved in D2O (99.99% D Cambridge Isotopes) containing 50 mM NaCl and 50 mM KH2PO4 to give a total volume of 250 µL in a 3 mm NMR tube for NMR analysis. Control samples were prepared in an identical manner without Tlp3 added. All NMR experiments were performed on a Bruker Avance 600 MHz spectrometer, equipped with a 5-mm TXI probe with triple axis gradients at 283 K without sample spinning. 1H NMR spectra were acquired with 32 scans, a 2 s relaxation delay over a spectral width of 6000 Hz. Solvent suppression of the residual HDO peak was achieved by continuous low-power presaturation pulse during the relaxation delay. In the STD-NMR experiments of the chemoattractants (isoleucine, purine and malic acid) and chemorepellents (lysine, arginine and glucosamine) in complex with Tlp3, the protein was saturated at −0.5 ppm in the aliphatic region of the spectrum and off-resonance at 33 ppm with a cascade of 40 selective Gaussian-shaped pulses of 50 ms duration (50 dB), which correlates to a strength of 190 Hz. A 100 µs delay between each soft pulse was applied, resulting in a total saturation time of 2 s and 2 K scans.
Data were obtained with an interspersed acquisition of pseudo-two-dimensional on-resonance and off-resonance spectra in order to minimize the effects of temperature and magnet instability. On- and off-resonance spectra were processed separately, and the final STD-NMR spectrum was obtained by subtracting the individual on- and off-resonance spectra, resulting in less subtraction artefacts. Relative STD effects were calculated according to the equation ASTD = (I0−Isat)/I0 = ISTD/I0 by comparing the intensity of the signals in the STD-NMR spectrum (ISTD) with signal intensities of a reference spectrum (I0). The STD signal with the highest intensity was set to 100% and other STD signals were calculated accordingly. A spin lock field of 10 ms was applied to remove unwanted background protein signals. Increased spin lock fields resulted in artefacts and reduced ligand signal intensities. Control STD-NMR experiments were performed using an identical experimental setup and the same ligand concentration but in the absence of the protein.
The tlp3 gene (cj1564) in C. jejuni 11168-O was inactivated by inverse PCR mutagenesis [69] to produce the strain 11168-OΔtlp3::Km (Table S1). The tlp3 periplasmic domain (amino acids 43–290) was amplified from the 11168-O genome and cloned into pGEM-TEasy to produce pGU0815 (Table S1). Inverse PCR with primers designed to incorporate a BglII restriction site and delete 52 bp of the tlp3 periplasmic domain was followed by insertion of a non-polar kanamycin resistance cassette with a consensus campylobacter promoter from pMW10 [70] in the same orientation as tlp3 to generate plasmid pGU0817 (Table S1). A kanamycin resistance cassette which lacks the transcription terminator was used in order to minimise effects on genes downstream of tlp3. Each construct was verified by DNA sequencing and subsequently electro-transformed into the motile variant of 11168 original clinical isolate, donated by D.G Newell, VIR, London [71]. Replacement of the mutant allele was verified by PCR and DNA sequencing. The complement of this mutant was generated by inserting the complete tlp3 gene including promoter into the pseudogene cj0046 using the plasmid pC46 (Table S1). The tlp3 complement plasmid was electro-transformed into C. jejuni 11168-O with insertion into the pseudogene cj0046 locus to produce strain 11168-OΔtlp3::KmΩcj0046::Cm (Table S1). Complementation was confirmed by PCR and sequence analysis.
Motility assays were performed as previously described [23] with the following modifications: C. jejuni strains were grown microaerobically at 42°C for 24 h in 10 ml Brucella broth (Oxoid) on a shaker (50 rpm/min). Cells were collected by centrifugation at 3000 g for 5 min and washed once with Brucella broth. Equal numbers of bacterial cells (5 µl) of 1 × 109 cfu/ml of each strain of C. jejuni were stabbed on top of 0.35% Mueller Hinton Agar plates (MHA Oxoid) and incubated microaerobically at 42°C for 48 h. Autoagglutination (AAG) was performed as previously described [72]. Briefly the cells were harvested with 2 ml of phosphate-buffered saline (PBS pH 7.2) and OD600 nm adjusted to 1.0. The bacterial suspension was poured into sterile glass tubes (13 × 100 mm) and incubated at 25°C, 37°C or 42°C for 24 h. 1.0 ml of the upper aqueous phase was carefully aspirated and the OD600 nm measured. Additionally viable bacteria were enumerated by plate counts. The lower 1 ml of solution containing the majority of the autoagglutinated cells were analysed by scanning electron microscopy (SEM). Biofilm assays were performed according to Reeser et al [73] and Tram et al [74].
Assays were performed as previously described [20] with modifications. Approximately 100 µl of bacterial suspension containing 2 × 108–4 × 108 bacteria per ml was inoculated into 24-well plate containing a confluent monolayer of Caco-2 cells. A 5 min centrifugation step at 500 g was included to facilitate the movement of C. jejuni onto the surface of the cell monolayer. The cells were incubated for 45 min at 37°C in a 5% CO2 humidified atmosphere to allow passive adherence and internalization. For the adherence assay, the total number of bacteria associated with the cell layer was enumerated by viable count. For the invasion assay, infected cells were washed three times with PBS and 1 ml of MEM (Dibco) containing 400 µg/ml gentamicin was added to the cell monolayer for 3 h. The monolayer was washed three times with PBS and the cells lysed with 200 ml of 0.2% Triton X-100; intracellular bacterial counts were enumerated by viable count. Each assay was performed in triplicate.
Cells were grown microaerobically for 24 h at 42°C, collected and washed three times with Brucella broth by centrifugation at 1000 g for 4 min. The final wash step was carried out using PBS and the OD600 nm adjusted to 0.025. Bacteria were fixed on plastic cover slips with 2% glutaraldehyde and 5% formaldehyde solution for 10 min. Slides were washed three times with H2O and dehydrated in gradient steps of water/ethanol (15, 30, 50, 75, 90 and 100%). Last steps of dehydration were performed in 50% HMDS/ethanol solution (Hexamethyldisilazane), followed by a final step of 100% HMDS. Slides were left to air dry and subsequently coated with gold (6 nm) prior to analysis on Jeol 5000 Scanning Electron Microscope.
Chicken colonisation analysis was performed as described previously [39] with an infective dose of 108 CFU (BDD/02/11).
RNA was extracted using RNeasy kit according to the manufacturer's protocol (Qiagen). cDNA synthesis and RT-qPCR was performed using 11168-O, Δtlp3 and Δtlp3c as previously described [33].
Nutrient depleted chemotaxis assay was performed as previously described [35] with the following modifications. One or two plugs (6 mm diameter) equidistantly apart (60 mm) were removed from Petri dishes containing 0.5% agar in H2O. Each well was filled with 0.5% agar containing 2 mM of selected amino acid. The plates were overlayed with 0.1% agar bacteriological (Oxoid) without nutritional supplements and left for 2 h to allow for diffusion of amino acids to create a chemical gradient. Cultures of C. jejuni were adjusted to OD600 nm 1.8 (109 cfu/ml). 100 µl drop of bacterial suspension was inoculated using a micropipette on top of the sloppy agar (0.1% agar bacteriological in H2O) in the centre of the Petri dish with the plates incubated at 37°C for 4 h to allow chemotactic migration of the bacteria. To determine the number of viable bacteria associated with each amino acid plug, a 5 mm area around and including each plug was removed and placed into a microcentrifuge tube containing 900 µl of Brucella broth. These were incubated microaerobically for 1 h at 37°C to allow the bacteria to dissociate from the plug into the media. Viable counts were performed with enumeration by serial dilution followed by plate counts. C. jejuni 81116 flaA−/flaB− isogenic mutant was used as a non-motile, non-chemotactic control; additionally agar plugs containing no amino acid were used as a negative control.
Yeast two-hybrid and three-hybrid analyses of protein interactions were performed as described previously [35].
|
10.1371/journal.ppat.1000029 | Beyond PrPres Type 1/Type 2 Dichotomy in Creutzfeldt-Jakob Disease | Sporadic Creutzfeldt-Jakob disease (sCJD) cases are currently subclassified according to the methionine/valine polymorphism at codon 129 of the PRNP gene and the proteinase K (PK) digested abnormal prion protein (PrPres) identified on Western blotting (type 1 or type 2). These biochemically distinct PrPres types have been considered to represent potential distinct prion strains. However, since cases of CJD show co-occurrence of type 1 and type 2 PrPres in the brain, the basis of this classification system and its relationship to agent strain are under discussion. Different brain areas from 41 sCJD and 12 iatrogenic CJD (iCJD) cases were investigated, using Western blotting for PrPres and two other biochemical assays reflecting the behaviour of the disease-associated form of the prion protein (PrPSc) under variable PK digestion conditions. In 30% of cases, both type 1 and type 2 PrPres were identified. Despite this, the other two biochemical assays found that PrPSc from an individual patient demonstrated uniform biochemical properties. Moreover, in sCJD, four distinct biochemical PrPSc subgroups were identified that correlated with the current sCJD clinico-pathological classification. In iCJD, four similar biochemical clusters were observed, but these did not correlate to any particular PRNP 129 polymorphism or western blot PrPres pattern. The identification of four different PrPSc biochemical subgroups in sCJD and iCJD, irrespective of the PRNP polymorphism at codon 129 and the PrPres isoform provides an alternative biochemical definition of PrPSc diversity and new insight in the perception of Human TSE agents variability.
| Prion diseases are transmissible neurodegenerative disorders characterized by accumulation of an abnormal isoform (PrPSc) of a host-encoded protein (PrPC) in affected tissues. According to the prion hypothesis, PrPSc alone constitutes the infectious agent. Sporadic Creutzfeldt-Jakob disease (sCJD) is the commonest human prion disease. Although considered as a spontaneous disorder, the clinicopathological phenotype of sCJD is variable and substantially influenced by the methionine/valine polymorphism at codon 129 of the prion protein gene (PRNP). Based on these clinicopathological and genetic criteria, a subclassification of sCJD has been proposed. Here, we used two new biochemical assays that identified four distinct biochemical PrPSc subgroups in a cohort of 41 sCJD cases. These subgroups correlate with the current sCJD subclassification and could therefore represent distinct prion strains. Iatrogenic CJD (iCJD) occurs following presumed accidental human-to-human sCJD transmission. Our biochemical investigations on 12 iCJD cases from different countries found the same four subgroups as in sCJD. However, in contrast to the sCJD cases, no particular correlation between the PRNP codon 129 polymorphism and biochemical PrPSc phenotype could be established in iCJD cases. This study provides an alternative biochemical definition of PrPSc diversity in human prion diseases and new insights into the perception of agent variability.
| Transmissible spongiform encephalopathies (TSE) are neurodegenerative disorders affecting a large spectrum of mammalian species that share similar characteristics, including a long incubation period (which in man may be measured in decades) and a progressive clinical course resulting in death [1].
The most common form of human TSE is an idiopathic disorder named sporadic Creutzfeldt-Jakob disease (sCJD). sCJD is not a uniform disorder in terms of its clinical and neuropathological phenotype. It remains unclear whether this variability is related to variations in the causative TSE agent strains, or to the influence of the methionine/valine polymorphism at codon 129 of the PRNP [2],[3].
A key event in the pathogenesis of TSE is the conversion of the normal cellular prion protein (PrPC, which is encoded by the PRNP gene) into an abnormal disease-associated isoform (PrPSc) in tissues of infected individuals. Conversion of PrPC into PrPSc is a post-translational process involving structural modifications of the protein and resulting in a higher β-sheet content [4]. PrPC is completely degraded after controlled digestion with proteinase K (PK) in the presence of detergents. PrPSc is N-terminally truncated under such conditions, resulting in a PK resistant core, termed PrPres [5]. PrPres, also named PrP 27–30, is a disease marker for TSE and the presence of PrPSc seems to correlate with infectivity [5],[6]. According to the prion hypothesis, PrPSc is the infectious agent in TSE [7] and, in the last decades, several lines of evidence have indicated that particular biochemical properties of PrPSc, such as solubility in N-lauroylsarcosine, PK resistance and electromobility in western blotting (WB) can be used to distinguish between different prion agents or strains [8],[9].
In sCJD, two major PrPres types have been described by WB: in type 1 PrPres, the unglycosylated fragment is 21 kDa, while in type 2, the apparent molecular weight of this unglycosylated fragment is 19 kDa [3]. Protein N-terminal sequencing revealed that type 2 isoform derives from preferential cleavage of the protein during PK digestion at amino acid 97, while in type 1 preferential cleavage occurs at amino acid 82 [10]. sCJD cases can be subclassified according the PrPres isoform and the PRNP codon 129 methionine (M)/valine (V) polymorphism, resulting in 6 major subypes: MM1, MM2, MV1, MV2, VV1 and VV2. Interestingly, these subtypes appear to carry distinct pathological and clinical features, [2],[3], and it has been proposed that type 1 and type 2 isoforms in sCJD might correspond to different TSE agent strains. However, the description of PrPres isoforms which appear to be distinct from type 1 and type 2, and the increasing number of reports describing the coexistence of type 1 and type 2 PrPres in different areas or the same area in the brain from a single sCJD patient, calls into questions the subclassification system described above in sCJD [11]–[14]. Here, in a large group of cases including 41 sCJD and 12 iCJD patients, we confirmed that type 1 and type 2 PrPres can be observed as a mixture in a substantial number of patients. However, using two novel assays described here, PrPSc from these patients with mixed PrPres types are homogeneous irrespective of the brain area considered. Moreover, based on these novel PrPSc biochemical properties, four distinct subgroups were observed in our cohort of sCJD patients. Similar findings were observed in iCJD cases from two countries and differing sources of infection.
A total of 41 French cases of sCJD, each of which had frozen tissue (2–4 g) available from preferentially 5 brain regions: (occipital, temporal and frontal cortex, cerebellum and the caudate nucleus), were included in this study. All six currently defined classes of s-CJD patients (MM1-MM2-MV1-MV2-VV1-VV2) were represented in our panel (Table 1). Moreover, 12 cases of iatrogenic CJD (iCJD), linked to contamination by growth hormone (GH) or dura mater grafts, from patients originating either from United Kingdom (UK) or France, were also investigated (Table 1). None of the patients had a familial history of prion disease and, in each case, the entire PRNP coding sequence was analyzed, either by denaturating gradient gel electrophoresis and/or direct sequencing. All patients died from CJD during the period 1997–2004. Additionally, five cases of Alzheimer's disease were included as non-CJD controls.
In all cases, informed consent for research was obtained and the material used had appropriate ethical approval for use in this project.
For each sample, a 20% brain homogenate (weight/volume) in 5% glucose was prepared using a high-speed homogenizer (TeSeE Precess 48 system). The homogenates were then filtered through a 20 Gauge needle before storage at −80°C.
Various factors have been reported to influence the results of PrPres analysis by WB, including tissue pH and the effect of Cu2+ ions [15]–[17]. In order to limit these factors, each homogenate was diluted a 100-fold in a single non-CJD control brain homogenate prior to further investigation.
A WB kit (TeSeE WB kit Bio-Rad) was used following the manufacturer's recommendations.
Three different monoclonal PrP-specific antibodies were used for PrP detection: Sha31 (1 µg/ml) [18], 8G8 (4 µg/ml) [19] and 12B2 (4 µg/ml) [20], which recognized the amino acid sequences YEDRYYRE (145–152), SQWNKPSK (97–104) and WGQGG (89–93) respectively. After incubation with goat anti-mouse IgG antibody conjugated to horseradish peroxidase, signal was visualized using the ECL western blotting detection system by enhanced chemiluminescent reaction (ECL, Amersham). Molecular weights were determined with a standard protein preparation (MagicMark, Invitrogen).
PrPSc detection was carried out using sandwich ELISA test (TeSeE CJD, Bio-Rad) used following the manufacturer's recommendations. The assay protocol includes a preliminary purification of the PrPSc (TeSeE purification kit) consisting in (i) digestion of PrPC with PK, (ii) precipitation of PrPSc by buffer B and centrifugation, (iii) denaturation of PrPSc in buffer C at 100°C, before immuno-enzymatic detection. In this ELISA, the capture antibody 3B5 recognizes the octarepeat region of PrP [19], while the detection antibody 12F10 binds to the core part of the protein [18].
PK resistance of the PrPSc portion recognized in the ELISA test was determined by measurement of the ELISA specific signal recovered from a series of homogenate aliquots digested with different concentrations of PK in buffer A′ reagent (TeSeE Sheep/Goat purification kit). Each sample was first diluted in normal brain homogenate (between 100- and 10,000-fold) until obtaining a signal between 1.5 and 2 absorbance units after digestion with 50 µg/ml of PK. Triplicate of equilibrated samples were then submitted to a PK digestion with concentrations ranging from 50 to 500 µg/ml, before PrPSc precipitation and ELISA detection. Results were expressed as the percentage of residual signal when compared to the 50 µg/ml PK digestion (lowest PK concentration). In each assay, two standardized controls (scrapie and BSE from sheep) were used as an internal standard. About 20% of samples were randomly selected and submitted to two independent tests separated in time as to assess inter-assay variation.
The ELISA test used in this study was adapted from the Bio-Rad TeSeE test, validated at CEA for EU strain typing studies in ruminants and designed to distinguish BSE in sheep from scrapie. The principle was to measure conformational variations in PrPSc by applying two differential PK digestions under the modification of detergent conditions (SDS sensitivity). For each sample, PK digestion was performed under two conditions: (i) two aliquots of 250 µl of 20% homogenate were mixed either with 250 µl of A reagent (TeSeE purification kit) containing 20 µg of PK, (ii) with 250 µl of A′ reagent (N-lauroylsarcosine sodium salt 5% (W/V), sodium dodecyl sulfate 5% (W/V) containing 55 µg of PK, All the tubes were then mixed by inversion 10 times and incubated at 37°C (in a water bath) for exactly 15 min. Subsequently, 250 µl of reagent B (Bio-Rad purification kit)/PMSF (final concentration 4 mM) were added, mixed and the tubes were centrifuged for 5 minutes at 20,000 g at 20°C. Supernatants were discarded and tubes dried by inversion onto an absorbent paper for 5 min. Each pellet was denatured for 5 min at 100°C with 25 µl of C reagent (Bio-Rad purification kit). The samples were diluted in 250 µL of R6 buffer containing 4 mM of serine protease inhibitor AEBSF (4-2-aminoethyl benzenesulfonyl fluoride hydrochloride), and, if desired, further serially diluted in R6 buffer. ELISA plates were then incubated for two hours at room temperature and, after three washes, antibody detection (TeSeE CJD, Bio-Rad) was added for two hours at 4°C. The ratio of the absorbance obtained in the two conditions (A/A′) was calculated using appropriate dilutions providing absorbance measurements ranging from 0.5 to 2.5 absorbance units in A conditions. For each plate, the same three control samples (one MM1, one VV2 and one MM GH) were included. To avoid inter-assay variations, final results were expressed as a normalized ratio established by dividing the ratio obtained for the analyzed sample by the one obtained for a VV2 sample selected as standard.
For each sCJD case, PrPres profile was determined from five brain areas using both Sha31 and 8G8 antibodies. A single PrPres type (type 1 or type 2) was observed in investigated brain areas of most of the MM1 (n = 11), and all the areas from MV1 (n = 8), VV1 (n = 1) and MM2 (n = 3) sCJD cases of our panel. However, in several cases initially classified as MM1 (n = 2), and in a majority of VV2 (n = 5) or MV2 (n = 6) cases, some brain areas harboured mixed electrophoretic pattern characterized by two distinct bands at 19 and 21 kDa, indicating the coexistence of PrPres type 1 and type 2 (Table 1 and Figure 1A). Moreover, in individual patients some brain areas were found to be type 1, while another area could be found to be type 2 (Table 1 and Figure 1A). Since both Sha31 and 8G8 gave similar results this phenomenon cannot be attributed to some antibody peculiarity in PrP recognition (Figure 1B and 1C).
Antibody 12B2 is specific for the amino acid sequence 89–93 that is located N-terminally of the type 2 cleavage site (amino acid 97). In principle, this antibody is unable to recognize type 2 PrPres. Systematic western blotting with 12B2 consistently demonstrated the presence of the 21 kDa band, characteristic for type 1 PrPres, in nearly all type 2 classified samples, regardless of the PRNP codon 129 polymorphism (Figure 1D). In a limited number of type 2 samples, 12B2 failed to detect a type 1 band (Figure 1E and 1F, lane 2,3).
Using Sha31 or 8G8, mixed type 1/type 2 PrPres profiles were observed in several iCJD cases (Figure 2A), regardless of their national origin or mode of infection. In most (but not all) samples initially classified as type 2, the 12B2 antibody revealed the presence of a 21 kDa band, characteristic of type 1 PrPres (Figure 2B).
Together these findings point to the existence of variable amounts of type 1 PrPres molecules in all or nearly all type 2 classified patients (Table 1).
In sCJD and iCJD patients who harboured a single WB PrPSc type in the different brain areas, as assessed by Sha31, a single ELISA PK resistance profile (Table 1 and Figure 3A) and a comparable ratio in strain typing assay (Table 1 and Figure 3B) were observed in all brain areas. Surprisingly, in each patient harbouring both type 1 and 2 PrPres, either in the same or in different brain areas, a single ELISA PK digestion profile (Table 1 and Figure 3C and 3D) and a comparable signal ratio in strain typing assay (Table 1 and Figure 3B) was also observed, irrespective of region assayed.
MM1 and VV2 samples but also MM2 and VV1 samples, which harboured similar apparent PrPSc content (as assessed by ELISA) were artificially mixed in different proportions. Using WB, a mixed type 1+2 profile could, or could not, be observed depending on the mixture proportions (Figure 3E and 3F). Both PrPSc resistance ELISA assay (Figure 3G and 3H) and strain typing ELISA (not shown) were able to discriminate the different mixtures from the original isolates and from each other. These results clearly demonstrate that the uniformity of PrPSc biochemical properties, as demonstrated by both PrPSc resistance ELISA and strain typing ELISA, in patients harbouring different PrPres isoforms cannot be attributed to a lack of discriminative power of these techniques.
Together, these data strongly indicate that, despite possible variations in PrPres type on WB analysis, patients with either sCJD or iCJD appear to harbour a single PrPSc isoform in their brain.
According to the results from PrPSc PK resistance assay and strain typing ELISAs, sCJD patients could be split into four groups (Table 1, and Figure 3A and 3B). The first group was characterized by a strong PK resistance (Figure 3A) and a low ratio in strain typing assay (Figure 3B). Group 1 could be readily differentiated from Group 2 which showed a higher sensitivity regarding PK digestion, as well as an increased signal ratio in strain typing assay, when compared to Group 1. Two other PrPSc groups were also observed. Group 3 harboured an intermediate PK lability in the PrPSc resistance ELISA and ratio in the strain typing ELISA, when compared to Group 1 and 2. Group 4 had a very high PK-sensitivity and ratio in the strain typing ELISA. No overlapping in PK resistance profile or ratio value in strain typing assay were observed between the four determined groups (Table 1).
Group1 was composed of sCJD MM and MV patients, harbouring predominantly type 1 PrPres while Group 2 consisted in VV and MV patients harbouring predominantly type 2 or type 1+2 PrPres. Groups 3 and 4 were respectively composed with VV1 and MM2 patients from our sCJD panel.
Striking differences were observed in the PrPSc properties between the different iCJD cases and all four groups relying on PrPSc signatures observed in sCJD cases were identified (Table 1, and Figures 3B and 4).
As it might have been expected from sCJD cases observations, Group 1 PrPSc properties was identified in MM1 UK dura mater graft patients (n = 2) (Figure 4A) while Group 2 PrPSc features were observed in UK VV2 (n = 2) (Figure 4D) and MV2 (n = 1) (Figure 4B) GH patients. Surprisingly, a typical Group 2 PrPSc signature was also observed in one out of the three MV1 French GH patients (type 1 in all brain areas). Meanwhile, all investigated MM1 and two out of the three MV1 French GH cases (Figure 4A) harboured identical PrPSc properties than Group 3 sCJD (Figure 4E). Finally, a Group 4 sCJD PrPSc signature (Figure 4F) was observed, using both PrPSc resistance ELISA (Figure 4E) and strain typing ELISA (Figure 3B), in a French dura mater VV1 case (n = 1), which harboured a type 1 PrPres WB profile in every investigated area.
Taken together, these observations support the concept that, in iCJD patients, variability in the PrPSc biochemical properties is not related to the route of infection or the PRNP codon 129 genotype. It also indirectly suggests that the range of different PrPSc properties observed in iCJD might be related to those in the source of infection (likely to have been a sCJD case).
In this study, detection, by WB, of the coexistence of two PrPres types in about 30% (13/41) of cases is consistent with already published data [12],[14]. This observation could suggest the existence in brain from a single patient of different abnormal PrP species. Although two main PK cleavage sites are associated with PrPres type 1 and type 2 (respectively amino acid 82 and 97), N-terminal sequencing revealed in all investigated cases the presence of a whole spectrum of overlapping cleavage sites. Moreover in a part of investigated cases this technique demonstrated the presence (i) of variable but consistent level of type 1 PrPres in patients classified type 2 using WB and (ii) in some patient classified type 1, of low amount of type 2 PrPres [10]. These observations could suggest that, rather than a pure type 1 or type 2 PrPres, PK digestion of a PrPSc specific conformer generate variable mixture of PrPres fragments (with presence of dominant or sub dominant type 1 or type 2 PrPres), which WB usually failed to reveal accurately because its intrinsic technical limits [14]. Antibodies either harbouring higher affinity to PrP (like Sha31) [18] or probing specifically type 1 PrPres (like 12B2) [20], now allow a better perception of such mixture. However, investigations carried out using artificial mixture of type 1 and type 2 brain homogenate, even using high affinity anti-PrP antibodies, clearly indicate the current limits of WB discriminative power [14]. Together, these data suggest that WB analysis of PrPres on its own could be misleading for adequate discrimination between PrPSc variants in CJD.
Both PrPSc PK resistance ELISA and strain typing ELISA are based on the characterization the N terminal part of the PrPSc PK digestion either by increasing PK amount or modifying detergent conditions. While WB profile could be compared to a snapshot picture of PrPres fragments generated by PK digestion process, these assays reflect the dynamics of the PK cleavage rather than its final result (different forms of PrPres). Consequently they could provide different but also more accurate perception of the PrPSc conformers.
Our findings from the PrPSc capture immunoassays clearly indicate that in a single patient, irrespective of brain area, sCJD associated PrPSc displays uniform biochemical properties, regardless of the regional variation of type 1 and type 2 isoforms determined by WB. Such findings support the idea of the presence of a specific TSE agent in each brain and the accumulation of a single associated PrPSc conformer.
Because the limited size of our cohort of cases, an in depth comparison between the PrPSc signature (as established in this study) and the Parchi classification system is not possible.
However, despite this limitation, two major groups were identified in our panel according to the PrPSc properties. The first major group was constituted with patients harbouring a highly PK resistant PrPSc (MM1 and MV1 patients). The second group included patients harboring a PK labile PrPSc (VV2 and MV2 patients). Using both lesion profile and clinical parameters [2], two major forms of sCJD are commonly recognized. The first sCJD form, named “classical”, is characterized by a “rapid evolution” (usually around 4 months), and affects most of the MM1 and MV1 patients. The second sCJD form, named “atypical”, affects VV2 and MV2 with a longer symptomatic evolution (usually longer than 6 months) and a late dementia. Despite inter-individual variations, sCJD Groups 1 and 2, as we defined them on biochemical criteria were consistent with this classification.
Both VV1 and MM2 sCJD cases are extremely rare; they respectively represent 1% and 4% of the identified sCJD cases. According to the literature, these patients have clinical features and lesion profiles that are very different from other sCJD patients [2]. However, in our study as in previously published studies, WB did not identify any distinct biochemical difference from other type 1 and type 2 cases. In contrast, both the strain typing ELISA and PrPSc resistance assays clearly differentiated these cases from Group 1 and Group 2 cases. This finding, which is consistent with clinico/pathological observations carried out in patients, could indicate that there are indeed differences in PrPSc that distinguish these VV1 and MM2 cases from other sCJD groups.
Although prion strains can only be identified definitively by bioassay, molecular in vitro tools to characterize PrPSc are more and more widely used for the rapid identification of particular agents, such as BSE in cattle, sheep, rodent and humans (vCJD) [20],[21]. This has come to be termed “molecular strain typing” and although widely employed, the exact relationship between PrPSc biochemistry and the biological properties of the agents responsible remain to be determined. In sCJD, the presence of four distinct PrPSc biochemical forms apparently correlated to clinico-pathological phenotypes as defined by Parchi et al. [2] could be an indication of the involvement of different TSE agents.
iCJD cases are a consequence of accidental human to human TSE transmission, most likely representing transmission of sCJD. The identification in iCJD cases of the four PrPSc signatures identified in sCJD is consistent with the existence of distinct prions associated with these biochemical forms.
Three examples of human-to-human transmission of variant CJD through blood transfusion have now been identified. While all blood donors were MM at codon 129 PRNP, the recipients had either a MM (n = 2) or a MV genotype (n = 1). Despite this genotype difference there appears to have been conservation of the disease phenotype and PrPres type in all “secondary” vCJD cases [22]–[25]. These observations could suggest that in case of inter-human transmission, difference in donor/recipient genotype could result in un-altered abnormal PrP signature.
Our identification of MM GH iCJD cases harbouring similar PrPSc signature as a VV1 sCJD case or of a VV dura mater iCJD case similar to MM2 sCJD might indicate preservation of a specific PrPSc biochemical signature after human to human transmission between individuals of different codon 129 genotypes.
Treatment with extracts of GH contaminated by CJD has lead to a high number of iCJD cases in France and the UK. The codon 129 genotypes of the affected individuals in the two countries differ, with the French cohort predominantly MM and MV and the British cohort MV and VV [26]. In the absence of any clear explanation for this finding, it was suggested that it might be due to contamination of different batches of GH with different prion strains from individuals of differing PRNP codon 129 genotypes. Our identification of different biochemical forms of PrPSc in GH French patients and in UK patients is consistent with this hypothesis. The variability observed within the French GH cases could signify involvement of different prion strains, consistent with multiple contaminated GH batches in the French epidemic.
The identification in this study of different PrPSc species in CJD patients with the same PRNP polymorphism at codon 129 and WB PrPres profile offers a new perspective on our understanding of the relationship between PrP biochemistry, prion disease phenotype and agent strain. We highlight two novel approaches to analysing PrPSc in sCJD and iCJD and offer evidence that these analyses provide potentially-strain associated information, which appears to be lacking from the conventional WB assay. |
10.1371/journal.pntd.0003877 | Soil-Transmitted Helminths in Southwestern China: A Cross-Sectional Study of Links to Cognitive Ability, Nutrition, and School Performance among Children | Empirical evidence suggests that the prevalence of soil-transmitted helminth (STH) infections in remote and poor rural areas is still high among children, the most vulnerable to infection. There is concern that STH infections may detrimentally affect children’s healthy development, including their cognitive ability, nutritional status, and school performance. Medical studies have not yet identified the exact nature of the impact STH infections have on children. The objective of this study is to examine the relationship between STH infections and developmental outcomes among a primary school-aged population in rural China.
We conducted a large-scale survey in Guizhou province in southwest China in May 2013. A total of 2,179 children aged 9-11 years living in seven nationally-designated poverty counties in rural China served as our study sample. Overall, 42 percent of the sample’s elementary school-aged children were infected with one or more of the three types of STH—Ascaris lumbricoides (ascaris), Trichuris trichuria (whipworm) and the hookworms Ancylostoma duodenale or Necator americanus. After controlling for socioeconomic status, we observed that infection with one or more STHs is associated with worse cognitive ability, worse nutritional status, and worse school performance than no infection. This study also presents evidence that children with Trichuris infection, either infection with Trichuris only or co-infected with Trichuris and Ascaris, experience worse cognitive, nutritional and schooling outcomes than their uninfected peers or children infected with only Ascaris.
We find that STH infection still poses a significant health challenge among children living in poor, rural, ethnic areas of southwest China. Given the important linkages we find between STH infection and a number of important child health and educational outcomes, we believe that our results will contribute positively to the debate surrounding the recent Cochrane report.
| Empirical evidence suggests that the prevalence of soil-transmitted helminth (STH) infections in remote and poor rural areas is still high among children, the most vulnerable to infection. There is concern that STH infections may detrimentally affect children’s healthy development, including their cognitive ability, nutritional status, and school performance. Medical studies have not yet identified the exact nature of the impact STH infections have on children. The objective of this study is to examine the relationship between STH infections and developmental outcomes among a primary school-aged population in rural China. We conducted a large-scale survey in Guizhou province in southwest China in May, 2013. Overall, 42 percent of elementary school-aged children were infected with one or more of the three types of STH—Ascaris lumbricoides (ascaris), Trichuris trichuria (whipworm) and the hookworms Ancylostoma duodenale or Necator americanus. After controlling for socioeconomic status, we observed that children infected with one or more STHs have worse cognitive ability, worse nutritional status, and worse school performance than their uninfected peers. While not causal, this study also presents evidence that children with Trichuris infection, either infection with Trichuris only or co-infected with Trichuris and Ascaris, have worse cognitive, nutritional and schooling outcomes than their uninfected peers or children infected with only Ascaris. Given these important linkages, we hope that our results will contribute positively to the debate surrounding the recent Cochrane report.
| A recent Cochrane Report [1] has raised questions about the nature of the relationship between infection with soil-transmitted helminths (STHs) and children’s healthy development. In a meta-analysis of 42 papers, the authors of the report found that there was no clear, consistent relationship between deworming and improvements in children’s cognitive ability, nutritional indicators, or school performance. The report ended with a call for more concerted research that would help clarify the nature of the relationship between STH infection and these outcomes in children. Health policymakers depend on this type of information when deciding how to allocate resources to different disease types, in general, and how to allocate for STH control and treatment, in particular.
Because of high STH prevalence, China is an especially suitable setting in which to conduct such additional research. According to Wang et al. [2], 40 percent of school-aged children in rural areas of Guizhou province are infected with one or more types of three STHs: Ascaris lumbricoides (ascaris), Ancylostoma duodenale (hookworm), and Trichuris trichuria (whipworm). In some villages, the prevalence is as high as 80 percent. Similarly high prevalence has also been reported elsewhere in China, such as in rural Guangxi and Hainan provinces [3].
The uncertainty in the international literature about the link between STH infections and child outcomes is reflected in the China-specific literature. Of six total China-based studies, five measured the link between STH infection and child health (either physical development or hemoglobin levels). Three of these [2–4] found a significant negative correlation between STH infection and children’s health, while two found no correlation [5–6]. None of the six studies measured school performance, although three attempted to measure the relationship between STH infections and cognitive ability using formal tests; however, the sample sizes in these studies were small, ranging from 140 to 200 children in 2 to 11 communities or clusters. With such small sample sizes, it is statistically improbable to produce meaningful results. In short, in the context of China—as in the rest of the world—there is uncertainty about the relationship between STH infections and child outcomes.
In this paper we will answer questions raised by the Cochrane report [1] and build evidence on the relationship between STH infections and health outcomes in children in rural China. To achieve this goal, we have three objectives. First, we will document the prevalence of STHs in the study areas, thus better defining the severity of the STH problem in poor areas of rural China. Second, we will document the levels of cognitive ability, nutritional indicators and school performance among our sample children in order to assess how children in poor rural areas fare in terms of these measured outcomes. Finally, we will examine the links between STH infection and cognitive ability, nutritional indicators, and school performance.
We collected the data used in May of 2013 as part of a large-scale survey of elementary school-aged children in Guizhou province. Our study was conducted in seven rural counties in Qiandongnan prefecture. We chose our sample to include regions that were poor and populated by ethnic minorities, the subpopulations that are at higher risk for STH infection. Fig 1 depicts the sample selection process. Based on rural per capita income levels reported in Guizhou Statistical Yearbook [7], the research team randomly selected a total of seven rural counties from the poorest half of the counties (8 out of 16) in Qiangdongnan. According to national statistics, at 4,625 yuan, the average rural individual in our sample areas has a per capita income in the bottom quartile of China’s rural income distribution [8].
Once we chose the sample counties, we selected the sample townships and villages. In each county, we included all townships except for the township which houses the county government. We did not include the township that housed the county government because such townships are almost always wealthier and more urban than a typical rural township. A total of 112 townships were selected. Then sample villages within each township were selected. Since our survey would take place in schools, we obtained a list of all the 9–11 year old children attending the central primary school in each township. We classified all 9–11 year old children by their home village, and then randomly selected 20 sample children from the home village with the largest number of children at that school. We excluded villages that housed the local township government, since (as discussed above in the context of towns/counties) these villages are typically wealthier and more urban than a typical village. If the home village had fewer than 20 children in our age group attending the school, we randomly selected children from the next-largest village to fill in the gap. Overall, a total of 20 school children were randomly chosen from either one or two villages in each township. A total of 2,240 children from 146 villages and 112 townships in the 7 poor rural counties were chosen as sample students.
The survey team collected four types of information: data from a socioeconomic survey; scores on a test of cognitive ability; measures of child health (including STH infection status, height, weight, and hemoglobin levels); and school absence and performance (as measured by absenteeism and performance on a standardized math test).
This study received ethical approval from the Stanford University Institutional Review Board (IRB) (Protocol ID 25027), and from the Sichuan University Ethical Review Board (Protocol ID 2013005–02). All participating children gave their assent for their involvement in the study, and the children’s legal guardians gave their written consent for both their own and their children’s involvement. Children who were found to have severe anemia were referred to the local hospital for treatment.
Anemia status was determined based on finger prick blood analysis for hemoglobin (Hb). Following internationally accepted standards, anemia was defined as Hb<115 g/L [20].
Physical indicators of height and weight were used to construct height-for-age z-scores (HAZ) and Body Mass Index (BMI)-for-age z-scores using WHO AnthroPlus, a software application of the WHO Reference 2007 for children aged 5–19 years that is used to monitor the growth of school-aged children and adolescents [21]. Weight-for-age z-scores (WAZ) were calculated using a SAS program for the 2000 CDC growth chart for children aged 0–20 years [22]. We followed internationally recognized cutoffs [23] to consider children whose HAZ, WAZ, or BMI-for-age z-score to fall more than two standard deviations below the international mean to be stunted, underweight, or malnourished, respectively.
Raw scores obtained from the four core subtests of the WISC-IV were converted to age-scaled index scores using tables of norms in the Mandarin version of the WISC-IV administration and scoring manual. Two index scores are considered for analysis: Working Memory Index (WMI) and Processing Speed Index (PSI). Scores are divided into internationally-recognized ranges. A score of 90–110 is considered “average”; a score of 80–89 is considered “low average”; a score of 70–79 is considered “borderline”; and a score of below 70 is considered “extremely low” and at risk for intellectual disabilities or mental retardation.
All statistical analyses were performed using STATA 12.0. P-values below 0.05 were considered statistically significant. All P-values were adjusted for multiple hypotheses testing by the Bonferroni method. The statistical significance of differences in all outcomes by subgroup populations was assessed using student’s t-test in STATA. STATA’s multiple linear regression model was used in the multivariate analysis for those continuous outcome variables: WMI, PSI, Hb, HAZ, WAZ, BmiAZ, as well as standardized math test scores. Meanwhile, STATA’s logistic regression model was used in the multivariate analysis for those binary outcome variables: anemic (yes/no) and school absence (yes/no). We included the following independent variables as potential confounders in the multivariate analysis: gender, age, boarding status, minority status, sanitation behaviors, and household characteristics. Definitions of key variables to be used in the rest of the paper are presented in Table B in S1 Text.
We examined 2,179 school-aged children. In our sample, 46 percent of the students were female and 54 percent were male, a ratio similar to those found in most poor areas in China [24]. The average age is 10.6 years. A total of 26 percent of sample students board at school.
The background characteristics of the sample by infection status are presented in Table 1. There are no significant differences between the infected and uninfected group in terms of deworming history or gender. However, infected children are more likely to be older, to board at school, and to be a member of the Miao, or Shui ethnic minority groups. Infected children are also significantly more likely than uninfected children to eat uncooked meat, drink unboiled water, and to live in households with a dirt floor. Uninfected children are significantly more likely than infected children to wash their hands after using the toilet, to live in households with a private latrine, to have a higher household income, and to have better educated parents.
Of the 2,179 children who provided fecal samples, participated in the socioeconomic survey, cognitive testing, health examination as well as school absence and performance tests, 42 percent were infected with one or more of the three types of STH (Table 2). The most prevalent type of STH in the survey areas is Ascaris (31 percent), followed by Trichuris (22 percent), and finally by hookworm (1 percent). The prevalence of Ascaris only and Trichuris only was 19 percent and 11 percent, respectively. Eleven percent of sample children were co-infected with Ascaris and Trichuris. In contrast, the prevalences of infection with hookworm only, co-infection with Ascaris and hookworm, co-infection with Trichuris and hookworm, or co-infection with Ascaris, Trichuris and hookworm were negligible.
A total of 63 percent of children had a Working Memory Index (WMI) that was either “extremely low” (<70) or “borderline” (70–79). The breakdown shows that 13 percent of children scored “extremely low” on the WMI portion of the test, and 50 percent of children scored in the “borderline” range.
A total of 36 percent of children had a Processing Speed Index (PSI) that was either “extremely low” (<70) or “borderline” (70–79). The breakdown shows that 10 percent of children scored “extremely low” on the PSI portion of the test, and 26 percent of children scored in the “borderline” range.
We find that 16 percent of our sample children are anemic (Table 2), 28 percent are stunted (HAZ < -2), 6 percent are malnourished (BMI-for-age < -2), and 26 percent are underweight (WAZ < -2).
Around 13 percent of children in our sample had been absent from school at least once during the most recent semester. The average child in our sample earned a failing score on the TIMSS test (score < 60), scoring an average of 52.6 out of 100 on the TIMSS test (Table 2).
There are significant differences in children’s cognitive ability between infected children and uninfected children (Fig 2 and Table 2). Our data show that the mean WMI of infected children is 76.6, significantly lower than that of the uninfected group (80.0, p < 0.005). Moreover, 71 percent of infected children had an “extremely low” or “borderline” WMI, significantly higher than that of uninfected children (57 percent, p < 0.005). These differences remain statistically significant after controlling for confounding factors (Table 3).
Our data show that the mean PSI of infected children is 83.4, significantly lower than that of the uninfected group (88.1, p < 0.005). Moreover, 45 percent of infected children in the infected group had an “extremely low” or “borderline” PSI, significantly higher than that of uninfected children (29 percent, p < 0.005). These differences remain statistically significant after controlling for confounding factors (Table 3).
There are no significant differences between children with Ascaris only and children with no infection of any of the three types of STHs in terms of WMI and PSI (See Panel A, Table C in S1 Text). The same is true when we compare children with both Ascaris and Trichuris against children with Trichuris only (Panel D2).
However, compared with children with no infection of any of the three types of STHs, children with Trichuris only infection had significantly lower WMI (p<0.001) and PSI (p<0.001) after controlling for confounding factors (Panel B). The same strong relationship remains when we compare children with both Ascaris and Trichuris against children with no infection with any of the three types of STHs (Panel C).
Our data also show that children with a co-infection of both Ascaris and Trichuris had significantly lower PSI (p<0.001) than children with Ascaris only infection, after controlling for confounding factors. However, there is no significant difference in WMI between these two groups of children (Panel D1).
There are significant differences in terms of both mean hemoglobin level and mean anemia rate between infected and uninfected children (Table 2). The hemoglobin level among infected children is 126.8 g/L, significantly higher than that among the uninfected group (125.4 g/L, p < 0.01). Meanwhile, the anemia rate among infected children is 19 percent, significantly higher than that among the uninfected group (15 percent, p < 0.05). However, after controlling for confounding factors, the differences between infected children and their uninfected peers were found to be statistically insignificant (Table 3).
There are significant group differences in height-for-age z-scores (HAZ) between infected and uninfected children (Table 2). The mean HAZ among infected children is -1.59, compared with -1.25 among uninfected children (p < 0.005). An average of 34 percent of infected children are stunted, compared with 23 percent of uninfected children (p < 0.005). After controlling for confounding factors, the difference in HAZ remains statistically significant (Table 3, p < 0.001).
There are significant group differences in weight-for-age z-scores (WAZ) between infected and uninfected children (Table 2). The mean WAZ among infected children is -1.55, compared with -1.27 among uninfected children (p < 0.005). An average of 31 percent of infected children are underweight, compared with 23 percent of uninfected children (p < 0.005). After controlling for confounding factors, the difference in WAZ remains statistically significant (Table 3, p < 0.005).
The mean BMI-for-age z-score among infected children is -0.64, compared with -0.55 among uninfected children (p < 0.05). The proportion of being malnourished does not vary significantly between infected and uninfected children (Table 2). Similar to the cases of hemoglobin level and anemia rate, after controlling for confounding factors, the difference in BMI-for-age z-scores becomes statistically insignificant (Table 3).
There are no significant differences between children with Ascaris only and children with no infection of any of the three types of STHs in terms of mean hemoglobin level, mean anemia rate, HAZ, WAZ and BmiAZ (Panle A, Appendix Table 3). The same is true when we compare children with both Ascaris and Trichuris against children with Trichuris only (Panel D2).
However, compared with children with no infection of any of the three types of STHs, children with Trichuris only infection had significantly lower Hb (p<0.001), lower HAZ (p<0.001) and lower WAZ (p<0.001) after controlling for confounding factors although their mean anemia rate and BmiAZ are not statistically different (Panel B). Our data also show that children with a co-infection of both Ascaris and Trichuris had significantly lower HAZ (p<0.001) and lower WAZ (p<0.001) than children with no infection of any of the three types of STHs although their mean hemoglobin level, mean anemia rate and BmiAZ are not statistically different (Panel B). Finally, compared with children with Ascaris only infection, children co-infected with both Ascaris and Trichuris also had significantly lower WAZ (p<0.001), but there are no significant differences in terms of mean hemoglobin level, mean anemia rate, HAZ or BmiAZ (Panel D1).
The absence rate among infected children is 16 percent, compared to 11 percent among uninfected children (p < 0.005) (Table 2). This difference becomes statistically insignificant after controlling for confounding factors (Table 3).
The mean TIMSS score among infected children is 47.5, compared with 56.4 among uninfected children (p < 0.005). A total of 70 percent of infected children failed the TIMSS test (scores < 60), compared with 56 percent of uninfected children (p < 0.005) (Table 2). These differences remain statistically significant after controlling for confounding factors (Table 3, p < 0.001).
Compared to children without infection of any of the three types of STHs, children with Ascaris only infection see no difference in the incidence of school absence (See Panel A, Table C in S1 Text). The same is true when we compare children with Trichuris only (Panel B) or children with a co-infection with both Ascaris and Trichuris against children with no infection of any of the three types of STHs (Panel C). Similarly, our data show no significant difference in incidence of school absence between children with a co-infection with both Ascaris and Trichuris with children with Ascaris only (Panel D1) or Trichuris only (Panel D2).
Our data also show that there is no difference in standardized test scores between children with Ascaris only infection and children with no infection of any of the three types of STHs, after controlling for confounding factors (See Panel A, Table C in S1 Text). The same is true when we compare children with a co-infection of both Ascaris and Trichuris to children with Trichuris only infection (Panel D2).
However, compared with children with no infection of any of the three types of STHs, children with Trichuris only infection or children had lower standardized test scores (p<0.001), after controlling for confounding factors (Panel B). The same strong correlation holds when we compare children with a co-infection of both Ascaris and Trichuris against children with no infection of any of the three types of STHs (Panel C). Our data also show that children with a co-infection of both Ascaris and Trichuris had lower standardized test scores (p<0.001) than children with Ascaris only infection (Panel D1).
In this paper we document the prevalence of STHs using results from stool sampling and socioeconomic testing of 2,179 school children living in seven nationally-designated poverty counties in Qiandongnan prefecture in Guizhou province. We observed that 42 percent of the sample children were infected with one or more of the three types of STH—Ascaris, Trichuris, and hookworm. This prevalence is consistent with previous, smaller-scale studies in China [2,3], but is more than twice the observed STH prevalence from the National Survey on Current Status of the Important Parasitic Diseases in Human Population in 2004 [25]. According to the WHO treatment guidelines, the prevalence we document warrants mass treatment.
We also document children’s cognitive ability, nutritional indicators, school absence and performance. Our data show that sample children are lagging far behind the international standard in terms of each of these measured outcomes. We further found that after controlling for a set of socioeconomic confounders, infection with one or more STHs is associated with worse cognitive ability (in terms of WMI and PSI), worse nutritional status (in terms of HAZ and WAZ), and worse school performance (in terms of standardized math test scores). Without implying there is any causal link per se, this study also presents evidence that infection with Trichuris, either infection with Trichuris only or co-infected with Trichuris and Ascaris, makes children experience worse cognitive, nutritional and schooling outcomes than their uninfected peers or children infected with only Ascaris.
These results are consistent with findings from other epidemiological studies and randomized controlled trials examining the relationship between STH infections and cognitive ability [3, 10–14, 26], nutritional indicators [2, 3, 13, 27–32], and school performance [33–34]. In a study of children between the ages of 5 and 14 years in China, Shang reported that STH infection was associated with high incidence of malnourishment, stunting and anemia. The Shang study also found infections were correlated with worse performance on WMI and PSI [3]. These observations support our findings that infections with one or more STHs are associated with worse cognitive ability, worse nutritional status and worse school performance than their uninfected peers. A randomized controlled trial of STH infected children who were treated with albendazole demonstrated less school absenteeism than children in the comparison group who were not treated with albendazole [35]. However, we do not find any strong correlation between STH infection with school absenteeism.
A previous cross-sectional study among schoolchildren in Brazil suggested that polyparasitised children experience worse cognitive outcomes than children with only one helminth infection [14]. We attempted to assess the effects of polyparasitism on child development outcomes, specifically cognitive ability, nutritional indicators, school absence and performance. To do so, a method was developed whereby development outcomes were compared between children with co-infections of STHs and children with only one STH infection. The findings suggest that infection with Trichuris, either infection with Trichuris only or co-infected with Trichuris and Ascaris, makes children experience worse cognitive, nutritional and schooling outcomes than their uninfected peers or children infected with only Ascaris.
Our study has several limitations. First, due to budgetary constraints, we collected two stool samples per child (on consecutive days), rather than three samples per child. While we believe that two samples adequately allows for the cyclical nature of roundworms’ egg laying patterns, three samples may have allowed for even greater sensitivity in the detection of the true prevalence of infection. Second, while we made every effort to keep samples refrigerated for as long as possible between sample production and laboratory testing, the samples were not produced on site, and therefore children may have waited up to several hours before delivering their samples to the nearest refrigeration facilities (either at the village clinic or at the school). This waiting period was outside of our control, but may have contributed to the degradation of hookworm eggs. Since both of these limitations may have resulted in an underestimate of total STH prevalence, the estimates presented here should be considered to be a lower bound. A third study limitation is that we were unable to collect data on the intensity of infection in the sample areas.
Our study shows that STH infection still poses a significant health challenge among children living in poor, rural, ethnic areas of southwest China. Given the important linkages we find between STH infection and a number of important child health and educational outcomes, we hope that our results will contribute positively to the debate surrounding the recent Cochrane report [1]. Although our results are correlational, we believe that the strength of the correlations is striking, and indicates a need for more rigorous research on the impacts of STH treatment on child outcomes.
Due to the cross-sectional nature of our data, we are unable to identify the precise reasons behind the linkages we observe through this study. However, we can speculate that one possible explanation might be that STH infections lead to nutritional deficits, which in turn contribute to cognitive impairments. Another possible explanation might be that children from disadvantaged households are both more likely to have poor sanitation practices that contribute to chronic STH infection and are also more likely to have poor nutritional intake that may lead to worse cognitive performance.
|
10.1371/journal.pbio.1002472 | NMNAT2:HSP90 Complex Mediates Proteostasis in Proteinopathies | Nicotinamide mononucleotide adenylyl transferase 2 (NMNAT2) is neuroprotective in numerous preclinical models of neurodegeneration. Here, we show that brain nmnat2 mRNA levels correlate positively with global cognitive function and negatively with AD pathology. In AD brains, NMNAT2 mRNA and protein levels are reduced. NMNAT2 shifts its solubility and colocalizes with aggregated Tau in AD brains, similar to chaperones, which aid in the clearance or refolding of misfolded proteins. Investigating the mechanism of this observation, we discover a novel chaperone function of NMNAT2, independent from its enzymatic activity. NMNAT2 complexes with heat shock protein 90 (HSP90) to refold aggregated protein substrates. NMNAT2’s refoldase activity requires a unique C-terminal ATP site, activated in the presence of HSP90. Furthermore, deleting NMNAT2 function increases the vulnerability of cortical neurons to proteotoxic stress and excitotoxicity. Interestingly, NMNAT2 acts as a chaperone to reduce proteotoxic stress, while its enzymatic activity protects neurons from excitotoxicity. Taken together, our data indicate that NMNAT2 exerts its chaperone or enzymatic function in a context-dependent manner to maintain neuronal health.
| Pathological protein aggregates are found in many neurodegenerative diseases, and it has been hypothesized that these protein aggregates are toxic and cause neuronal death. Little is known about how neurons protect against pathological protein aggregates to maintain their health. Nicotinamide mononucleotide adenylyltransferase 2 (NMNAT2) is a newly identified neuronal maintenance factor. We found that in humans, levels of NMNAT2 transcript are positively correlated with cognitive function and are negatively correlated with pathological features of neurodegenerative disease like plaques and tangles. In this study, we demonstrate that NMNAT2 can act as a chaperone to reduce protein aggregates, and this function is independent from its known function in the enzymatic synthesis of nicotinamide adenine dinucleotide (NAD). We find that NMNAT2 interacts with heat shock protein 90 (HSP90) to refold protein aggregates, and that deleting NMNAT2 in cortical neurons renders them vulnerable to protein stress or excitotoxicity. Interestingly, the chaperone function of NMNAT2 protects neurons from protein toxicity, while its enzymatic function is required to defend against excitotoxicity. Our work here suggests that NMNAT2 uses either its chaperone or enzymatic function to combat neuronal insults in a context-dependent manner. In Alzheimer disease brains, NMNAT2 levels are less than 50% of control levels, and we propose that enhancing NMNAT2 function may provide an effective therapeutic intervention to reserve cognitive function.
| Robust neuronal maintenance mechanisms are required to minimize or repair damage arising from intrinsic and extrinsic stressors. Nicotinamide mononucleotide adenylyl transferases (NMNATs) play important roles in neuronal maintenance in flies [1,2] and human [3–6], and their overexpression provides neuroprotection in diverse neurodegenerative models [7,8]. NMNATs are known as nicotinamide adenine dinucleotide (NAD)-synthesizing enzymes that enable proper flux of NAD, an essential cofactor for many cellular processes [7,8]. The literature regarding the importance of this enzymatic activity in axonal health is mixed [8]. The majority of evidence that supports the importance of NMNAT’s enzymatic activity comes from studies using Wallerian degeneration models. Upon nerve cut, NAD levels are quickly reduced, and exogenous NAD application offers protection [9–12]. Abolishing NMNAT enzymatic function impairs their axon-protective function against injuries [9,10,12–14]. However, NMNAT enzymatic activity is not required to maintain neural integrity in Drosophila photoreceptors [2], suggesting that NMNATs may protect neurons by different mechanisms in a context-dependent manner [7,8].
NMNAT2 has the shortest half-life among three mammalian NMNATs and its function in axonal survival cannot be compensated by endogenous NMNAT1 or 3 [15,16]. The constant axonal supply of NMNAT2 from Golgi-derived vesicles is critical to maintain axonal health [16,17]. Constitutive NMNAT2 removal results in neurite outgrowth deficits [18,19], while knocking down NMNAT2 in developed neurons causes axonal degeneration [15], a prominent feature of many neurodegenerative diseases [20–23]. Importantly, NMNAT2 is highly expressed in the mammalian brain [24,25]. nmnat2 mRNA levels are reduced in Parkinson, Huntington, and Alzheimer diseases (AD), as well as in tauopathies [26–31], suggesting a role of NMNAT2 in maintaining neuronal health in the aging brain. In rTg4510 transgenic mice, a Frontotemporal Dementia and Parkinsonism-17 (FTDP-17) tauopathy model, NMNAT2 abundance declines prior to the onset of neurodegeneration or memory deficits [32]. Moreover, elevating NMNAT2 levels in rTg4510 mice ameliorates their neurodegenerative phenotype. These finding suggest a role for NMNAT2 in neuronal maintenance in the brain, but the mechanism underlying neuroprotection remains to be elucidated.
Drosophila NMNAT (dNMNAT) also functions as a molecular chaperone [2]. Molecular chaperones are defined as a class of proteins that interact with, stabilize, or assist proteins to retain their native and functionally active conformation [33]. They are critical in maintaining protein homeostasis (proteostasis) and to facilitate the clearance of pathological protein aggregates [34,35]. Furthermore, the identification of synaptic chaperones has highlighted the importance of chaperones in preserving neuronal function [36–40]. For example, the synaptic vesicle protein cysteine string protein α (CSPα) forms a chaperone complex with heat shock cognate 70 (Hsc70) and the small glutamine-rich tetratricopeptide repeat protein SGT to promote SNARE-complex assembly and maintain the presynaptic release machinery [37,38]. Haploinsufficency of CSPα in humans causes autosomal dominant adult onset neuronal ceroid lipofusinosis and leads to dementia [41,42].
Proteinopathies, including AD and tauopathies, are characterized by stereotypic aggregated proteins, and this pathology is associated with cognitive impairment [43,44]. It has been proposed that enhancing chaperone activity helps to establish a cytoprotective state, defending against the cellular damage caused by misfolding and/or aggregation associated proteinopathies [45]. The role of chaperones in AD has been studied extensively with respect to tau aggregation and fibrillization. Chaperones, such as HSP70, HSP90, and the C-terminus of HSC70 Interacting Protein (CHIP) have been shown to preferentially bind to hyperphosphorylated human Tau (p-hTau) as well as paired helical filamentous tau, but not to nonphosphorylated tau [46]. Indeed, overexpression of cytosolic HSP70 and HSP90 inhibit the early stages of amyloid aggregation in AD models [47]. The reduced Tau burden upon elevating NMNAT2 levels in rTg4510 mice [32] prompted us to ask the following three questions. Does NMNAT2 act as a chaperone? Is this chaperone activity required for NMNAT2 to reduce tauopathy and protect neurons against protein stress? How does NMNAT2 exert its chaperone function?
In this study, we validate the clinical relevance of human NMNAT2 levels by examining the relationship between nmnat2 mRNA levels and the cognitive capabilities and AD pathology in a large cohort of aged human subjects. In AD brains, NMNAT2 shifts into the insoluble fraction together with chaperones like HSP70 [48–51]. Next, we show that NMNAT2 functions as a chaperone in several in vitro and in vivo assays. We define the domain that acts as a chaperone and show that NMNAT2 acts with HSP90 to clear protein aggregates. Furthermore, we demonstrate that NMNAT2’s enzymatic or chaperone function is differentially recruited to protect neurons depending on the nature of insult. Deleting NMNAT2 in cortical neurons increases their vulnerability to proteotoxic stress and excitotoxicity triggered by excessive neurotransmission. The chaperone function of NMNAT2 is required to defend against proteotoxic stress, while its enzymatic function is indispensible for maintaining viability under excitotoxic conditions. Taken together, our findings suggest that the dual-functions of NMNAT2 protect neurons and preserve cognitive function during aging.
To evaluate the clinical relevance of NMNAT2 in human brain, we examined postmortem brain tissue and data available from the Religious Orders Study and the Rush Memory and Aging Project [52–54]. These studies combine longitudinal clinical and cognitive evaluations with brain neuropathological evaluations at death. A global cognitive summary measure is based on subject performance on a battery of 17 standardized cognitive tests, while a global AD neuropathologic burden is computed from counts of neuritic plaques and neurofibrillary tangles on silver-stained tissue sections from five brain regions [55,56]. nmnat1/2 mRNA levels were extracted from an RNA-sequencing dataset profiling the transcriptome in dorsolateral prefrontal cortex samples from 541 deceased subjects (mean age at death = 88.4 y, standard deviation (sd) = 6.7; S1 Table). Higher nmnat2 mRNA levels are associated with better cognitive performance proximate to death (p = 0.0007; Fig 1A–1D) as well as with a lower AD neuropathological burden (p = 0.004; S2 Table). These associations are robust when adjusted for age of death, postmortem interval and RNA quality (RNA integrity number; RIN) (S1 Table). By contrast, there is no relationship between nmnat1 mRNA levels with cognition or AD pathology in this cohort (Fig 1B and 1C, S2 Table).
Causality cannot be inferred from such human data analysis. Path analysis, developed by the geneticist Sewall Wright [57], is a standard approach in human genetics to examine the complex interrelationships among multiple variables and to assess the relative importance of these variables upon outcome(s) of interest, including a range of potential mediators [57–60]. We conducted path analysis to examine the interrelationship between nmnat2 expression, AD pathology and cognition. Fig 1D illustrates our hypothetical structural model, including a direct path that links nmnat2 expression to cognition, as well as an indirect path that links its expression to AD pathology and, subsequently, to cognition. Standardized path coefficients and the corresponding standard error for each path were reported to assess both the hypothesized direct and indirect effect of nmnat2 expression on cognition, using standard notation [60]. The relations represented by arrows in Fig 1D reflect our a priori hypothesis and are consistent with, but are not proof of, causality. We find that about 30% of the effect of nmnat2 mRNA abundance on cognition is explained by global AD pathology (p = 0.003). This result is consistent with a hypothetical model in which at least part of the relation between nmnat2 and cognition is mediated by AD pathology. The placement of AD pathology downstream of nmnat2 expression in this model suggests the role of NMNAT2 in modulating AD pathology.
An independent cohort of human cortical samples consisting of control and AD brains were acquired from the Oregon Brain Bank. In this cohort, we observed a significant reduction of nmnat2 mRNA as well as NMNAT2 protein but not NMNAT1 (S1 Fig; S3 Table). Soluble and insoluble proteins were extracted sequentially from human postmortem brain tissues with progressively more stringent detergents thereby fractionating human brain proteins based on their solubility [61,62], allowing the more soluble proteins to be extracted first. This was followed by the extraction of membrane-bound proteins, and finally the most insoluble proteins were isolated in a 2% SDS buffer. Aggregated proteins such as neurofibrillary tangles and amyloid plaques are typically present in insoluble fractions prepared from AD brains [63]. The majority of NMNAT2 in control brains was extracted in the soluble fraction (Fig 1E and 1F), while very little NMNAT2 was found in the insoluble fraction. However, in brains of AD patients, abundant NMNAT2 protein was detected in the insoluble fraction (p < 0.001). This fraction also contains hyperphosphorylated hTau and HSP90 (Fig 1E). The strong shift in solubility of NMNAT2 in AD brains is reminiscent of the behavior of chaperones, such as HSP70, HSP90 and HSP27, and cochaperones such as CHIP, which have been linked to pathological aggregates in AD [64].
To assess the chaperone activity of human NMNAT2, we first conducted a cell-based luciferase assay [65,66]. In this assay, treatment with cycloheximide (200 uM for 3 h) to block protein synthesis, followed by heat denaturation (at 42°C for 15 min), renders the endogenous chaperone machinery incapable of preventing luciferase aggregation and refolding (Fig 2A). This allows for a direct measurement of the chaperone activity of the introduced test protein by its ability to prevent luciferase from undergoing heat shock-induced denaturation (holdase activity) and to promote proper refolding (foldase activity) of heat-denatured luciferase during recovery (at 37°C for 3 h). We found that NMNAT2’s holdase and foldase activity are comparable to that of HSP70 (S2 Fig). In other words, NMNAT2 reduces luciferase aggregation upon thermal stress and promotes its refolding, as indicated by higher luciferase activity postrecovery in NMNAT2 overexpressing cells.
To assess if NMNAT2 can directly bind to substrates and reduce protein aggregation in the absence of other proteins, a cell-free chaperone activity assay was performed. In the presence of thermal stress, citrate synthase (CS) aggregates leading to increased optical absorbance that can be detected as a result of Raleigh scattering [67]. This assay with recombinant proteins allows us to examine whether NMNAT2 can exert holdase activity to maintain CS in a thermostable conformation in the absence of other chaperones, cochaperones, or ATP. Both NMNAT2 and HSP70, but not lysozyme, are capable of preventing thermally-induced CS aggregation in a dose-dependent manner (S3 Fig).
Homology-based structural modeling of NMNAT2 shows that both the N- and C-terminal domains are highly conserved among NMNATs [24], while the central domain of NMNAT2 is different [68]. Both NAD synthesis activity [68,69] and chaperone function are comparable among NMNAT1-3 (S2 and S3 Figs). Previous studies have shown that mutating W92G eliminates more than 95% of NMNAT2’s catalytic activity [14,70] whereas mutating two palmitoylation sites (Cys-164 and Cys-165) in the central region of the protein increases stability [16,17]. We therefore generated an enzyme-dead (ED) NMNAT2 by mutating residue W92 (NMNAT2W92G; ED) or by deleting residues 1 to 100 of the N-terminus (Δ1–100; ΔNT) to entirely eliminate NMNAT2’s NAD catalytic activity (Fig 2B, S4A and S4B Fig). As shown in Fig 2C, both NMNAT2-ΔNT (p = 0.0004) and -ED (p = 0.0098) retained the ability to decrease luciferase aggregation and increase refolding similarly to NMNAT2-WT and HSP70.
Mutating the palmitoylation sites at Cys-164 /Cys-165 to serines (C164S;C165S; PM) significantly improved enzymatic (p < 0.001) and chaperone activity (p < 0.0001) for either holdase or foldase activity (Fig 2C) in addition to its stability (S4C Fig). However, deleting the C-terminal domain (Δ200–308; ΔCT) severely impaired NMNAT2’s foldase (p < 0.0001) but not its holdase activity (p = 0.87, compared to WT; Fig 2C), suggesting that this domain is selectively required for foldase activity. ATP binding has been shown to enhance chaperone activity of the large heat shock protein families by suppressing protein unfolding and aggregation [35,71–74]. The C-terminal region of NMNAT is predicted to contain an ATP-binding site spanning residues 269–274. These residues are highly conserved among NMNATs. Deleting 5 aa in the C-terminal ATP site (Δ269–274; ΔcATP) renders NMNAT2 incapable of promoting luciferase refolding (p < 0.0001, Fig 2C) but does not affect NAD synthesis (p = 0.79, S4A and S4B Fig). The deficits in foldase activity observed in NMNAT2-ΔCT and -ΔcATP are unlikely to be caused by alterations in their stability, as the half-lives of these proteins are similar to NMNAT2-WT, -ΔNT, and -ED (S4C Fig). Hence, our data indicate that the C-terminal ATP site is critical for NMNAT2’s foldase function, and the domains implicated in NAD enzymatic and chaperone function are mutually exclusive.
Intracellular aggregation of the microtubule-associated protein tau into filamentous inclusions is a defining characteristic of AD. Because the appearance of tau-aggregate bearing lesions correlates with both cognitive decline and neurodegeneration, it has been hypothesized that hyperphosphorylated tau may be directly toxic to the cells that harbor them. Our previous studies showed that NMNAT2 overexpression in the hippocampi of rTg4510 mice, a tauopathy model [75,76], is sufficient to reduce p-hTau burden [32]. To determine whether NMNAT2’s chaperone function is required for p-hTau clearance, we first tested NMNAT2 WT and various mutants in a HEK293-tau cell line stably expressing a doxycycline-inducible human tau40 (Fig 3A and 3B) [77]. In these cells, an increase in p-hTau is induced 48–72 h after doxycycline treatment. This p-hTau is easily detected by the PHF-1 antibody that detects p-S396/404 Tau [77,78]. Overexpression of NMNAT2-WT (p = 0.009), -ED (p = 0.013), -PM (p = 0.011), or HSP70 (p = 0.024) significantly reduced p-hTau by half, but not total, hTau in the doxycycline-induced cells (Fig 3A and 3B). However, NMNAT2-ΔCT (p = 0.682) or -ΔcATP (p = 0.763) overexpression failed to reduce p-hTau levels, suggesting that NMNAT2’s chaperone function is required to reduce p-hTau in this in vitro Tau model.
To further demonstrate that NMNAT2’s chaperone activity is required for p-hTau clearance in vivo, we examined the impact of NMNAT2-WT, -ED, -ΔcATP and GFP overexpression on p-hTau levels in the hippocampus of rTg4510 mice (Fig 3C). 1 μl of 109 particles of recombinant adenoassociated virus (rAAV) carrying GFP- or HA-tagged NMNAT2-WT, -ED, -ΔcATP was injected into the hippocampi of 6-wk-old rTg4510 mice (n = 17), which corresponds to the age when endogenous NMNAT2 begins to decline [32]. The impact of GFP or NMNAT2-WT, -ED, -ΔcATP expression on hTau was evaluated 1-mo post injection with immunostaining (Fig 3C). The MC1-antibody that recognizes conformational changes in hTau [79] was employed to evaluate the impact of NMNAT2 and its variants on hTau processing, while an HA-antibody identified exogenous NMNAT2 or its variants (Fig 3C and 3D). The overall intensity of MC1 immunoreactivity in the CA1 region of the hippocampi was quantified, and NMNAT2-WT (p < 0.001), -ED (p < 0.001), but not -ΔcATP (p = 0.68) significantly reduced the levels of conformationally altered hTau recognized by the MC-1 antibody (Fig 3C and 3D). To quantify the impact of NMNAT2 overexpression on hTau processing, western blot analysis was conducted with tissue homogenates prepared from EGFP and NMNAT2-WT, -ED, -ΔcATP rAAV transduced hippocampi of rTg4510 mice one mo postinjection (Fig 3E and 3F). The levels of p-hTau were evaluated with AT8 (recognizing p-S202/205 hTau [80]) and PHF-1 antibodies. Overexpressing NMNAT2-WT (p < 0.001), -ED (p < 0.001), but not NMNAT2-ΔcATP (p = 0.59) or GFP significantly reduced p-hTau levels in rTg4510 hippocampi while total hTau levels remained unchanged (Fig 3F). These data provide compelling evidence that NMNAT2’s chaperone activity but not its enzymatic function is critical in ameliorating proteotoxic stress caused by hTau overexpression.
How does NMNAT2 act as a chaperone? Does NMNAT2 function together with known chaperones HSP70/90 or cochaperones HOP and CHIP [81] to reduce protein aggregates? To determine if NMNAT2 interacts with other chaperones, we conducted immunoprecipitation experiments of NMNAT2 to determine whether HSP70, HSP90, HOP, or CHIP complex with NMNAT2 in hTau40 cell lines after induction of hTau40 expression. As shown in Fig 4A, NMNAT2 complexes with HSP90 but not HSP70, HOP or CHIP in NMNAT2-transfected HEK293-tau cells, via its C-terminus (S6 Fig). Similar to NMNAT2, HSP90 is also present in the insoluble fraction prepared from human AD brains but not control brains (Fig 1E). In aged rTg4510 cortex, NMNAT2 is also present in the insoluble fraction where p-hTau and HSP90 are concentrated (S5 Fig). NMNAT2 and HSP90 not only form complexes with each other (S6 Fig) in rTg4510 cortex (Fig 4B) but also bind to hTau (Fig 4B). These interactions are likely to be evolutionarily conserved as proteomic studies in Drosophila also identified HSP83, the Drosophila HSP90 homolog, as a dNMNAT-interacting protein (S4 Table).
HSP90 binds to non-native polypeptides to prevent aggregate formation, mainly at the late stages of substrate folding, a function that is tightly regulated [81]. To determine whether HSP90 is required for NMNAT2’s ability to reduce protein aggregates, we reduced HSP90 levels by either using HSP90-specific siRNA or by pharmacological blockade with 17-allylamino-17-demethoxygeldanamycin (17-AAG), an HSP90-specific inhibitor [82] (Fig 4C). We first examined the impact of reducing HSP90 expression level on NMNAT2’s ability to attenuate p-hTau burden. Briefly, cells were transfected with NMNAT2-WT, and either HSP90 siRNA or scrambled siRNA 12 h prior to doxycycline induction. p-hTau abundance in these cells was examined 48 h post doxycycline induction. We found that decreasing HSP90 function severely impaired NMNAT2’s ability to reduce p-hTau (Fig 4D; p < 0.001for siRNA).
To determine whether NMNAT2 requires HSP90 to exert its function as a holdase or a foldase, we performed luciferase-refolding assays (Fig 2A). The cells were transfected with Luciferase and NMNAT2-WT together with scrambled or HSP90 siRNA (Fig 4E). To test the drug, the cells were treated with 500 nM 17-AAG, 24 h prior to heat shock treatment. We found that HSP90 reduction by either HSP90-siRNA (p < 0.001) or 17-AAG (p < 0.001) prevented NMNAT2’s facilitation of the recovery of aggregated luciferase, but these treatments have a minimal impact on NMNAT2’s holdase activity (Fig 4E). These data suggest that NMNAT2 requires HSP90 for its foldase activity to reduce the proteotoxic burden.
HSP90 can regulate the stress response by modulating the activity of Heat Shock Factor 1 (HSF1). During nonstressed condition, HSP90 binds to HSF1 and p23-immunophilins [83]. This binding prevents HSF1 from activating the transcription of the heat shock response genes. Upon stress or heat shock, protein aggregates sequester HSP90 [83,84], freeing HSF1 to activate transcription of the heat shock response genes [83]. Hence, reduction or inhibition of HSP90 promotes the heat shock response (Fig 4C, S7 Fig). To assess if HSF1 plays a role in the chaperoning activity of NMNAT2, we blocked HSF1 effects on HSP transcription using 10 μM KRIBB-11 (KRI) (Fig 4C) [85,86]. KRI application minimally impacted NMNAT2’s chaperone activity (p = 0.47, Fig 4E). Moreover, coapplication of KRI with either AAG or an siRNA for HSP90 resulted in changes that were similar to those seen with AAG or siRNA alone (p = 0.54, compared to siRNA alone). Consistent with the data obtained from the luciferase assay, HSF-1 inhibition had no impact on NMNAT2-mediated clearance of p-hTau (Fig 4D). These observations rule out a contribution by HSF1 to NMNAT2’s chaperone function. Taken together, our results show that NMNAT2 requires HSP90 for its foldase activity to reduce the proteotoxic burden in this hTau model.
ATPase activity is often required for molecular chaperone activity. As demonstrated above, NMNAT2’s C-terminal ATP binding site is essential for its foldase activity. Hence, we tested whether this site has ATPase activity. Unlike HSP90, recombinant NMNAT2 shows no measurable ATPase activity in the presence of denatured CS substrate (Fig 4F, S8 and S9 Figs). However, addition of HSP90 stimulated NMNAT2’s ATPase activity (p = 0.0002 compared to HSP90; Fig 4F, S8B Fig). The importance of the C-terminal mutant missing this ATP binding site for this ATPase activity is demonstrated by the observation that the C-terminal mutant without this ATP site lacked ATPase activity in the presence of misfolded CS and HSP90 (p = 0.38 compared to HSP90; Fig 4F, S9 Fig). Our findings thus suggest that a complex of NMNAT2 with HSP90 induces a conformational change that stimulates NMNAT2’s ATPase activity, allowing it to function as a foldase (Fig 4G).
As nmnat2 transcript levels are reduced in various proteinopathies (S5 Table), we examined whether NMNAT2’s chaperone function is able to reduce aggregates of polyglutamine repeat expanded Spinocerebellar Ataxia type 1 (SCA1) protein. We conducted a cell-based human Ataxin1-[82Q]-GFP aggregation assay [87], in which the polyglutamine protein forms nuclear aggregates, observed in SCA1 pathology (Fig 5). Ataxin1-[82Q]-GFP forms numerous and often large protein aggregates in nucleus and cytoplasm of Cos7 cells (Fig 5A and 5B). In the presence of NMNAT2-WT (p < 0.0001), -ΔNT (p < 0.0001), -ED (p < 0.0001), -PM (p < 0.0001), the size of these GFP aggregates are significantly smaller than the aggregates found in mCherry expressing cells, or neighboring untransfected cells (empty). However, NMNAT2-ΔCT (p = 0.52) or -ΔcATP (p = 0.631) overexpression failed to reduce the size of Ataxin1-[82Q]-GFP aggregates, again suggesting that the chaperoning activity of NMNAT2 plays a critical role in dampening the formation of protein aggregates.
To distinguish between soluble and insoluble Ataxin1-[82Q]-GFP aggregates, cells were lysed and fractionated [87]. NMNAT2 overexpression reduced the amount of both soluble (p = 0.0004) and insoluble (p < 0.0001) Ataxin1-[82Q]-GFP aggregates (Fig 5C and 5D). Again, NMNAT2’s chaperone function, but not NAD synthase activity, is required for ameliorating the proteotoxic burden in SCA-1. Furthermore, in Ataxin1-[82Q]-GFP and NMNAT2 expressing Cos7 cells, NMNAT2 also complexes with HSP90, but not with HSP70, CHIP, or HOP (Fig 5E). Knocking-down HSP90 with HSP90 siRNA also prevented the reduction of soluble or insoluble ataxin1 aggregates by NMNAT2 (Fig 5F and 5G). However, HSF-1 inhibition with KRI minimally affected NMNAT2-mediated clearance of Ataxin1-82Q-GFP aggregates. Hence, HSF1-mediated transcription does not mediate NMNAT2’s chaperone function in this model system. All the above results provide strong evidence that the NMNAT2: HSP90 complex is capable of reducing aggregated proteins in different proteinopathies (S14 Fig).
Loss of NMNAT2 in mice causes death at birth [18]. To assess NMNAT2’s function in the central nervous system (CNS), cortical neurons were prepared from NMNAT2 WT (NMNAT2+/+), HET (NMNAT2+/-), and KO (NMNAT2-/-) embryos. We quantitatively compared the viability of NMNAT2 WT, HET, and KO cortical neurons at different days-in-vitro (DIV) using the MTT (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide) reduction assay, a colorimetric assay routinely used to estimate the number of viable cells [88]; Fig 6). We found that the viability of NMNAT2 KO neurons decreases as they mature and form numerous synapses (S10A Fig). Next, we treated DIV14 cortical neurons with 10 μM MG132, a proteosomal inhibitor, for 12 h to promote their demise via proteotoxic stress. Interestingly, loss of NMNAT2 also increases the vulnerability of DIV14 cortical neurons to proteotoxic stress (Fig 6A; p < 0.0001), suggesting that NMNAT2 deletion renders neurons sensitive to proteosomal stress. To determine whether NMNAT2’s chaperone function is required for its ability to protect cortical neurons against MG132-induced proteotoxicity, we transduced NMNAT2 KO cortical neurons with a lentiviral vector (LV) carrying NMNAT2 WT or various mutants at DIV2. At DIV14, the viability of these neurons was assessed by the MTT reduction assay after MG132 treatment. Overexpression of NMNAT2-WT (p = 0.0024), or -ED (p = 0.0013), but not NMNAT2-ΔcATP (p = 0.784), significantly increased the viability of NMNAT2 KO neurons post MG132 treatment (Fig 6A, S11 Fig). Interestingly, in WT neurons, NMNAT2 overexpression completely mitigated the toxic impact of protein stress (Fig 6A). Such protection by NMNAT2 depends on its chaperone but not enzymatic function. Hence, the chaperone function of NMNAT2 is required to protect cortical neurons against MG132 induced proteotoxicity, whereas its NAD synthesis activity is dispensable.
In Drosophila photoreceptors, dNMNAT protects neurons against activity-related insults [1]. Here, we found that upon massive glutamate release triggered by KCl treatment (30 mM for 24 h), the viability of NMNAT2 KO neurons was significantly decreased even at DIV10, while the viability of NMNAT2 WT and HET neurons was not affected by KCl treatment (Fig 6B, S10B and S11A Figs). At DIV20, both NMNAT2 HET (p < 0.01) and KO (p < 0.0001), neurons are significantly more sensitive to KCl-triggered excitotoxicity compared to NMNAT2 WT neurons (S10B Fig). Overexpression of NMNAT2-WT (p < 0.0001), -ΔcATP (p < 0.0001), but not NMNAT2-ED (p = 0.162), in NMNAT2 KO neurons significantly restored their resistance to KCl-induced cell death (Fig 6B, S11B Fig). Similarly, NMNAT2 overexpression in WT neurons completely prevented the excitotoxicity and this protection required NMNAT2’s enzymatic activity. Thus, cortical neurons require NMNAT2’s enzymatic but not chaperone function to protect them against excitoxicity. Taken together, both the enzymatic and chaperone activity of NMNAT2 are important to maintain neuronal health in a context-dependent manner.
Synapses are often the first structures lost in neurodegenerative disease [89]. Vertebrate NMNAT2 has been detected in synaptic membranes and vesicles of cortical neurons [17,24] suggesting a role for NMNAT2 in synapse maintenance. Immunostaining of DIV10 cultured neurons reveals a decrease of presynaptic proteins such as VGluT1 and synaptophysin (SYPH) in NMNAT2 HET neurons when compared to WT (S12 Fig. Almost no VGluT1 or SYPH signals are seen in the axons of NMNAT2 KO neurons. Loss of NMNAT2 significantly reduces levels of synaptic vesicle proteins such as SNAP25 (p = 0.0071) and SYPH (p = 0.032) as well as the active zone protein RIM1α (Fig 7A; p = 0.0028). In 8-mo-old NMNAT2 WT and HET brains, NMNAT2 protein levels are variable (Fig 7B and 7C). Interestingly, the abundance of presynaptic proteins like SYPH (p < 0.0001, r = 0.9633), VGluT1 (p < 0.0001, r = 0.9577), SNAP25 (p < 0.0001, r = 0.9526) and RIM1α (p < 0.0001, r = 0.9701) are positively correlated with NMNAT2 levels (Fig 7C). In addition, the levels of neurofilament (p = 0.033, r = 0.6159) and microtubule binding protein MAP2 (p = 0.0067, r = 0.7325) are also significantly correlated to NMNAT2 levels, whereas levels of HSP90 and N-methyl-D-aspartate receptor 1 (NR1) were not significantly related. Taken together, our data indicate that NMNAT2 plays a significant role in synaptic maintenance.
Here, we provide compelling evidence that NMNAT2 acts as a potent chaperone for denatured luciferase, p-hTau, and ataxin1-82Q-GFP. NMNAT2’s C-terminal ATP site is required for its chaperone activity but not for its NAD synthesis activity. The ED (NAD synthesis incompetent) NMNAT2 mutant has comparable chaperone activity to wildtype NMNAT2 in restoring luciferase activity as well as in reducing p-Tau and ataxin1-82Q-GFP. In rTg4510 mice, both WT- and ED-, but not chaperone-compromized-NMNAT2, were effective in reducing the abundance of both p-hTau and pathological hTau. This finding implicates an in vivo requirement of NMNAT2’s chaperone activity in p-hTau clearance. Its partial loss may therefore promote neurodegeneration in AD and other neurodegenerative diseases, whereas increased levels of NMNAT2 may protect against the demise of neurons.
NMNAT2 exerts both a holdase function, whereby it binds to aggregation-prone proteins to reduce their aggregation, and a foldase function, whereby it binds to and refolds aggregated or conformationally abnormal proteins. Using a CS aggregation assay, we found that NMNAT2 alone can act as holdase, while it requires HSP90 to refold CS aggregates. Importantly, the interaction with HSP90 activates NMNAT2’s C-terminal ATPase activity when protein aggregates are present. This C-terminus ATP binding site is required for NMNAT2’s foldase activity. It is likely that the ATPase activity triggered by HSP90 binding is required for NMNAT2 to refold its client proteins. Independently, a Drosophila proteomic screen also identified the HSP90 homolog, HSP83, as a dNMNAT interacting partner, suggesting the NMNAT2 and HSP90 interaction is conserved during evolution. In addition to hTau, HSP90 is also required for NMNAT2 to reduce CS aggregation as well as refolding of denatured luciferase and Ataxin1-polyQ aggregates. The conservation of such interactions across diverse systems attests to a broad role for the HSP90:NMNAT2 complex in mitigating proteotoxic aggregation.
Previous studies using nerve injury models have proposed that NMNAT2 supplies NAD to maintain axonal integrity [15–17,90]. Here, we found that deleting NMNAT2 in cortical neurons renders them sensitive to both proteotoxic stress and excitotoxicity triggered by sustained neuronal depolarization. In wildtype neurons, NMNAT2 overexpression has no impact on baseline cell viability but strengthens neuronal defense mechanisms to combat neural insults triggered by excessive protein stress or neurotransmission. Interestingly, the chaperone function of NMNAT2 is required to counter protein stress while the enzymatic function of NMNAT2 provides protection against excitotoxicity. The restoration of NMNAT2 KO neurons’ resistance to activity induced insults by over expression of enzymatically competent NMNAT2 reveals a similar context-dependent protection mechanism.
It remains to be determined how NMNAT2 switches between these two functional modes to protect against these distinct insults. Using Cos7 cells, we demonstrate that NMNAT2 only complexes with HSP90 upon protein stress (S13 Fig) despite both proteins being highly expressed. The observation of increased NMNAT2:HSP90 complex formation in the hippocampi of rTg4510 mice compared to wildtype mice suggests that protein stress triggered NMNAT2:HSP90 complex formation also occurs in vivo. Interestingly, NMNAT2’s ATPase activity is only activated when both HSP90 and protein aggregates are present. Taken together, we propose that NMNAT2 forms complexes with different partners depending on the nature of the insult and complex formation allows it to switch modes of actions to offer neuroprotection.
The localization of NMNAT2 in synaptosomal fractions prepared from whole mouse brains suggests the presence of NMNAT2 in synapses [24]. In Drosophila, dNMNAT maintains synaptic structure by stabilizing the active zone protein Bruchpilot [91], and photoreceptors that lack dNMNAT are sensitive to light or neural activity [1]. Loss of NMNAT2 significantly reduces the levels of presynaptic proteins. Previous studies have demonstrated a critical role of NMNAT2 in neurite outgrowth [18,19]. The reduction in axon number and length can account for the reduction of presynaptic proteins we detected in NMNAT2 KO neurons. However, immunostaining reveals reduced immunoreactivity for synaptic vesicle proteins VGluT1 and synaptophysin in NMNAT2 KO axons (S12 Fig). This suggests that NMNAT2 is required not only for axonal outgrowth but also for maintaining synaptic proteins in axonal arbors. The finding of a positive correlation between NMNAT2 levels and presynaptic proteins in adult mice provide additional support for the role of NMNAT2 in maintaining synaptic integrity. The dual-function of NMNAT2 makes it a very potent neuronal maintenance factor; it can provide sufficient local NAD to maintain neurons after high frequency neurotransmission and can also act as a chaperone to reduce proteinopathies. Such dual protection is likely to be important in preventing synaptic loss and preserve cognitive function during aging.
Our studies of NMNAT2 transcript levels, AD pathology, and cognitive performance in the human cohort studies represent associations, and on their own, do not establish causality. It is interesting that NMNAT2 but not NMNAT1 levels are reduced in AD brains. The shift of NMNAT2 solubility to an insoluble compartment in AD brains and its partitioning with hyperphosphorylated hTau, provides strong support for NMNAT2’s role as a molecular chaperone in the human brain. Due to NMNAT2’s short half-life [15,17], the appropriate level of NMNAT2 is likely to be actively regulated to ensure optimal neuronal function. In rTg4510 mice, NMNAT2 levels are reduced prior to neurodegeneration, a reduction due in part from decreased CREB-mediated nmnat2 transcription [32]. Decreased CREB activity has been reported in human AD brains and mouse AD models [92]. Thus, it is plausible in human that reduced CREB activity results in a decrease in NMNAT2 and synaptic loss, culminating in dementia. Deleting NMNAT2 function in mouse cortical neurons decreases presynaptic markers and increases sensitivity to excitotoxicity and protein stress. NMNAT2’s enzymatic activity is required to reduce excitotoxicity upon excessive neurotransmission, while it works together with HSP90 to reduce protein stress through its chaperone function. Elucidating how nmnat2 levels are regulated may help to reveal preventive therapies to protect neurons from nmnat2 down-regulation and cognitive decline. NMNAT2’s chaperone function is a promising therapeutic target to protect against CNS proteinopathies.
The ROS (ORA#91020181) and MAP (ORA#86121802) studies were conducted in accordance with the latest version of the Declaration of Helsinki and were approved by the Institutional Review Board of Rush University Medical Center. Written informed consent was obtained from all subjects, followed by an Anatomic Gift Act for organ donation. For human brain biochemistry, frozen cortical tissue was obtained from the Oregon Brain Bank at Oregon Health and Science University in Portland, OR. Tissue use conformed to institutional review board-approved protocols. Animal protocol 4307, 14–031 were approved by the IACUC at Baylor College of Medicine and Indiana University, respectively.
The ROS cohort, established in 1994, consists of more than 1,100 Catholic priests, nuns, and brothers from 40 groups in 12 states who were at least 55 y of age and free of known dementia at the time of enrollment. The MAP cohort, established in 1997, consists of more than 1,600 men and women primarily from retirement facilities in the Chicago area who were at least 53 y of age and free of known dementia at the time of enrollment. All participants in ROS and MAP signed an informed consent agreeing to annual detailed clinical evaluations and cognitive tests, and the rate of follow-up exceeds 90%. Similarly, participants in both cohorts signed an Anatomical Gift Act donating their brains at the time of death. The mean for the interval between last study evaluation and death is 8.2 mo. The overall autopsy rate exceeds 85%. All aspects of these studies were approved by the Institutional Review Boards of Rush University Medical Center and Partners Healthcare. More detailed information regarding the two cohorts can be found in previously published literature [52,53]. The global cognitive performance score has been previously described and was determined based on 17 cognitive tests, measuring 5 domains of cognitive function (episodic memory, visuospatial ability, perceptual speed, semantic memory, and working memory), and collected at annual evaluations proximate to death [55]. Results of each test were converted into a Z score using the mean and SD from baseline evaluations of all participants and then averaged to compute the composite measure of global cognition. For descriptive purposes and for the subgroup analysis presented in both S1 and S2 Tables, and Fig 1, the clinical diagnoses of dementia and AD were made following National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer’s Disease and Related Disorders Association recommendations [93,94]. Mild cognitive impairment (MCI) referred to those individuals rated as cognitively impaired by the neuropsychologist but not demented by the examining physician, as previously described [95]. The quantitative global AD pathology measure has been extensively described in prior publications [56]. Bielschowsky silver stain was used to visualize and count neuritic plaques and neurofibrillary tangles on representative sections from midfrontal, middle temporal, inferior parietal, and entorhinal cortices, and the hippocampal CA1 sector. Raw counts were divided by the population standard deviation of the region-specific counts, and the scaled counts were averaged over the five brain regions to create a global summary measure. RNA-seq data was generated using postmortem brain tissue (dorsolateral prefrontal cortex) from ROS/MAP subjects. The RNA was extracted from the tissue after the quality control evaluation based on RIN score [96] or RNA were extracted first, and then RIN was used to filter out low-quality samples. The library was sequenced using Illumina Hi-Seq with 101 bp reads and 4-plex pooling. All the paired-end reads were mapped using RSEM alignment using Bowtie as the aligner [97] using the human genome transcriptomic database from Ensemble (http://www.ensembl.org). The Fragment Per Kilobase Per Million (FPKM) was calculated for the nmnat1 or nmnat2, and these values were regarded as expression quantity for each gene and its isoform for further analysis. Linear regression was used to evaluate the relationship between transcript levels (nmnat1 or nmnat2) and either global cognition or global AD pathology summary score. Since the scaled outcome of AD pathologic burden was right skewed, we applied square root transformation prior to the analyses. All analyses were adjusted for age at death, gender, postmortem interval, and RIN score. p < 0.0125 was considered significant after adjusting for multiple hypothesis testing (α = 0.05 divided by 4 [2 transcripts + 2 outcomes]. Path analysis was used to evaluate structural models linking nmnat2 mRNA expression, global AD pathology, and global cognition proximate to death. Standardized path coefficients along individual pathways were used to assess the total direct and indirect effects of nmnat2 expression on cognition. All tests were performed using SAS software, version 9.3 (SAS Institute, Cary, NC) and Mplus, version 7.0 (Muthen & Muthen, 1998–2012).
Frozen cortical tissue was obtained from the Oregon Brain Bank at Oregon Health and Science University (OHSU) in Portland, OR. These subjects were enrolled and clinically evaluated at the NIH-sponsored Layton Aging and AD Center (ADC) at OHSU. Every subject received annual neurological and neuropsychological evaluations, with a clinical dementia rating (CDR) assigned by an experienced clinician. The AD subjects were diagnosed according to the National Institute for Neurological and Communicative Disorders and Stroke-Alzheimer’s Disease and Related Disorder Association diagnostic criteria for clinical AD, followed by neuropathologic confirmation at autopsy. Neuropathologic assessment conformed to National Institute on Aging-Reagan consensus criteria. All brain tissue was examined by a neuropathologist for neurodegenerative pathology including neurofibrillary tangles and neuritic plaques and assigned an amyloid score based on the deposition of amyloid plaques in the brain (0, no plaques; 1, sparse plaques; 2, moderate plaques; 3, dense plaques), and a Braak stage (0–6; with 6 being the most severe) indicative of the level and location of hyperphosphorylated tau tangles [98]. In addition to the pathological information detailed above, demographical data including age, sex, and MMSE score for each case were received along with the frozen tissue (S3 Table).
For each sample, RNA concentration was determined by spectrophotometry at 260 nm. Two micrograms of RNA were used for the reverse transcription reaction with the High Capacity RNA to cDNA kit (Applied Biosystems, Carlsbad, CA). Quantification of nmnat2 and nmnat1 mRNA was performed using a CFX Touch Realtime PCR System (BioRad). The housekeeping genes, 18s and gapdh, were used as internal controls to normalize mRNA expression. Amplification mix (25 μl) consisted of 0.1 μg cDNA, 12.5 μl TaqMan Universal PCR Master Mix (Applied Biosystems) and 1 μl TaqMan Gene Expression Assay (Hs00322752_m1 nmnat2, Hs00276702_m1 nmnat1, Hs99999901_s1 18s, Hs02758991_g1 gapdh, Mm99999915_g1). Samples were amplified by a PCR program of 40 cycles of 10 s at 95°C, 15 s at 55°C, and 1 min at 72°C. The Ct value was defined as the number of cycles required for the fluorescence to exceed the detection threshold, and the data were analyzed using the 2−ΔΔCt method to quantitatively assess relative changes in gene expression [99]. Briefly, the raw Ct values were averaged for samples per group (AD, NDAN, and age-matched controls) for nmnat1, nmnat2, gapdh, and 18s. ΔCt values were calculated by subtracting the geometric means of gapdh and 18s Ct value from that of nmnat1 and nmnat2. The relative expression of nmnat1 and nmnat2 in AD and NDAN samples compared with controls was obtained by calculating the ΔΔCt value for each group by subtracting the average ΔCt value of control nmnat1/nmnat2 from AD/NDAN average ΔCt. The relative fold change of AD/NDAN groups to control was expressed as 2−ΔΔCt.
Total RNA was extracted from human brain samples using Trizol reagent (Life Technology, Grand Island, NY). 0.5 mL of 100% isopropanol was added to the aqueous phase, per 1 mL of TRIzol Reagent used for homogenization and the samples were incubated at room temperature for 10 minutes before centrifuging at 12,000 × g for 10 minutes at 4°C. RNA was purified from the extracts using RNeasy Lipid Tissue Mini kit (Qiagen, Gaithersburg, MD) according to manufacturer’s protocol. RNA concentration and purity were determined using a Nanodrop spectrophotometer.
Human brain samples were prepared for protein extraction using the differential fractionation method previously described [61,62]. This method relies on the use of increasingly stringent detergents that allow for soluble and peripheral membrane proteins to be extracted first, followed by extraction of membrane-bound proteins, and finally the most insoluble proteins are extracted into a 2% SDS buffer. Briefly, brain tissue was collected and frozen immediately with liquid nitrogen. Prior to extraction, the frozen brain tissue was weighed and homogenized in 15 volumes of ice-cold TBS (50 mM Tris-HCl, pH 7.4 and 150 mM NaCl) containing protease and phosphatase inhibitors (Roche, Branford, CT). The samples were centrifuged at 175,000 x g for 1 h at 4°C and the supernatants collected as the TBS fraction. The pellets were resuspended in ice-cold TBS buffer, spun at 175,000 x g for 1 h at 4°C, and the supernatants were discarded. Ice-cold TBS (pH 7.4) containing protease and phosphatase inhibitors and 1% Triton X-100 (TBS-X) were added to the pellet and the suspension homogenized. The samples were centrifuged at 170,000 x g for 1 h at 4°C and the supernatants collected as the TBS-TritonX fraction (soluble fraction). During the next myelin flotation phase, the pellet from the previous step was washed in TBS-TritonX buffer with 30% sucrose at room temperature for 1 h (with rotational shaking). This fraction was centrifuged at 175,000 x g for 1 h at 4°C, and the supernatants were discarded. The final pellet containing the insoluble material was re-suspended in 2% SDS buffer (insoluble fraction). Protein concentrations were determined using the Bradford Assay (BioRad, Hercules, CA).
Upon measurement of total protein concentrations, this extraction scheme applied to control brains yields 32% of total proteins in the TBS fraction, 65% in the TBS-TritonX fraction, and around 2% in the insoluble fraction. In comparison, in AD brains, around 26% of total protein is present in the TBS fraction, 68% in TBS-TritonX fraction and roughly 5% in the insoluble fraction, reflecting the higher level of insoluble protein present in AD brains.
For analysis of protein expression, 10 μg of total protein lysate per sample were resolved by sodium dodecyl sulfate polyacrylamide gel electrophoresis and transferred onto a nitrocellulose membrane, and probed with antibodies against the HA-tag (1/5,000; Cell Signaling, Danvers, MA), ß-actin (1/4,000; Sigma), hTau (1/1,000; DAKO, Glostrup, Denmark), GFP (1/1,000, Invitrogen), GAPDH (1/5000, Millipore, Temecula, CA), PHF-1 (1/1,000, gift from Dr. Peter Davies), AT8 (P1/1,000 Pierce/Thermofisher, Waltham, MA), Neurofilament-M (1/2,000, Millipore, Temecula, CA), HSP70 (1/1,000, Abcam, Cambridge, MA), HSP90 (1/1,000, Novus Biological, Littleton, CO), CHIP (1/1,000, Novus Biological, Littleton, CO), HOP (1/1,000, Novus Biological, Littleton, CO). Western blot analysis was performed with infrared dye conjugated secondary antibodies, IR700 and IR800 (1/10,000, LI-COR Biosciences, Lincoln, NE). Blots were imaged and processed on an Odyssey Infrared Imaging System. Densitometry analysis was performed on all the blots using ImageJ software (NIH). Data are represented as means ± SEM.
To study changes in neuronal proteins in aged NMNAT2-HET and wild type littermate brains, the cortex and hippocampus were dissected out and homogenized in Syn-PER Synaptic Protein Extraction buffer (ThermoFisher Scientific, GrandIsland, NY). Part of the total homogenate from each brain was stored, while the rest was used for synaptic fractionation, to enrich for synaptic proteins, using manufacturer’s protocol. Protein concentrations were calculated using Bradford Assay (BioRad). For analysis of protein expression, 10 μg of total protein lysate per sample were resolved by sodium dodecyl sulfate polyacrylamide gel electrophoresis, transferred onto a nitrocellulose membrane, and probed with antibodies against Synaptophysin (1/4,000, Millipore, Temecula, CA), SNAP-25 (1/1,000, Synaptic Systems, GmbH, Germany), Neurofilament (1/1,000, Millipore, Temecula, CA), VGlut2 (1/1,000, Synaptic Systems, GmbH, Germany), GAPDH (1/5,000, Millipore, Temecula, CA).
pCMV-HSP70, pCMV-NMNAT1, pCMV-NMNAT2, pCMV-NMNAT3 and pCMV-cytLuciferase were gifts by Dr. R. Grace Zhai (University of Miami). pEGFP-hAtaxin-1[82Q] was a gift by Dr. Huda Y. Zoghbi (Baylor College of Medicine). Site-directed mutagenesis was performed on the pCAG-HA-NMNAT2 backbone using a Quik-ChangeII Site-Directed Mutagenesis kit (Agilent, Santa Clara CA). The sequences of primers used for mutagenesis are listed below. NMNAT2W92G-forward: 5’-TGC TAC CAG GAC ACC GGT CAG ACG ACC TGC AGC-3’. NMNAT2W92G-reverse: 5’-gct gca ggt cgt ctg acc ggt gtc ctg gta gca-3’. NMNAT2ΔNT-forward, 5’- gga aaa cag ggc ctc gtg tca agc cgg-3’ NMNAT2ΔNT-reverse. 5’- cgt ggg tct tgg tgg tct cgg t-3’. NMNAT2ΔCT-forward, 5’- ctg tac atc aat gcc tcc ggc tag-3’. NMNAT2ΔCT-reverse, 5’- cac tac cac aca gca gca gga tcc gta gct c-3’. NMNAT2ΔcATP-forward, 5’-gtg gat tac ctg tcc cag ccg gtc atc gac tac-3’. NMNAT2ΔcATP-reverse, 5’-ctt ggt tga gct gac aac aga cat ggg-3’. NMNAT2C164S/C165S-forward, 5'-aag cct cag ccg gat ctc ctc tgt ccg ccc-3'. NMNAT2C164S/C165S-reverse, 5'-ggg cgg aca gag gag atc cgg ctg agg ctt-3'. For rAAV generations, cDNAs carrying NMNAT2-WT, -ED, -ΔcATP were respectively cloned into BamHI or NotI sites 5’ to the expression cassette-IRES-EGFP driven by the elongation factor-1 (EF-1) promoter within the AV4-EF1-GFP shuttle plasmid as described [32].
The CS aggregation assay was performed as described previously [67]. Substrate CS (Sigma, St. Louis, MO) was desalted and mixed with either egg white lysozyme (Sigma), NMNAT1/2/3 (R&D Systems, Minneapolis, MN), or HSP70 (R&D Systems, Minneapolis, MN) at varying concentrations in HEPES (pH 7.4) buffer for 30 min. To generate recombinant NMNAT2 proteins, the coding sequences were cloned into pET28b vector for bacterial expressions. The recombinant proteins of NMNAT2 and its variants were prepared by the MD Anderson Proteomic Core. Aggregation of denatured CS was initiated at 43°C and was monitored as Raleigh scattering absorbance at 360 nm as a function of time. A FluoStar Optima plate reader (BMG Labtech, Cary, NC) was used for absorbance measurements. The relative chaperone activity of NMNAT was calculated as the scattering of CS aggregates with time versus NMNAT concentration.
The refolding assay of chemically denatured CS was performed as described previously [100]. Briefly, the concentration of desalted CS was measured at A280 (for CS, A280(0.1%) = 1.78). 1 μM CS was denatured in 6 M guanidine hydrochloride, 1 M Tris-HCl (pH = 8) and 50 mM dithiothreitol (DTT) for 1.5 hours at 25°C. Renaturation of aggregated CS was initiated by adding 0.5 M ATP and 1 μM of respective chaperones and studied by measuring light scattering at 360 nm as a function of time. A FluoStar Optima plate reader (BMG Labtech, Cary, NC) was used for absorbance measurements.
The luciferase aggregation assay was performed as described [101,102], with a few modifications. Cos7 cells were transfected with pCMV-cytLuciferase and one of the following plasmids: pCMV-HSP70, pCMV-NMNAT1, pCMV-NMNAT3, pCAG-mCherry, HA-NMNAT2, HA-NMNAT2W92G(ED), HA-NMNAT2C164S/C165S (PM), HA-NMNAT2ΔNT, HA-NMNAT2ΔCT, or HA-NMNAT2ΔcATP, using Lipofectamine (Invitrogen, Grand Island, NY). 48 h post-transfection, the cells were treated with the protein synthesis inhibitor cycloheximide (50 ug/mL for 3 h). One batch of cells was lysed immediately with luciferase lysis buffer (Promega, Madison, WI), while two similar batches were heat-shocked at 42°C for 15 min (which induced efficient unfolding of luciferase without killing the cells). Second batch of cells was lysed immediately after heat shock, while the final batch was allowed to recover at 37°C for 3 h. Luciferase activity was measured with the Luciferase Assay System (Promega).
NMNAT enzyme activity was measured using Amplite Colorimetric Total NAD and NADH Assay Kit (AAT Bioquest, Sunnyvale, CA). Briefly, Cos7 cells overexpressing different NMNAT isoforms and mutants were lysed 48 h post transfection using manufacturer’s protocol. For measuring NAD levels from hippocampi, tissue was extracted in the lysis buffer provided in the kit and NAD was measured using 50 ug total protein per reaction. NAD/NADH levels were measured using supplied standards using an absorbance microplate reader at ~576 nm.
The ATPase activity of NMNAT2 variants and HSP90 was measured with or without aggregated CS using the EnzCheck Phosphate Assay kit (Life Technologies, Grand Island, NY). Briefly, this spectrophotometric microplate-based assay detects a change in absorbance from 330 nm to 360 nm, when the 2-amino-6-mercapto-7-methylpurine riboside (MESG) substrate is converted by the purine nucleoside phosphorylase (PNP) enzyme to the ribose MESG product, in the presence of inorganic phosphate. The reactions were performed in a 96-well plate, in the presence of 500 mM ATP and either 1 mM NMNAT2-WT, -ΔcATP or HSP90, in the absence or presence of denatured CS. Measurements were made after 5 min of incubation according to the manufacturer’s protocol.
Inducible tau stable Trex-293 cell lines expressing human 2N4R tau (hTau40) were a gift from Dr. Jeff Kuret [77]. To maintain stably-transfected cells containing the Tau transgene, these were cultured in selective medium containing 500 μg/ml G418 (Sigma). Cells were transfected with pCMV-mCherry, HSP70, HA-NMNAT2, HA-NMNAT2-ED, HA-NMNAT2-PM, HA-NMNAT2-ΔNT, HA-NMNAT2-ΔCT or HA-NMNAT2-ΔcATP using Lipofectamine. 6 h post transfection, Tau expression was induced by adding 1 μg/mL tetracycline (Sigma) to cell culture media. 48 h postinduction, the cells were lysed in RIPA buffer to extract total protein, which was assayed for PHF-1 levels using western blotting.
HA-NMNAT2 was expressed in Cos7 cells using lipofectamine. 48 h post transfection, cells were treated with DMSO or 10 μM MG132 for 12 h. Cells were lysed in RIPA buffer and protein concentration assayed using the Bradford method. 1 mg total protein was used as the starting material for immunoprecipitation with HA-antibody or HSP90 antibody. Mouse IgG was used as a negative control in the experiment. Immunoprecipitates were probed with specific antibodies to reveal protein–protein interactions.
The generation and genotyping of rTg(tauP301L)4510 (rTg4510) and NMNAT2-Blad mice have been described previously [18,76]. rTg4510 mice over-express the P301L mutation in 4R0N human tau associated with FTDP-17 [76]. NMNAT2 null mutation was generated by transposon-mediated gene-trap mutagenesis [18]. Animal housing and use were in compliance with the NIH Guidelines for the Care and Use of Laboratory Animals and were approved by the institutional animal care committee at Baylor College of Medicine and Indiana University.
To quantitatively compare the in vivo efficacy of NMNAT2 and its variants in reducing toxic tau species, we overexpressed GFP, and NMNAT2-WT, -ED, -ΔcATP in the hippocampi of 6-wk-old rTg4510 mice through rAAV. Serotype 6 rAAVs carrying cDNA’s of NMNAT2-WT, -ED, or -ΔcATP were prepared by cloning into the expression cassette-IRES-EGFP driven by the EF-1 promoter within the AV4-EF1-GFP shuttle as described [32]. GFP-rAAV was also generated as control rAAV. 1 μl rAAV-NMNAT2-WT, -ED, -ΔcATP, or 1 μl of rAAV-EGFP (108 particles per μl) was injected into the CA1/dentate gyrus area of the hippocampus in one hemisphere (control mice: n = 6 for GFP, n = 2 per NMNAT2-WT, -ED, and ΔcATP; rTg4510 mice: n = 6 for GFP, n = 6 for NMNAT2-WT, n = 5 for -ED, and n = 6 for ΔcATP) at the following coordinates measured from bregma: A: 2 mm, L: 1–1.5 mm, V: 1.2 mm (tooth bar set at zero). The efficacy of these rAAVs on toxic tau clearance were quantitatively evaluated with immunostaining and western blotting using the tissue prepared from GFP-AAV and NMNAT2 or NMNAT2 variants-AAV-transduced hippocampi of rTg4510 mice at 1 mo postinjection.
One month after rAAV injections, rTg4510 and control mice were deeply anesthetized, transcardially perfused with PBS, followed by 4% Paraformaldehyde (PFA) solution in PBS, after which the brains were removed and postfixed overnight in 4% PFA. The brains were serially sectioned in the coronal plane into 50 μm thick sections using a Leica VT-1000 vibrating microtome (Leica Microsystems, Bannockburn, IL). Brain slices containing comparative complete hippocampal morphology were selected and stained with MC-1 antibody (recognize pathological conformation of hTau, a generous gift from Peter Davies, Feinstein Institute for Medical Research, Manhasset, NY; 1:500), HA-antibody (1:1,000; Cell Signaling, Danvers, MA) and DAPI (nuclear stain). For quantitative analysis of MC1-positive neurons in the hippocampus, we imaged areas enriched with HA-positive pyramidal neurons in the CA1 and quantified HA and MC1-immunoreactivity in the striatum pyramidale layer using NIH ImageJ software, as described previously [103].
Cortical neurons were prepared from E16.5 NMNAT2 WT, HET, or KO embryos and cultured on Poly-D-Lysine coated coverslips. DIV10 neurons were fixed with 4% PFA and 3% Sucrose in PBS and immunostained for synaptic markers Synaptophysin (1/1,000, Millipore, Temecula, CA), and VGlut1 (1/500 (1/1,000, Synaptic Systems, GmbH, Germany), the axonal marker ßIII-Tubulin (1/1,000, Millipore, Temecula, CA) and MAP2 (1/2,000, Millipore, Temecula, CA) as described [104,105]. Cells were imaged using fluorescent Alexa Fluor 555, 488, and 647 conjugated antibodies, with a Leica TCS SPE personal confocal.
Cortical neurons were prepared from the cortices of E16.5 embryos from NMNAT2 wild type, heterozygous, or homozygous [18] using the Worthington Papain Dissociation Kit (LK003153, Worthington Biochemical Corporation, Lakewood, NJ). 50,000 neurons were plated per well in polylysine-D coated 96-well plates (BD Biosciences, San Jose, CA) in Neurobasal media (Gibco, Grand Island, NY). Following plating, half of the media was replenished with B-27 and L-glutamine supplemented fresh Neurobasal media on Days-in-Vitro (DIV2), DIV7 and DIV14 and DIV18. Neurons were treated with prior to treatment with MG132 (10 μM) or DMSO (control) on DIV14 for 12 h. For KCL treatment, neurons were treated at DIV5, 10, 15, or 20 with 30 mM KCl for 24 h. Following either treatment, the MTT assay was performed as described [106]. MTT assay measures cell viability based upon the ability of viable cells to reduce the tetrazolium dye MTT 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide to its insoluble form formazan [88]. MTT assay was performed on primary cortical neurons grown in 96-well plates, using Vybrant MTT Cell Proliferation Assay Kit, according to manufacturer’s guidelines (Life Technologies). The absorbance values for cells treated with MTT were measured using the 540 nm wavelength, Opsys MR spectrophotometer, and Revelation Quicklink software.
Wild type NMNAT2, ED, PM, and cATP deletion mutants were cloned into LV provided by the DERC Gene Vector Core at BCM. These LV vectors were concentrated by ultracentrifugation and supplied at roughly 1 x 105 tu/uL. Neurons plated at a concentration of 100K/mL onto Poly-D-Lysine coated coverslips in 24-well plates or 96-well plates for MTT assay, were infected on DIV2 with 1 uL (105) LV particles to transduce almost all the neurons plated. For imaging, coverslips were either fixed at DIV5 or DIV10 with 4%PFA+ 3%Sucrose solution and processed for immunostaining (see below).
Heads from approximately 5 ml of Transgenic flies carrying P{w[+mW.hs] = GAL4-da.G32}UH1; UAS-HA:Nmnat [1] were homogenized in a dounce homogenizer in lysis buffer containing 25 mM Tris-HCl (pH 7.5), 150 mM NaCl, 5 mM EDTA, 1% NP-40, 1 mM DTT and protease and phosphatase inhibitor (Thermo Scientific, Rockford, IL). The resulting supernatant was collected without centrifugation after 5 min settling on ice and used for coimmunoprecipitation using Sigma Ezview Red Anti-HA affinity gel along with the supplied protocol. Bound proteins were eluted using Sigma HA peptide at 100 ug/ml.
The mass spectrometry analysis was performed in the Mass Spectrometry-Proteomics Core Facility in Baylor College of Medicine. Two microliter aliquots of each sample were quantified using NanoOrange protein quantification kit (Invitrogen). All samples were normalized to one microgram per 100 μL with 50 mM ammonium bicarbonate buffer (pH 7.9). The samples were reduced with 100 mM DTT (BioRad) in a rotary shaker at 800 rpm at room temperature for 30 min. The reduced cysteine residues were further treated with 400 mM iodoacetamide (Sigma) in the dark at room temperature for 30 min in the same shaker. 40 ng of sequencing grade trypsin (Sigma) were added to the samples for digestion. The digests were supplemented with acetonitrile to 10% of total volume. The digestion was allowed to proceed at 37°C for 16 h by mild shaking. Digestion was stopped by the addition of formic acid to 5% final volume and tubes were Speedvac dried. Digests were resuspended in 0.1% formic acid and 5% acetonitrile solution. The concentration of each digest was measured using NanoOrange and 200 ng of each sample was fed into the Eksigent nanoLC system. The Ekisgent nanoLC and the ABCIEX TripleTOF 5,600 mass spectrometer were controlled by Analyst software, version 1.6 (ABCIEX Inc., Framingham, MA).
The Eksigent nanoLC system has a cHiPLC system. The chips contained a trap column (200 μm × 0.5 mm, ChromXP C18-CL, 3 μm, 120Å) that trapped the injected peptides in a flow rate of 3 uL/min in a buffer composed of 50% of 0.1% formic acid and 50% of acetonitrile for 5 min at 23°C. The second section was an analytical column (75 μm × 15 cm, ChromXP C18-CL, 3 μm, 120 Å) with an organic mobile phase gradient set at a 90 min gradient starting from 5 to 35% acetonitrile in 0.1% formic acid at a flow rate of 300 nL/min. The acetonitrile concentration was increased to 80% in 5 min and then held in 80% acetonitrile for an additional 5 min. The final acetonitrile concentration was reduced to 5% within 5 min and then the column was equilibrated for 20 min to prepare for the next sample.
The Eksigent nanoLC was linked via the nanoflex to the ABCIEX TripleTOF 5600 mass spectrometer. The spray tip was the PicoTip emitter (360 μm OD, 20 μm ID, 12 cm long with 10 μm tip opening size) from New Objective. The GS 1 gas was set at 3 units and the curtain gas was set at 24 units. The ionization voltage was set at 2,400 V. The interface heater temperature was set at 150°C. The MS precursor ions were selected from 400 to 1,250 amu. The cycle time was 0.25 s and the forty most abundant ions were allowed to pass to the q2 for product ion production. The MS/MS spectra were acquired at 0.1 second at the m/z range of 100 to 2,000. A beta-galactosidase tryptic digest was used as a calibration standard every sixth run. Raw WIFF files were fed into Protein Pilot version 4.5 (Paragon-based) software (ABCIEX Inc.) for peak generation and database search with <10% false discovery rate. The organism was set as Drosophila and the database was the NCBI nonredundant database.
The ataxin-1 aggregation assay was performed as described [87]. Briefly, Cos7 cells were doubly transfected with either pEGFP-hAtaxtn-1[82Q] and pCAG-mCherry, HA-NMNAT2, HA-NMNAT2W92G (ED), HA-NMNAT2C164S/C165S (PM), HA-NMNAT2ΔNT, HA-NMNAT2ΔCT or HA-NMNAT2ΔcATP using Lipofectamine (Invitrogen). After 48 h, cells were fixed and immunolabelled to detect HA-tagged NMNATs. Cell nuclei were labeled with DAPI (Molecular Probes, Grand Island, NY). The size of GFP-positive aggregates was measured using Neurolucida software.
To distinguish between soluble and insoluble ataxin, the cells were lysed in RIPA buffer (50 mM Tris (pH 8), 150 mM NaCl, 1 mM EDTA, 1% NP40, 0.5% sodium deoxycholate, 0.1% SDS, 1 mM PMSF and protease inhibitor cocktail (Roche, Branford, CT) and fractionated according to an established protocol [87]. The detergent-soluble fraction was defined as proteins remaining in the supernatant after centrifugation of the cell lysates at 120,000 g for 15 min. After removal of the supernatant, the pellets were washed with RIPA buffer and then suspended in 2% SDS Buffer. One assumption made in this assay was that all cells transfected would have likely chance of incorporating both pEGFP-hAtaxin-1[82Q] and the test chaperone protein.
Immunofluorescence staining was performed on cultured cells. Cells were fixed in 4% PFA for 20 min and washed with PBS/0.01% Triton X-100 (PBST) and permeabilized with 0.2% Triton X-100 in PBS at room temperature for 5 min. Next, blocking was performed for 1 h using 5% normal goat serum in PBS. Following this step, the cells were stained for the HA-tag (1:1,000, Cell Signaling) diluted in PBST/1% normal goat serum at 4°C overnight. The next day, cells were washed with PBST, and incubated with goat anti-rabbit or goat anti-mouse 594 secondary antibodies (all at 1/500, Invitrogen) in PBST at room temperature for 2 h. Following this incubation, cells were washed with PBST three times for 10 min each and mounted in Vectashield mounting media with DAPI (Vector Labs, Burlingame, CA) and cover slipped for imaging. Fluorescence microscopy was performed using a Zeiss AxioImager M1 system equipped with epifluorescence filters, a Zeiss monochrome digital camera and AxioVision software. All images were processed in Adobe Photoshop for brightness/contrast, orientation and background correction to better illustrate staining patterns.
Means were compared across groups using one-way analysis of variance (ANOVA) for ≥ three groups or t-tests for two groups for data with normal distribution that met variance homogeneity. Significance was assessed using the Tukey criterion for pairwise mean comparisons under the ANOVA model. Normality was assessed using the Kolmogorov-Smirnov test. When homogeneity of variance was not met under the ANOVA model, a robust ANOVA was carried out. Two-tailed p-values <0.05 were considered significant. Values are expressed as mean ± SE. Computations were carried out using Graphpad Prism for Windows, version 6.0 (La Jolla, CA).
|
10.1371/journal.pgen.1005445 | The Evolutionary Potential of Phenotypic Mutations | Errors in protein synthesis, so-called phenotypic mutations, are orders-of-magnitude more frequent than genetic mutations. Here, we provide direct evidence that alternative protein forms and phenotypic variability derived from translational errors paved the path to genetic, evolutionary adaptations via gene duplication. We explored the evolutionary origins of Saccharomyces cerevisiae IDP3 - an NADP-dependent isocitrate dehydrogenase mediating fatty acids ß-oxidation in the peroxisome. Following the yeast whole genome duplication, IDP3 diverged from a cytosolic ancestral gene by acquisition of a C-terminal peroxisomal targeting signal. We discovered that the pre-duplicated cytosolic IDPs are partially localized to the peroxisome owing to +1 translational frameshifts that bypass the stop codon and unveil cryptic peroxisomal targeting signals within the 3’-UTR. Exploring putative cryptic signals in all 3’-UTRs of yeast genomes, we found that other enzymes related to NADPH production such as pyruvate carboxylase 1 (PYC1) might be prone to peroxisomal localization via cryptic signals. Using laboratory evolution we found that these translational frameshifts are rapidly imprinted via genetic single base deletions occurring within the very same gene location. Further, as exemplified here, the sequences that promote translational frameshifts are also more prone to genetic deletions. Thus, genotypes conferring higher phenotypic variability not only meet immediate challenges by unveiling cryptic 3’-UTR sequences, but also boost the potential for future genetic adaptations.
| The rarity of genetic mutations limits the likelihood of adaptation. However, transcriptional and translational errors, so-called phenotypic mutations, are >105-fold more frequent, thus generating protein mutants from unmodified genes. We provide the first evidence that phenotypic mutations paved the path to what later, after gene duplication, became newly compartmentalized enzymes. Thus, gene duplication followed rather than initiated the divergence of this new trait. Our findings also show that translational infidelity and phenotypic variability comprise the origins of evolutionary innovations, and how selection for enhanced phenotypic variability also promotes the appearance of genetic mutations that lead to the very same outcome.
| Latent, promiscuous protein functions serve as starting points for evolving new functions, thus resolving the evolutionary ‘catch’ of no new trait can evolve unless it already exists and can confer an immediate survival benefit [1,2,3]. Along the same veins, it has been proposed that other forms of molecular infidelity, such as transcriptional and translational errors, may also underlie the evolution of new protein traits [4,5,6]. Indeed, these so-called ‘phenotypic mutations’ yield protein variability from an unmutated gene and are up to 105 times more frequent than genetic mutations [7,8,9]. Phenotypic mutations may thus bridge the crucial and relatively long time gap between the appearance of a new challenge and the emergence and fixation of changes in genotype, i.e., evolutionary adaptations as often manifested in new, paraloguous genes. To this date, however, no direct evidence exists for a phenotypic mutation paving the path to a genetic, evolutionary adaptation.
Gene duplication is the source of new paralogs, including proteins with new activities or new subcellular localizations. However, different mechanisms may underlie the emergence of new functions via gene duplication. The first proposed mechanism, now known as Ohno’s model, or neo-functionalization, is initiated by duplication as a random event generating a redundant gene copy that acquires mutations under no selection. If and when a new function becomes beneficial, and if the drifted copy happens to provide this new function, the duplicated gene becomes under selection thus giving rise to a new paralog [10]. The discovery of multi-functional proteins prompted alterative models by which gene duplication follows rather than precedes the emergence of new functions. By these models, ‘gene sharing’ [11], sub-functionalization [12], or more explicitly, ‘divergence before duplication’ [13,14], the new function initially develops in the original, pre-duplicated gene. Mutations, that are largely neutral with respect to the primary, original function, may give rise to latent, promiscuous functions, which, in turn, may become under selection if and when needed [15,16]. The new function therefore becomes under selection alongside the original one, giving rise to a bi- or multi-functional protein (gene sharing). Duplication may occur at a much later stage, thus allowing the two functions to be split between two paraloguous genes (sub-functionalization).
The current literature primarily addresses how new binding, regulatory or enzymatic functions evolve via duplication, thus providing ample evidence for divergence via multifunctional ancestors [17,18,19,20,21]. However, protein function relates not only to what a protein does but also to where it functions and with which partners. In eukaryotic cells, for example, proteins localize to different subcellular compartments to perform their designated functions. Indeed, about a third of duplicate protein pairs derived from the yeast whole-genome duplication (WGD) that occurred along the lineage leading to S. cerevisiae localize to different subcellular compartments. However, the evolutionary mechanisms underlying the divergence of gene paralogues with new subcellular localizations remain largely unknown [22]. Divergence before duplication, and a subsequent sub-functionalization to two paralogues, demands the appearance of the new trait within the ancestral, pre-duplicated gene while maintaining its original function [14,23,24,25]. In the case of localization, this means dual localization, a phenomenon that is in fact well recorded [26,27,28]. Amongst other mechanisms, the partial expression of protein forms carrying targeting signal sequences may occur via alternative splicing or transcriptional/translational errors [29,30,31]. To this date, however, no particular example exists whereby a phenotypic mutation led to the divergence of a new paralog in a recently diverged species.
To study the history of evolution of new protein localizations, we sought to examine duplicate gene pairs that derived from the yeast WGD and diverged in their cellular localizations. Of the potential candidates listed in the literature [22,32,33], one gene stood out in having a clear-cut selectable phenotype, and hence being amenable to laboratory evolution experiments. Saccaromyces cerevisiae IDP3 is an NADP-dependent isocitrate dehydrogenase that following the WGD diverged towards peroxisomal localization. Peroxisomes are ubiquitous eukaryotic subcellular compartments where oxidative reactions occur, most notably the degradation of fatty acids via β-oxidation. IDP3 is selectively essential for yeast growth on unsaturated fatty acids as main carbon source, providing the reducing agent NADPH in peroxisome for the β-oxidation of these fatty acids such as petroselinate [34]. S. cerevisiae has three differently compartmentalized IDP paralogues: mitochondrial IDP1, cytosolic IDP2 and peroxisomal IDP3. While the divergence of IDP1 is an ancient event, IDP2 and IDP3 derived from the WGD and share >77% sequence identity. Indeed, in species that diverged prior to the WGD, apart from IDP1, a single IDP copy exists corresponding to cytosolic IDP2. We thus examined the evolutionary mechanisms that underlie the divergence of the ancestral cytosolic IDP2 to give the newly localized peroxisomal paralogue IDP3.
Peroxisomal proteins are transported from the cytosol, most commonly, as with IDP3, via a carboxy-terminal peroxisomal targeting signal. This signal, dubbed PTS1 primarily comprises a tripeptide motif: (S/A/C)-(K/R/H)-(L/M)-*; whereby * represents a stop codon [35]. While additional C-terminal residues affect targeting efficiency, the last 3 C-terminal residues are most crucial for peroxisomal targeting [36]. We first analyzed the IDP gene sequences from S. cerevisiae-related species that diverged before and after the WGD. The pre-WGD species posses only a cytosolic IDP2 gene with no PTS1 signature, whereas the post-WGD species all have a peroxisomal IDP3 paralogue with a C-terminal PTS1 as well as a cytosolic IDP2 gene (Table 1). By this generally-accepted analysis [22], the ancestral IDP2 was presumably localized to the cytosol, and following the WGD, IDP3 neo-localized to the peroxisome by acquiring a PTS1 while IDP2 remained a cytosolic isozyme.
To examine how IDP3’s new, peroxisomal localization diverged from a cytosolic IDP2, we replaced the coding and regulatory regions of S. cerevisiae IDP3 with those of IDP2, and measured the effects on yeast growth in a petroselinate containing medium. A ΔIdp3 strain was constructed from a wild-type strain that spontaneously adapted to growth on petroselinate. The wild-type IDP2 and modified IDP2 with addition of IDP3’s PTS1 at its C-terminus (IDP2+CKL) were cloned into a chromosomal plasmid and transformed into the ΔIdp3 strain. Whilst wild-type IDP2 failed to complement the ΔIdp3 growth on petroselinate, IDP2+CKL gave an IDP3-like growth phenotype (S1A Fig), as previously studied [37]. Like-wise, relative to the PTS1, the divergence of upstream regulatory elements is minor, as indicated by the same petroselinate growth when IDP3’s promoter region was replaced with IDP2’s (S1B Fig). These results suggest that acquisition of the PTS1 motif may have been necessary, and possibly even sufficient, to support divergence of IDP3 from IDP2. It also appears that other changes in IDP3’s open reading frame, and changes in its regulatory regions, were less critical.
We thus focused on unraveling the evolutionary origin of PTS1 motif, that is, when and how IDP3’s peroxisomal signal peptide emerged. By the classical Ohno’s model, the key steps towards divergence occur after duplication, and initially as drift, namely not under adaptive selection [10]. Nonetheless, we searched the C-termini and 3’-UTR sequences immediately after the stop codon of the pre-duplication IDP2s, attempting to identify possible starting sequences from which a PTS1 motif may have evolved via few mutations. We discovered intact, putative PTS1 motifs including an adequate stop codon located shortly after the original stop codon, in the 3’-UTRs. Putative PTS1s were found in 4 out of 5 of Saccaromycetaceae species that are phylogenetically closest to S. cerevisiae but not in more distant species including Candida (so-called CTG fungi group; the bottom clade in Fig 1A). However, unlike previously discovered cryptic PTS1 motifs [30,31,38], these cryptic PTS1 motifs relate not to the enzyme’s coding frame but to a +1 frameshift (Fig 1B). Accordingly, when the cryptic PTS1 was revealed in the coding frame by a single nucleotide deletion upstream to the stop codon, A. gossypii IDP2 (A.gos IDP2) enabled growth of the ΔIdp3 strain on petroselinate (Fig 2A).
Regulated translational frameshifts are known, but they typically occur at long homorepeats such as 8A that are not observed in the coding sequences prior to the stop codons of the pre-duplication IDP2s. Do, then, the cryptic PTS1 motifs within the 3’-UTR regions comprise relics of ancestral PTS1 motifs that were non-functionalized; or do they still encode a functional peroxisomal targeting signal and are thereby maintained under selection for dual localization? The latter seems likely given that sequences that perfectly match a functional PTS1 (Table 1 and Fig 1B) are found in 4 out of 5 of the pre-WGD Saccaromycetaceae species. To investigate the possibility that the pre-duplication IDP2 genes partially produce a peroxisomal isoform carrying a PTS1 via a transcriptional or translation frameshift, we tested whether wild-type A.gos IDP2 can enable petroselinate growth of the S. cerevisiae ΔIdp3 strain. Indeed, growth on petroselinate could be observed with wild-type A.gos IDP2 at about half the rate observed with S. cerevisiae’s original, peroxisomal IDP3, and mutating the cryptic PTS1 within the 3’-UTR abolished the growth (Fig 2A). Thus, a peroxisomal IDP isoform carrying the PTS1 motif seems to be co-expressed alongside the original, cytosolic form.
How are the cryptic PTS1 recruited in the coding region? Homonucleotide repeats show consistently higher tendency for slippage of RNA polymerases, and the ribosome, thus inducing phenotypic frameshift mutations [39,40,41]. Indeed, a 6T repeat exists shortly before the original stop codon of A.gos IDP2 and within a highly diverged segment of the C-terminus (S.cerevisiae Leu403 is the last conserved position in IDP alignments; Fig 1B). Accordingly, silent mutations replacing 2 out of the 6 T within this repeat gave no complementation (Fig 2A).
We further examined whether the phenotypic frameshift at the 6T repeat is due to a transcriptional or translational error. Total RNA from S. cerevisiae expressing A.gos IDP2 and grown on petroselinate was extracted. The cDNA derived from mRNAs of A.gos IDP2 gene was amplified by RT-PCR and cloned for sequencing. In randomly picked 36 clones, all carried a 6U site, corresponding to the original gene’s 6T sequence, and no other sequence changes were detected along 500 bp flanking IDP2’s stop-codon. The phenotypic frameshift is therefore caused by translational errors, consistently with the fact that they are ~10-fold more frequent than transcriptional errors [9]. Overall, these results suggest that the A. gossypii IDP2 partially produces an alternative isoform carrying a PTS1 motif via translational error that bypasses the stop codon and unveils the cryptic, frame-shifted PTS1, and thus exhibiting peroxisomal IDP activity that enables growth on petroselinate.
We further validated the coexistence of two isoforms of A.gos IDP2 by mass spectrometry. Alongside the expected A.gos IDP2 gene product, a higher mass form corresponding to the predicted frame-shifted product at the 6T repeat including the C-terminal AKL was observed with ~30% of the total mass (Fig 2B). Peroxisomal targeting by the cryptic PTS1 was also observed by fluorescent cell imaging. Red fluorescent protein (mCherry) was C-terminally tagged with the C-terminal fragment of A.gos IDP2 containing the cryptic PTS1 in the 3’-UTR. We observed clear punctate co-staining with the peroxisomal marker protein Pex14 fused to GFP when the cryptic PTS1 was revealed by the frameshift (one T deletion in the 6T repeat). On the other hand, wild-type A.gos IDP2 fragment fused to mCherry was primarily visualized in the cytosol, yet with also weak, punctate co-staining with the peroxisomal marker, likely indicating dual localization (Fig 2C).
We subsequently tested the cryptic PTS1 motifs of the other pre-duplication IDP2 genes. The C-terminus of S.cer IDP2 was replaced with the C-termini of 5 different pre-duplication IDP2 genes, including their 3’UTRs (Fig 1B). As expected, K. lactis IDP2 that contains no cryptic PTS1 showed no growth complementation. Three pre-duplication IDP2s containing the cryptic PTS1 appeared to partially express peroxisomal isoforms at a level similar to A.gos IDP2, although only A.gos IDP2 has a >3 bp long homorepeat site (S2A Fig). In contrast, K. waltii IDP2 (K.wal IDP2) failed to show complementation despite the existence of the cryptic PTS1 motif. Nonetheless, growth rate on petroselinate was significantly enhanced when the cryptic motif was revealed by a genetic frameshift (one nucleotide deletion before the stop codon; S2B Fig). We subsequently tested complementation with the full length of K.wal IDP2 open-reading-frame including 150 bp downstream after the stop codon. The full-length K.wal IDP2 showed complementation although with lower growth rates than the other pre-duplication IDP2s. Thus, although K. waltii IDP2 appears to have a functional cryptic PTS1, unveiling it by a phenotypic frameshift seems to be dependent on having the broader context of the K. waltii gene and not just the region around the stop-codon, as was the case with the other pre-duplication IDP2 genes. The frequency of ribosomal slippage may therefore depend on the secondary structure of mRNA, as well as on environmental factors that regulate translational fidelity in the host organism [42,43].
Following the above findings, and a report that appeared while this work was ongoing on cryptic peroxisomal targeting of two cytosolic enzymes [30], we performed a systematic computational search of the 3’-UTR regions of four closely related post-WGD Saccharomyces genomes: S. cerevisiae, S. paradoxus, S. bayanus and S. mikatae (S1 and S3 Tables). The search was based on PTS1 motifs containing all possible variation of amino acids ((S/A/C/E/I/H/Q)-(K/R/H)-(L/F)-stop) from 20 peroxisomal proteins in the Saccaromyces genome database. We searched for the motif starting up to 30 bp downstream the stop codon in all frames. PTS1-like motifs were found in around 1% of total genes of the genomes. However, about 40% of these were interrupted by another stop codon (S1 Table). Further, only a small number of these potentially cryptic motifs were found in more than one species, suggesting that these motifs are under functional selection. We thus focused on few interesting candidates that are conserved among the post-WGD species, and foremost on pyruvate carboxylase 1 (PYC1)—a cytosolic enzyme converting pyruvate to oxaloacetate. The NADPH used for peroxisomal β-oxidation could be produced from pyruvate by a putative pathway that includes four enzymes: PYC: pyruvate carboxylase; CIT: citrate synthase; ACO: aconitase; and finally IDP: isocirate dehydrogenase. Although not established as a peroxisomal NADPH providing pathway, this reaction sequence comprises part of the TCA cycle. Among these, two enzymes have known peroxisomal paralogues in S. cerevisiae: CIT2 and IDP3. The other two, PYC and ACO, are thought to act in the cytosol and mitochondria, respectively, thus demanding the shuttle of their substrates and products to and from the peroxisome (Fig 3A) [44]. We identified, however, a cryptic PTS1-like motif (SHL*) in the PYC1 genes of all four Saccharomyces species. The motif of S. cerevisiae PYC1 is located at 11 bp downstream from the stop codon, in a +1-shifted frame, and predicted as a weak motif by the PTS1 predictor [45] (S3A Fig).
We examined the functionality of the PTS1-like motif of S.cer PYC1 by tagging S.cer IDP2 at the C-terminus with the C-terminal fragment of S.cer PYC1 (the last 11 amino acids and the 3’UTR ending with SHL*). The PYC1 motif showed functional targeting when recruited within the coding frame via a single nucleotide deletion, as indicated by ΔIdp3 complementation for growth on petroselinate, while not functional with the native sequence (Fig 3B). Peroxisomal localization was also observed by fluorescent imaging with mCherry C-terminally-tagged with the 3’UTR motif revealed by a single nucleotide deletion (S3B Fig). These results suggest that this motif is relevant for peroxisomal targeting of S.cer PYC1 via phenotypic errors. At a minimum, our results indicate that a duplicated S.cer PYC1 is within a single genetic mutation from becoming a functional peroxisomal paralog, or perhaps that PYC1 was dually localized in the past.
Our computational search did not identify consensus PTS1 motifs ((S/A/C/E/I/H/Q)-(K/R/H)-(L/F)-stop) in ACO genes. However, upon a closer look we identified a PTS1-like motif (-NKF*) located at +1-shifted frame shortly after the stop codon of S.cer ACO2 (S4A Fig). This motif gave very weak functional targeting when inserted in-frame at the C-terminus of IDP2 and tested for growth on petroselinate (S4B Fig). However, in a continuous passage culture, the slow growth was dramatically accelerated and eventually matched the growth rate of IDP3 (Fig 3C). Sequencing of randomly chosen clones from the petroselinate culture identified a single nucleotide exchange that occurred spontaneously, converting NKF* to NKL* and thus yielding a stronger targeting signal (S4C Fig). The rapid fixation of this mutation demonstrates the ease by which the latent ACO2 motif can further evolve to yield an efficient targeting signal.
How do phenotypic mutations, e.g. the slippage in pre-duplication IDP2s, become eventually ‘imprinted’ via a genetic mutation, thus leading to evolutionary adaptation as observed in the extant, peroxisomal IDP3s? Homonucleotide repeats of 3–8 bases are prone to phenotypic, transcriptional/translational errors as exemplified here with A. gossyppii IDP2 (Fig 2B). However, homorepeats are also highly prone to genetic, frame-shifting InDels (insertions and deletions). In fact, these two phenomena are strongly correlated: the longer the homorepeat, the higher is the frequency of both phenotypic and genetic frameshifts (see Ref. [39] and references therein). However, apart from A. gossyppii IDP2, the 3 other pre-duplication IDP2 genes have no homorepeats of >3 bases length in the region before the stop-codon (Fig 1B). We therefore sought to identify the sites of slippage that unveil the cryptic PTS1 sequences, and to also establish whether the very same sites also comprise hotspots for genetic deletion mutations that result in the exclusive expression of a peroxisomal form.
In fact, we began our exploration with the latter—namely, we sought to identify hotspots for genetic, single base deletions that may occur upstream to K. waltii IDP2’s stop-codon and result in its cryptic PTS1 becoming in-frame (K. waltii IDP2 was the most poorly bypassed pre-duplication IDP2; and, as mentioned above, has no >3 bases repeats in its C-terminus; Fig 1B). We randomly mutated the segment of 100 bases around K. waltii IDP2’s stop-codon, transformed the mutated gene library to the S. cerevisiae ΔIdp3 strain and selected the transformed yeast cells for growth on petroselinate. After 200 hours, the culture’s growth rate dramatically increased (Fig 4A). The selected pool was analyzed by sequencing seven randomly chosen clones. We identified 3 different single base deletions that all occurred within a stretch comprising 3 repeats of 3 bases each just before the stop codon (AAATCCCAAA; Fig 4B). To examine whether the phenotypic frameshifts occur within the very same stretch, we applied the same test applied to validate the 6T repeat as the site of ribosomal slippage in A. gossyppii IDP2. Namely, we introduced silent mutations at each of the 3 deletion sites (AAA TCC CAA A; in bold, the sites of silent mutations; Fig 4B) and examined whether the frequency of slippage, as reflected by the rate of growth on petroselinate, would be reduced. Indeed, silent mutations in the two deletion sites that are closer to the stop-codon showed a marked inhibition of growth, and the triple mutant showed effectively no growth (Fig 4C). It therefore appears that the phenotypic mutations leading to cryptic peroxisomal localization in the cytosolic IDP2s are readily ‘immortalized’ via genetic deletion mutations that occur within the very same site.
Taken together, our results show that S. cerevisiae IDP3 diverged from an ancestral, pre-duplicated gene that, although primarily localized to the cytosol, had the capacity for peroxisomal localization via a phenotypic mutation—a frameshift induced by translational slippage. We appear to be witnessing all the putative intermediates along this evolutionary trajectory. Specifically, S. cerevisiae IDP3, and the other post-duplication IDP3s, all have a ‘legitimate’ in-frame PTS1. The pre-duplication cytosolic IDP2s in Saccaromycetaceae species have a cryptic PTS1 within their 3’-UTR regions that are unveiled by translational and/or transcriptional errors. Further, the contemporary S. cerevisiae PYC1 and ACO2 genes appear to contain cryptic PTS1 signals in their 3’-UTRs that are readily revealed by a single genetic deletion mutation. Indeed, our findings also suggest that four enzymes, that together comprise a putative pathway providing NADPH for peroxisomal β-oxidation, either have a known peroxisomal paralogue (CIT and IDP), or have been partially localized to the peroxisome in the past or are evolving towards peroxisomal localization (PYC and ACO). Finally, our laboratory evolution experiments confirm that the pre-duplication IDP2s carrying cryptic PTS1 sequences readily evolve via genetic mutations to yield ‘legitimate’ peroxisome-targeted genes.
We also conclude that IDP3’s mechanism of divergence does not fit Ohno’s model, namely, neo-functionalization/localization. Rather, IDP3’s peroxisomal targeting emerged in the ancestral Saccaromycetaceae species long before the IDP gene was duplicated to give the newly diverged IDP3. As shown here, 4 out of the 5 pre-duplication IDP2s in Saccaromycetaceae species are dually localized via a cryptic PTS1 whilst the more distant species do not possess such cryptic PTS1s (Fig 1). Thus, IDPs represent a clear case of divergence via ‘gene sharing’ [11,21] and of ‘divergence before duplication’ [13,14]. Following duplication, the ancestral dual, cytosol-peroxisome localization function was split between two paralogous genes, thus representing a case of sub-functionalization/localization [22,24]. The latter evolutionary mechanisms rely on one gene executing multiple functions, and accordingly on weak trade-off—namely, that mutations that endow the newly emerging function do not abolish the original function. Weak trade-offs, i.e., ‘something for nothing’, at least at the early stages of evolution, is a key feature that makes divergence before the duplication a far more plausible scenario than Ohno’s model [46]. Indeed, the assumption of tradeoffs underlies Ohno’s model—the existence of a redundant copy relieved from the burden of selection enables mutations to freely accumulate, including mutations that undermine the original function [47]. Our analysis indicated a ratio of ~1:3 of the peroxisome-cytosol isoforms in A. gossyppii IDP2 (Fig 2B). Thus, a reduction of ~25% in the levels of the cytosolic IDP is enough to shift from no growth on petrosalinate to a growth rate that is only half of that observed with the ‘legitimate’ peroxisomal IDP3 (Fig 2A). However, growth levels comparable to wild-type were afforded only upon a genetic mutation that leads to exclusive peroxisomal targeting. Assuming that cytosolic IDP is essential, such a mutation could only follow duplication.
IDPs’ divergence prior to duplication was driven by transcriptional /translational errors that result in dual localization to both the cytosol and peroxisome from a single gene. Dual localization is commonly observed (see also refs [27,28,48,49]. Specifically, stop-codon read-through, or alternative splicing, were previously shown to mediate the dual cytosol-peroxisome targeting of several glycolytic enzymes in various yeast species [30]. However, there is no evidence indicating that these genes duplicated and diverged into ‘legitimate’ peroxisomal paralogues, as is the case with the pre-duplication IDP2s. Divergence before duplication may apply to localization signals other than PTS1. Most protein localizations, such as to the endoplasmic reticulum, mitochondria, or chloroplasts, are mediated by N-terminal target signals that are ~20 amino acids long. Translational errors can also produce various N-terminal isoforms from a single mRNA owing to alternative translation initiation sites (“leaky scanning”) [50] thus enabling dual targeting [27,29,51].
Foremost, our results provide unequivocal evidence that phenotypic mutations led to the evolution of new traits [43,52]. Noise and infidelity in general, and transcriptional and translational errors specifically, may comprise a “look-ahead” effect [6] thus underlining the phenotype of the yet to emerge duplicated, diverged gene [5]. The rate of phenotypic mutations is >105-fold higher than genetic mutations, and may be further enhanced under stress due to the malfunction of translational fidelity [42], or under the yeast prionic state ([PSI+]), thus promoting phenotypic diversity that mediates survival in challenging environments [53]. Further, although phenotypic mutations are not inherited as such, the capacity to induce them is inherited via DNA sequences that favor slippage, as manifested, for example, in the 6T repeat inducing dual localization of A. gossyppii IDP2.
Finally, our results indicate another intriguing aspect of phenotypic mutations—the same gene context is prone to both phenotypic mutations (transcriptional/translational errors) and genetic mutations (Fig 4). Thus, selection for dual localization, hence favoring gene sequences whereby slippage downstream the stop-codon occurs at relatively high frequency (as seems to be the case in the pre-duplication IDP2 genes) also creates a hotspot for a genetic mutation. In this manner, a coincidental error becomes a ‘frozen accident’ under selection, as well as a hotspot for evolutionary adaptation.
All strains were derived from By4741 (MATa, his3Δ1, leu2Δ0, met15Δ0, ura3Δ0) [54]. The strains used are listed in S2 Table. To obtain the wild-type strain harboring a selection marker (WT; idp3:: IDP3-kanMX4), the IDP3 open reading frame (ORF) of the By4741 genome was replaced with the IDP3 gene fusing to kanMX4 cassette by homologous recombination [55]. This strain (WT) was subjected to serial passage culture in the petroselinate containing medium (1% yeast extract, 2% Bacto-peptone, 0.2% Tween-40, 0.1% petroselinate) containing G418 (200 μg/ml)) until spontaneously adapting and exhibiting higher growth rate. The IDP3 knockout strain (ΔIdp3; idp3:: hphNT1) was constructed from this adapted strain, whereby the locus containing the IDP3 gene—the kanMX4 cassette was replaced with a hpnNT1 cassette by homologous recombination.
The kanMX4 cassette was PCR amplified from a pBS7 vector (Yeast resource center). The IDP3 gene—kanMX4 fusion was constructed as follows: The IDP3 gene, including the ORF and 150 bps downstream, was PCR amplified from genomic DNA. The amplified gene was introduced into the pBS7 vector containing kanMX4 cassette by using the SmaI /BglII sites. The IDP3 gene—the kanMX4 cassette was thus PCR amplified from the sub-cloned pBS7 vector. The hpnNT1 cassette was amplified from pRS41H plasmid [56] using primers franking the 5’ end of IDP3’s ORF and the 3’ end of the kanMX4 cassette. DNA fragments encompassing the coding region, plus the 500 bps upstream (5’) and 150 bps downstream regions (3’) of the various IDP genes (S. cerevisiae IDP3, IDP2, A. gossypii IDP2, and K. waltii IDP2) were amplified from genomic DNA. For testing the effects of the coding vs. 5’-UTR (promoter) and 3’-UTR (cryptic PTS1s), the coding and non-coding fragments were separately amplified and combined by assembly PCR. These assembled fragments were subcloned by SmaI/NotI sites into pRS41K [56], a centromere-based plasmid for single-copy expression in yeast, generating: pRS41K-ScIDP3Pro/ScIDP3/ScIDP3Ter, pRS41K-ScIDP2Pro/ScIDP3/ScIDP3Ter, pRS41K-ScIDP3Pro/ScIDP2+CKL/ScIDP3Ter, pRS41K-ScIDP3Pro/ScIDP2/ScIDP2Ter, pRS41K-ScIDP3Pro/AgIDP2/AgIDP2Ter, pRS41K-ScIDP3Pro/KwIDP2/KwIDP2Ter; whereby ScIDP3pro and ScIDP2pro are 500 bp upstream regions of S. cerevisiae IDP3 and IDP2, respectively; ScIDP2, ScIDP2+CKL, ScIDP3, AgIDP2, and KwIDP2 are coding region; ScIDP2ter, ScIDP3ter, AgIDP2ter, and KwIDP2ter are 150 bp downstream regions of S. cerevisiae IDP3, IDP2, A. gossypii IDP2, and K. waltii IDP2 respectively: e.g. ScIDP2Pro/ScIDP3/ScIDP3Ter represents the assembled fragment of the upstream region of ScIDP2, ScIDP3 coding region, and the downstream region of ScIDP3. To introduce mutations (shown with underbars in primers sequences below) in the PTS1, or in the polyT region of A.gossypii IDP2, site-directed mutagenesis was performed by using pRS41K-ScIDP3Pro/AgIDP2/AgIDP2Ter plasmid as a template with primer sets ptsdel_f (5’-GAAAAAGCAAGCATAATTATAGCCTAGGCTGCCT-3’) and ptsdel_r (5’-AGGCAGCCTAGGCTATAATTATGCTTGCTTTTTC-3’) for cPTS1 mutation, delt_f (5’-GGCTACAAGCGTCTTTTTGTGAATAAGAAAAAGC-3’) and delt_r (5’-GCTTTTTCTTATTCACAAAAAGACGCTTGTAGCC-3’) for single T deletion on the polyT, and tsyn_f (5’-GGCTACAAGCGTCTCTTCTGTGAATAAGAAAAAG-3’) and tsyn_r (5’-CTTTTTCTTATTCACAGAAGAGACGCTTGTAGCC-3’) for silent mutations on the polyT, generating pRS41K-ScIDP3Pro/AgIDP2ΔAKL/AgIDP2Ter, pRS41K-ScIDP3Pro/AgIDP2Δt/AgIDP2Ter, and pRS41K-ScIDP3Pro/AgIDP2+silent /AgIDP2Ter, respectively. Replacement of the C-terminus of S.cer IDP2 with the C-termini and 3’-UTRs of various genes was performed using Leu403 as the 5’ crossover point (as it comprises the last conserved residue in yeast IDPs), and the stop codon as the 3’ crossover.
The peroxisome targeting potential of cryptic PST1 sequences was tested by measuring the growth rates of ΔIdp3 strain complemented with various IDP genes in the petroselinate containing medium. The IDP constructs tested for targeting (native and chimeras alike) were cloned into a chromosomal pRS41K plasmid and transformed to the ΔIdp3 strain. Cells were first grown on YPD media (1% yeast extract, 2% Bacto-peptone, 2% Glucose) for at least 18 hours, and then used to inoculate into the YP-petroselinate medium (1% yeast extract, 2% Bacto-peptone, 0.2% Tween-40, 0.1% petroselinate) at an initial OD600 0.1. Growth was monitored by absorbance at 600 nm (error bars represent standard deviations of three independent cultures).
The 100 bp region centered around the stop codon of K. waltii IDP2 was randomly mutated by error-prone PCR using GeneMorph II random mutagenesis kit (Agilent technologies, CA) and integrated by MEGAWHOP cloning [57] into the IDP2 encoding plasmid pRS41K-ScIDP3Pro/KwIDP2/KwIDP2Ter (construction details above). Sequencing indicated an average mutation rate of ~1 mutation per gene. The plasmid library was transformed to the ΔIdp3 strain. Cells were cultured for 300 hours in the YP-petroselinate medium as described above. The resulting culture with increased growth rates was plated on YPD plates and randomly chosen colonies were analyzed by DNA sequencing. Silent mutations were introduced in the plasmid pRS41K-ScIDP3Pro/KwIDP2/KwIDP2Ter by site-direct mutagenesis. These plasmids were used to transform the ΔIdp3 strain and test growths on petroselinate. Note that the ΔIdp3 strain used here was derived from another adaptation experiment where the ΔIdp3 strain complemented with K. waltii IDP2 gene (pRS41K-ScIDP3Pro/KwIDP2/KwIDP2Ter) was cultured in the YP-petroselinate medium until spontaneously adapting.
IDP sequences were obtained from the Fungal Orthogroups Repository [58]. Sequence alignment was created by MUSCLE [59]. Maximum likelihood phylogenic trees were created by PhyML [60] based on the yeast species tree [58] by using the JTT substitution matrix.
The ΔIdp3 strain transformed by plasmid pRS41K-ScIDP3Pro/AgIDP2/AgIDP2Ter was cultured in the YP media containing petroselinate until the mid-log phase (OD600 ~0.6). One mL culture was centrifuged and the collected cell pellet was subjected to total RNA extraction using total RNA extraction kit (Epicentre). The cDNA of A. gossypii IDP2 was amplified from the total RNA by RT-PCR using gene-specific primers: xhoI_agidp500f (5’-ATTGGGTACCCTCGAGAGGACGGGGACAAGTCCAAG-3’) and agter_notIr (5’-CACCGCGGTGGCGGCCGCAGATATGCTAGACTAGTAATAAATAGACGC-3’). The amplified PCR products were subcloned into plasmid pRS41K using XhoI/NotI. The plasmids were transformed into E. coli. and 36 randomly selected colonies were subjected to DNA sequencing.
For expression in S.cerevisiae and purification, A.gossypii IDP2’s ORF plus 150 bps of the 3’-UTR region was amplified by PCR (from plasmid pRS41K-ScIDP3Pro/AgIDP2/AgIDP2Ter) with primers encoding an N-terminal Histidine-tag. The DNA fragment was subcloned into plasmid pFA6a-nat [61] using XhoI/SpeI sites, for expression and the strong constitutive TEF2 promoter and ADH1’s terminator. The resulting construct including the promoter and terminator was excised using SacI sites and subcloned into plasmid pRS42H [56], a multicopy 2μ-based yeast plasmid, generating pRS42H-His:AgIDP2/AgIDP2Ter. This plasmid was transformed into the adaptive wild-type strain (WT). For purification of histidine-tagged protein, transformants were pre-cultivated in YPD for 18 hours at 30°C, transferred to YPD at a starting OD 0.05. The culture was harvested at OD 2.2 after 22 hours incubation at 30°C. Harvested cells were resuspended in the two-fold cell volumes of lysis buffer (50 mM potassium phosphate (pH 8.0), 300 mM sodium chloride, 2 mM sodium citrate, 10 mM Imidazole, 10% glycerol, 1 mM DTT, 0.1% Triton-100, 1% protease inhibitor cocktail (Sigma)) and lysed by voltex with the same cell volume of glass beads (425–600 nm, Sigma; G8772) and subsequent sonication. Cell debris was removed by centrifugation for 30 min at 11,500 rpm, and the supernatant was passed through a open column containing Ni-NTA resin. The column was washed with 30-fold resin volumes of wash buffer (50 mM potassium phosphate (pH 8.0), 300 mM sodium chloride, 2 mM sodium citrate, 20 mM Imidazole, 10% glycerol). The proteins were finally eluted with elution buffer (50 mM potassium phosphate (pH 8.0), 300 mM sodium chloride, 2 mM sodium citrate, 250 mM Imidazole, 10% glycerol). The eluted samples were buffer-exchanged with PBS (137 mM sodium chloride, 2.7 mM potassium chloride, 1.76 mM potassium dihydrogenphosphate, 10 mM disodium hydrogenphosphate, pH 7.4) containing 10% glycerol and 2 mM sodium citrate and concentrated by ultrafiltration (Vivaspin 500-10K, GE). Typical yield was ~ 0.2 mg/L at >90% purity as judged by SDP-PAGE.
Microcapillary reverse phase liquid chromatography (LC) was performed with a nanoAcquity UPLC system (nUPLC) (Waters Corp.), using the HEMA/EDMA (Hexyl methacrylate/Ethylene glycol dimethacrylate 60/40 v/v) monolithic column prepared in-house, as previously described [62]. Proteins (5μl of 50 ng/mL) were loaded onto the column and separated using a linear gradient of 20% to 60% solvent B over 40 min, at a flow rate of 10–15 μl/min, at 60°C. Solvent A was water + 0.05% formic acid+ 0.035% Trifluoroacetic acid and solvent B was acetonitrile + 0.05% formic acid+ 0.035% Trifluoroacetic acid. The LC eluant was sprayed into a Qstar XL mass spectrometer (MDS Sciex, Canada) by means of an electrospray ion (ESI) source. The following experimental parameters were used: capillary 5.3 kV, declustering potential of 40 V, focusing potential of 200 V, and second declustering potential of 20 V. The covered mass range was 500–5,000 m/z. Minimal smoothing and centering parameters were used. Spectra were calibrated using a solution of Reserpine (1 μM). The experimental masses of both the original and alternative isoforms corresponded to the theoretical values minus 130 Da due to the removal of the initial methionine [63].
Peroxisomal localization was confirmed by mating strains expressing Pex14p fused to GFP and the C-terminal proteins of interest fused to mCherry. The reference haploid strain (By4742: MATα, his3Δ1, leu2Δ0, met15Δ0, ura3Δ0) expressing Pex14p C-terminally tagged with GFP was constructed as described before [64]. The C-terminal fragments from A.gos IDP2 or S.cer PYC1 were inserted to the C-terminus of mCherry in plasmid pbs69_PRX_tdh3_mCherry, derived from pBS35 (yeast resource center), by whole plasmid PCR. The plasmid was digested in the Mfe1 site and integrated into the TDH3 promoter site of the adapted wild-type strain (WT) by homologous recombination. The transformed haploid WT strain was selected in the presence of Hygromycin (300 μg/ml), and analysed for positive RFP signal by fluorescent microscopy. The RFP-tagged strains were then mated with the GFP-tagged reference strain in SD medium lacking Histidine containing Hygromycin, and the resulting diploid strains were visualized by fluorescent microscopy (Nikon Ecripse Ti, Japan). Images were taken using a cooled CCD camera with an exposure time of 40–300 ms and processed using ImageJ (National Institutes of Health).
The motif sets used for PTS1 search were built from 20 known peroxisomal genes (13 genes containing the canonical PTS1, 7 genes either containing resembled PTS1 or C-terminus responsible for its localization) from Saccaromyces Genome Database. Each position in the motif was defined as the union of amino acids that appeared in these genes ((S/A/C/E/I/H/Q)-(K/R/H)-(L/F)-stop), in order to minimize the possibility for false negative. Searched were the motifs starting within the first 30 bps of the 3’ UTR, excluding the one disturbed by an additional stop codon. This search was performed in four genomes (downloaded from the Saccaromyces Genome Database): S. cerevisiae (S288C reference genome version R64), S. pardoxus (strain NRRL Y-17217), S. bayanus (WashU version) and S. mikatae (strain IFO1815). For each cryptic PTS1 candidate, 12 amino acids upstream to the PTS1 (including it) was extracted and manually scored by using the PTS1 predictor (http://mendel.imp.ac.at/pts1/ PTS1predictor.jsp) [45]. The cryptic PTS1 appeared in the coding frame were scored by replacing the original stop codons UAA/UAG and UGA with glutamine and arginine, respectively, according to the known stop codon read-through [65,66].
|
10.1371/journal.ppat.1003696 | Activation of the Cph1-Dependent MAP Kinase Signaling Pathway Induces White-Opaque Switching in Candida albicans | Depending on the environmental conditions, the pathogenic yeast Candida albicans can undergo different developmental programs, which are controlled by dedicated transcription factors and upstream signaling pathways. C. albicans strains that are homozygous at the mating type locus can switch from the normal yeast form (white) to an elongated cell type (opaque), which is the mating-competent form of this fungus. Both white and opaque cells use the Ste11-Hst7-Cek1/Cek2 MAP kinase signaling pathway to react to the presence of mating pheromone. However, while opaque cells employ the transcription factor Cph1 to induce the mating response, white cells recruit a different downstream transcription factor, Tec1, to promote the formation of a biofilm that facilitates mating of opaque cells in the population. The switch from the white to the opaque cell form is itself induced by environmental signals that result in the upregulation of the transcription factor Wor1, the master regulator of white-opaque switching. To get insight into the upstream signaling pathways controlling the switch, we expressed all C. albicans protein kinases from a tetracycline-inducible promoter in a switching-competent strain. Screening of this library of strains showed that a hyperactive form of Ste11 lacking its N-terminal domain (Ste11ΔN467) efficiently stimulated white cells to switch to the opaque phase, a behavior that did not occur in response to pheromone. Ste11ΔN467-induced switching specifically required the downstream MAP kinase Cek1 and its target transcription factor Cph1, but not Cek2 and Tec1, and forced expression of Cph1 also promoted white-opaque switching in a Wor1-dependent manner. Therefore, depending on the activation mechanism, components of the pheromone-responsive MAP kinase pathway can be reconnected to stimulate an alternative developmental program, switching of white cells to the mating-competent opaque phase.
| The pathogenic yeast Candida albicans can switch from the white yeast form to the mating-competent opaque form. Opaque cells are less virulent than white cells, but they can avoid recognition by phagocytes, indicating that white-opaque switching has evolved as an adaptation mechanism of C. albicans to specific host niches. Both white and opaque cells respond to mating pheromone by activating the Ste11-Hst7-Cek1/Cek2 MAP kinase pathway, but with different outcomes. Opaque cells utilize the transcription factor Cph1 to induce the mating response, whereas white cells recruit a different downstream transcription factor, Tec1, to promote biofilm formation. We used a comprehensive protein kinase expression library to gain insight into the signaling pathways that regulate switching from the white to the opaque phase and found that a hyperactive form of the upstream kinase Ste11 induced white opaque-switching, a behavior that did not occur in response to pheromone. Hyperactive Ste11 functions via the opaque-specific transcription factor Cph1 instead of the white-specific transcription factor Tec1 to promote this alternative developmental program. Therefore, depending on the activation mechanism, components of the pheromone-responsive MAP kinase pathway can be rewired to stimulate a transition from the more virulent white form to the less aggressive, but mating-competent opaque form.
| The yeast Candida albicans is a member of the microbiota in the gastrointestinal and genitourinary tracts of most healthy persons, but it can also cause superficial as well as life-threatening systemic infections when host defenses are compromised. Morphological transitions play a major role in the biology of C. albicans and in the interactions of the fungus with its host. For example, in response to various environmental stimuli, C. albicans alters its morphology from a unicellular budding yeast to a multicellular hyphal form. The switch from yeast to filamentous growth facilitates tissue invasion and is associated with the transition of C. albicans from a harmless colonizer to a pathogen that causes symptomatic infections [1].
C. albicans can also switch from the normal, round-to-oval yeast morphology (white) to an elongated yeast cell type (opaque), which is the mating-competent form of this diploid fungus [2]. Opaque cells can mate with opaque cells of opposite mating type to generate tetraploid fusion products, which may then undergo random chromosome loss to generate recombinant progeny in a parasexual cycle [3]. Switching of white cells to the opaque phase requires the transcription factor Wor1, the master regulator of white-opaque switching [4], [5], [6]. WOR1 is expressed at very low levels in white cells, but an increase in the amount of Wor1 above a threshold induces switching to the opaque phase. Wor1 activates its own expression, resulting in a positive feedback loop that provides the high Wor1 levels required for maintenance of the opaque phenotype. Additional transcription factors, including the positive regulators Wor2 and Czf1 and the negative regulator Efg1, which are themselves controlled by Wor1, ensure bistable expression of WOR1 (low in white and high in opaque cells) and epigenetic inheritance of the two phases [7]. Switching to the mating-competent opaque cell type is restricted to strains that are homozygous at the mating type locus (MTLa/a or MTLα/α). In heterozygous MTLa/α strains, switching of white cells to the opaque phase is inhibited by a heterodimeric repressor consisting of the homeodomain proteins a1 (encoded by MTLa) and α2 (encoded by MTLα), which prevents WOR1 expression [2], [4], [5], [6]. Most C. albicans strains in nature are MTL heterozygotes, but they can become homozygous by genomic rearrangements, which relieves them from repression by the a1-α2 repressor and renders the cells switching-competent [8].
White and opaque cells differ not only in their mating capacity, but also in many additional phenotypes and in the expression of genes that are unrelated to mating, suggesting that they are adapted to different environments within a mammalian host [9], [10], [11]. Opaque cells colonize skin more readily than do white cells, but they are much less virulent than white cells during a systemic infection [12], [13]. Although opaque cells can form hyphae, they do not undergo the yeast-hypha transition under many conditions that stimulate hyphae formation in white cells [14], [15]. This may result in a decreased capacity of opaque cells to escape from the bloodstream and invade into tissues. On the other hand, opaque cells can avoid recognition and phagocytosis by macrophages and neutrophils under conditions in which white cells are efficiently phagocytosed [16], [17], [18]. Therefore, switching to the opaque phase not only results in the acquisition of mating competence but may also allow escape from the native immune system in certain host niches in which the more aggressive white cells will be attacked, especially after hyphae formation [19], [20].
Interestingly, both white and opaque cells respond to the presence of mating pheromone. α-pheromone, which is produced by opaque α-cells, binds to the receptor Ste2 on a-cells, and a-pheromone, produced by opaque a-cells, binds to the receptor Ste3 on α-cells. But while opaque cells induce a mating response that results in shmoo formation and cell fusion, pheromone binding to white cells induces the production of a biofilm, which stabilizes the pheromone gradient and facilitates mating of opaque cells [21]. This differential response of white and opaque cells is achieved by the pheromone-induced activation of a common MAP kinase cascade, consisting of the MAPKKK Ste11, the MAPKK Hst7, and the partially redundant MAPKs Cek1 and Cek2, and the cell type-specific recruitment of different downstream transcription factors, Cph1 in opaque cells and Tec1 in white cells [22], [23].
It was believed for a long time that white-to-opaque switching is a stochastic process that occurs spontaneously in few cells of a population. However, it has recently become evident that white cells can be induced under certain environmental conditions to switch en masse to the opaque phase [24], [25], [26], [27], [28]. Although several transcription factors have been identified that regulate white-opaque switching [4], [5], [6], [7], [28], [29], [30], little is known about the upstream signal transduction pathways that allow white cells to respond to these inducing signals. Protein kinases are common components of signaling pathways that mediate cellular reactions to external and internal signals, and the protein kinase Tpk2 has recently been implicated in the induction of switching by environmental signals [26]. In the present study, we generated a comprehensive, tetracycline-inducible protein kinase expression library to investigate whether additional kinases are involved in the control of white-opaque switching. Intriguingly, we found that the pheromone-responsive MAP kinase pathway, which promotes the mating response in opaque cells and biofilm formation in white cells, can be rewired such that white cells recruit the opaque-specific transcription factor Cph1 instead of Tec1, which then induces switching to the opaque phase.
Artificial expression of the master regulator WOR1, and also of the positive regulator CZF1, from a constitutive or inducible promoter induces switching of white cells to the opaque phase [4], [5], [6], [7], [28], [30]. We reasoned that forced expression of protein kinases that act upstream of the transcriptional regulators to promote switching in response to environmental signals may similarly induce white cells to switch to the opaque phase. Therefore, we cloned all C. albicans genes encoding known or putative protein kinases and their regulators in a tetracycline-inducible gene expression cassette (see Materials and Methods). We also included 21 putative hyperactive or dominant negative alleles of these kinases in the collection. The resulting library of 160 Tet-inducible protein kinases and regulators (supplemental Table S1) was then integrated into the genome of the C. albicans MTLα/α strain WO-1, in which white-opaque switching was originally discovered and which has been widely used as a model strain to study this developmental program. In each case, two independent transformants were kept to confirm the reproducibility of phenotypes that were induced by the expression of a kinase.
To discover kinases whose forced expression from the Tet promoter induces switching to the opaque phase, white cells of the parental strain WO-1 and the strains containing the inducible protein kinase library were grown overnight in liquid medium in the presence or absence of doxycycline and then spread at an appropriate dilution on agar plates with or without doxycycline to allow the formation of colonies from individual cells. The results of this screening are summarized in supplemental Table S2. As expected, all strains behaved like the parental strain WO-1 when they were grown in the absence of doxycycline and showed only basal levels of spontaneous switching. In contrast, forced expression of the protein kinases MPS1, RAD53, TPK1, TPK2, and the hyperactive STE11ΔN467 allele resulted in a strongly increased frequency of switching from the white to the opaque phase when the inducer doxycycline was present in the preculture and/or during subsequent colony growth on the agar plates. Ste11 is the MAPKKK of the pheromone-responsive MAP kinase signaling pathway (but also functions in other pathways), and in the present study we focus on the role of this pathway in the regulation of white-opaque switching in C. albicans. The other identified kinases will be subject of future investigations.
The pheromone-responsive signaling pathway contains two partially redundant MAP kinases, Cek1 and Cek2, which activate the downstream transcription factors Cph1 and Tec1 [22], [23], [31]. Of note, the functions of the two transcription factors in the pheromone response differ. While Cph1 induces the expression of mating-specific genes in opaque cells and is required for mating, Tec1 has no role in mating of opaque cells, but promotes biofilm formation of white cells in response to pheromone produced by opaque cells. This response of white cells is thought to stabilize the pheromone gradient and facilitate mating of opaque cells in a population that contains a majority of white cells [21]. We therefore investigated if these downstream MAP kinases and transcription factors are also involved in the induction of white-opaque switching by the activated Ste11ΔN467. To this aim, we generated deletion mutants of strain WO-1 lacking CEK1, CEK2, CPH1, or TEC1 and expressed the hyperactive STE11ΔN467 allele from the Tet promoter in the various mutants, all of which were constructed twice independently (see supplemental Table S3). As can be seen in Fig. 1, inactivation of CEK1 abolished STE11ΔN467-induced white-opaque switching, while deletion of CEK2 had no effect. The switching defect of the cek1Δ mutants was complemented by reintroduction of a functional CEK1 copy. These results indicate that the two MAP kinases, which have redundant functions in the mating pheromone response of opaque cells and in pheromone-induced biofilm formation of white cells [23], [31], also have divergent roles: CEK1 is required for the Ste11-induced switching of white cells to the mating-competent opaque form, whereas CEK2 is dispensable for this developmental program.
Deletion of TEC1 did not affect the ability of the hyperactive Ste11 to induce white-opaque switching. In contrast, Ste11-induced switching was abolished in the cph1Δ mutants, and this defect was complemented by reintroduction of a functional CPH1 copy (Fig. 1). Therefore, Cph1 not only functions in the mating response of opaque cells but is also required for Ste11-induced switching of white cells to the opaque phase. In contrast, Tec1 induces biofilm formation in white cells in response to pheromone [22], but is not required for the induction of white-opaque switching. We hypothesized that, if Cph1 is the downstream transcription factor that mediates Ste11-induced white-opaque switching, forced expression of CPH1 might also promote switching of white cells to the opaque phase. Indeed, expression of CPH1 from the Tet promoter in strain WO-1 strongly induced white-opaque switching (Fig. 2). Doxycycline-induced expression of TEC1 also caused an increase in the frequency of white-opaque switching, but this was not comparable to the stimulation by CPH1, and TEC1 was also not required for Cph1-induced white-opaque switching (Fig. 2). Together, these results demonstrate that activation of the pheromone-responsive MAP kinase cascade in white cells by a hyperactive form of the MAPKKK Ste11 induces one arm of this signaling pathway, including the MAPK Cek1 and the transcription factor Cph1, to promote switching to the opaque phase.
White-opaque switching is controlled by a network of feedback loops comprising the positive regulators Wor1, Wor2, and Czf1 and the negative regulator Efg1 [7]. To investigate if the core positive regulators were required for Ste11/Cph1-induced white-opaque switching, we expressed CPH1 and the hyperactive STE11ΔN467 allele in mutants of strain WO-1 from which WOR1, WOR2, or CZF1 were deleted. Initial experiments showed that the colony phenotypes of the mutants were altered in some cases, making it difficult to decide whether cells had truly switched to the opaque phase (see Fig. 3B, top panels). We therefore generated derivatives of strain WO-1 and of the mutants that expressed GFP or RFP from the opaque-specific OP4 promoter to determine whether opaque-like colonies generated after Tet-induced expression of CPH1 or STE11ΔN467 contained bona fide opaque cells (see Fig. 3B, bottom panels). CPH1 and STE11ΔN467 induced white-opaque switching in the labeled wild-type reporter strains with the same efficiency as in the original parental strain WO-1 (Fig. 3A, compare with Fig. 1 and 2). No switching to the opaque phase was observed in wor1Δ and wor2Δ mutants, demonstrating that the master regulator Wor1 as well as Wor2 were also required for Cph1-induced switching. However, doxycycline-induced CPH1 expression promoted switching to the opaque phase also in the absence of CZF1, while forced expression of STE11ΔN467 failed to increase the switching frequency above background levels (Fig. 3A). Apparently, the overexpression of CPH1 was strong enough to promote the switch in the absence of Czf1, which is part of a positive feedback loop that facilitates switching. In contrast, the comparatively weaker induction by the hyperactive STE11ΔN467 allele was not sufficient to overcome the threshold for the switch to occur in the absence of Czf1. These results demonstrate that, when sufficiently active, Cph1 does not depend on Czf1 to promote white-opaque switching, but it still requires the master regulator Wor1.
Tet-induced expression of TEC1 or STE11, but not CPH1, in white cells of a C. albicans MTLa/a strain has been shown to promote biofilm formation under normally noninducing conditions [22]. In line with these results, we found that Tet-induced expression of the hyperactive STE11ΔN467 allele in the MTLα/α strain WO-1 also induced biofilm formation in addition to white-opaque switching (Fig. 4). To investigate which downstream MAP kinases and transcription factors were required for STE11ΔN467-induced biofilm growth, we assayed biofilm formation in cek1Δ, cek2Δ, cph1Δ, and tec1Δ mutants expressing the hyperactive STE11 allele. As expected, STE11ΔN467-induced biofilm formation was abolished in tec1Δ mutants. Surprisingly, however, the hyperactive Ste11 was also unable to promote biofilm formation in mutants lacking CPH1. Similarly, no induction of biofilm formation was seen in the absence of CEK1, while deletion of CEK2 had no effect in these assays, demonstrating that the two MAP kinases are not redundant for Ste11ΔN467-induced biofilm development in strain WO-1. Reintroduction of an intact copy of CPH1 and CEK1 into the respective mutants restored STE11ΔN467-induced biofilm growth, confirming that the mutant phenotype was caused by the deletion of these genes (Fig. 4).
As both CPH1 and TEC1 were required for the induction of biofilm formation by the hyperactive STE11ΔN467 allele, we tested whether forced expression of either of these transcription factors would promote biofilm formation in the absence of the other in strain WO-1. Tet-induced CPH1 expression indeed caused biofilm formation in this strain background, but this induction depended on the presence of TEC1 (Fig. 5). In contrast, Tet-induced expression of TEC1 promoted biofilm formation both in the presence and absence of CPH1.
The results presented above demonstrate that Tet-induced expression of the hyperactive STE11ΔN467 allele or the downstream transcription factor CPH1 promoted both biofilm formation and white-opaque switching in strain WO-1. Biofilm formation apparently was not a prerequisite for the induction of white-opaque switching, because switching was efficiently induced also in tec1Δ mutants, which did not form biofilms. Since biofilm formation has often been linked to hyphal morphogenesis, we examined the phenotype of the cells after Tet-induced STE11ΔN467 or CPH1 expression before plating for subsequent colony formation. For comparison, we also tested cells that expressed the known positive regulator CZF1 from the Tet promoter. These latter cells did not yet exhibit the opaque morphology after induction of CZF1 expression in liquid culture (Fig. 6A, bottom left panel), but were programmed to switch to the opaque phase, as almost all of them formed opaque colonies after plating (see supplemental Table S2). Wild-type cells expressing STE11ΔN467 or CPH1 mainly grew as filaments, similar to cells expressing TEC1, explaining the formation of biofilms on the plastic surface (Fig. 6A, middle panels). In the absence of TEC1, filamentation of cells expressing STE11ΔN467 was strongly reduced and most cells exhibited the normal white yeast morphology (Fig. 6A, top right panel). In contrast, Tet-induced expression of CPH1 in a tec1Δ background resulted in switching to the opaque phase already during growth in liquid medium, indicating that when filamentation is blocked by the absence of Tec1, upregulation of CPH1 expression in a switching-competent strain directly promotes opaque cell formation (Fig. 6A, bottom right panel). Strikingly, many of the opaque cells formed shmoos, in accord with the previously reported finding that CPH1 overexpression in opaque cells induces the mating response [32].
In order to understand why the filamentous cells observed after Tet-induced STE11ΔN467 or CPH1 expression were programmed to switch to the opaque phase after plating, we determined the expression levels of the master regulator WOR1 in these filamentous cells by RT-qPCR. Despite some variation between biological replicates, WOR1 transcript levels were consistently increased upon STE11ΔN467 expression in all experiments with the two independently constructed strains, on average ca. 15-fold above those in the parental control strain (Fig. 6B, top left panel). CPH1 transcript levels were also elevated, whereas TEC1 mRNA levels remained unchanged (Fig. 6B, top middle and right panels). The latter result is in agreement with the fact that Cph1, but not Tec1, is the downstream transcription factor that promotes STE11ΔN467-induced white-opaque switching. An even stronger upregulation of WOR1 (>800-fold) was observed upon Tet-induced expression of CPH1 itself (Fig. 6B, bottom left panel), which is explained by the significantly higher CPH1 transcript levels in these cells compared to those seen after expression of STE11ΔN467 from the Tet promoter (Fig. 6B, middle panels). These data demonstrate that the activation of the MAP kinase cascade in white cells by the hyperactive STE11ΔN467 has a different outcome compared to the induction by pheromone. Instead of causing TEC1 upregulation, Ste11 lacking its N-terminal inhibitory domain increases CPH1 expression, which in turn induces WOR1 expression, thereby programming the cells to switch to the opaque phase.
White-opaque switching can be induced by different environmental signals, but there are strain-specific differences in the response of switching-competent strains to the various stimuli [24], [25], [26], [28]. In strain WO-1, switching is strongly induced by a transient incubation in an anaerobic environment (0% O2, 18% CO2), while other tested strains did not switch to the opaque phase under these conditions [28]. We therefore investigated if Cph1-induced white-opaque switching might be a peculiar characteristic of strain WO-1 or if Cph1 can promote switching also in other C. albicans strains and in both mating types. For this purpose, we deleted either the MTLa or the MTLα locus in the commonly used reference strain SC5314 to generate switching-competent α and a derivatives, respectively, into which the Ptet-CPH1 fusion was subsequently introduced. Doxycycline did not induce white-opaque switching in these strains; however, using a control Ptet-GFP reporter fusion we found that the Tet promoter was much less efficiently induced in this strain background than in strain WO-1 (unpublished results). Hence, we expressed CPH1 from another regulatable promoter, POPT3, which is efficiently induced in strain SC5314 when the cells grow on BSA as a nitrogen source [33]. Expression of CPH1 from the OPT3 promoter during growth in YCB-BSA-YE medium strongly stimulated white-opaque switching in independently generated a and α derivatives of strain SC5314 (Fig. 7). Incubation of the untransformed parental strains in the same growth medium did not promote white-opaque switching, confirming that switching was induced by CPH1 expression. We consistently observed a higher switching frequency when the POPT3-CPH1 fusion was integrated into the OPT3-1 allele as compared to the OPT3-2 allele. Allele-specific differences in the activity of the OPT3 promoter did not seem to be the reason, because a POPT3-GFP reporter fusion was expressed at comparable levels after integration at either of the two loci (data not shown). Therefore, minor differences in the resulting OPT3-CPH1 hybrid transcripts (e.g., stability or translational efficiency) may specifically affect Cph1 levels and, consequently, the switching frequency. Regardless, these results demonstrate that Cph1-induced white-opaque switching is not specific to strain WO-1 and is independent of mating type.
The finding that activation of the Cph1-dependent MAP kinase pathway in white cells promoted switching to the opaque phase suggested that the presence of mating pheromone might also stimulate white-opaque switching. However, in contrast to opaque cells, white cells do not [22], [23] or not strongly (Fig. 8C) upregulate CPH1 expression in response to pheromone and no pheromone-induced white-opaque switching has been reported so far. In line with this, the addition of synthetic α-pheromone to white a-cells derived from strain SC5314 did not induce switching to the opaque phase under various growth conditions tested, and a mixture of opaque a- and α-derivatives of strain SC5314 (used as a source of a-pheromone) also did not stimulate white-opaque switching in the MTLα/α strain WO-1 when the cells were coincubated (data not shown, see Materials and Methods for details). As white cells use Tec1 to induce biofilm formation in response to pheromone, we also tested the effect of a-pheromone on tec1Δ mutants of strain WO-1; however, no pheromone-induced switching was observed in the mutants. Consequently, other signals may activate the MAP kinase pathway in a different way from that stimulated by pheromone to result in CPH1 instead of TEC1 upregulation in white cells, similar to the effect of the hyperactive Ste11ΔN467. We therefore investigated whether Cph1 is required for white-opaque switching in response to signals that efficiently stimulate switching in strain WO-1. For this purpose, we determined the switching frequency of the wild-type strain WO-1 and the cph1Δ mutants under various inducing conditions; mutants lacking the master regulator Wor1 or the positive regulator Czf1 were included for comparison. In contrast to czf1Δ mutants, in which the frequency of switching was drastically reduced under all tested conditions (no switching was observed in wor1Δ mutants, as expected), cells lacking Cph1 switched to the opaque phase with the same efficiency as the parental wild-type strain WO-1 when switching was induced by incubation in an anaerobic jar, in the presence of ketoconazole, or by growth on GlcNAc as carbon source (Fig. 10). Therefore, conditions that induce white cells to switch to the opaque phase by activating CPH1 remain to be discovered (see discussion).
The recent observation by several groups that switching of white cells to the opaque phase does not only occur stochastically in few cells of a population, but can be strongly stimulated by certain environmental cues [24], [25], [26], [28], suggested that expression of the master regulator WOR1 is induced by upstream signal transduction pathways in response to these signals. The cAMP/PKA signaling pathway has been implicated in the environmental induction of white-opaque switching, because deletion of components of this pathway reduced, albeit did not abolish, the stimulation of switching by GlcNAc and elevated CO2 concentrations [25], [26]. In this work, we used an overexpression approach to identify protein kinases that stimulate white-opaque switching. By generating and screening a comprehensive library of C. albicans strains that express all protein kinases of this organism from a tetracycline-inducible promoter, we discovered several kinases that could induce white cells to switch to the opaque phase. In addition to the recently identified Tpk2 [26], we found that the homologous kinase Tpk1 as well as the kinases Mps1 and Rad53 also stimulated white-opaque switching when overexpressed from the Tet promoter. The involvement of different protein kinases in the regulation of white-opaque switching probably reflects the fact that a variety of signals stimulate white cells to switch to the opaque phase. It should be noted that additional kinases not identified in our screening could nevertheless be involved in the control of white-opaque switching, because overexpression of a wild-type kinase may not necessarily have a phenotypic effect (as was the case for wild-type STE11, which did not stimulate switching).
A particularly intriguing result of our present study was that a hyperactive form of the MAPKKK Ste11, which acts in the pheromone response pathway, also promoted switching. This finding came as a surprise, because the pheromone-responsive MAP kinase signaling pathway is known to induce the mating response in opaque cells and biofilm formation in white cells, but so far it has not been implicated in the regulation of the switching event itself. No induction of white-opaque switching by pheromone has been reported in previous studies investigating the pheromone response of white cells [21], [23], and we also did not observe stimulation of switching by pheromone, suggesting that the MAP kinase pathway can be activated in an alternative way in white cells to promote switching to the opaque phase. The other kinases identified in our study do not seem to function via the MAP kinase pathway, because Tet-induced expression of MPS1, RAD53, TPK1, and TPK2 stimulated switching also in cph1Δ and ste11Δ mutants (see supplemental Fig. S1), arguing that these kinases function via alternative signaling pathways to activate WOR1. Unlike the hyperactive STE11, these other kinases also did not promote biofilm formation in our assays (data not shown), supporting the assumption that they do not activate the MAP kinase pathway under these conditions.
White and opaque cells use the same upstream components of the pheromone-induced MAP kinase signaling pathway, but different transcription factors to effect their specific responses. White cells induce expression of TEC1, but not CPH1, whereas opaque cells induce expression of CPH1, but not TEC1, in response to pheromone [22], [23], [34]. It is not currently known how the upregulation of the alternative downstream transcription factor, CPH1 in white cells and TEC1 in opaque cells, in the presence of pheromone is blocked in the two cell types. In Saccharomyces cerevisiae, components of the pheromone-responsive MAP kinase cascade are also used for the induction of a different developmental program, invasive growth, under starvation conditions [36]. Here, the scaffold protein Ste5, which binds all three components of the MAP kinase cascade, acts as a conformational switch that gates the flow of information to ensure a proper physiological response to different inducing signals [37]. In C. albicans, the homologous scaffold protein Cst5 is required for both the Cph1-mediated mating response of opaque cells and Tec1-dependent biofilm formation of white cells [32]. The hyperactive Ste11ΔN467, which lacks the binding site for Cst5 [35], caused upregulation of CPH1 instead of TEC1 in white cells, i.e., these cells had overcome the block in CPH1 induction upon activation of the MAP kinase pathway (Fig. 11). However, the absence of Cst5 was not sufficient to enable wild-type Ste11 to induce switching of white cells to the opaque phase, indicating that additional signals are required for Ste11 to recruit Cph1 and promote switching. Overexpression of CPH1 has been reported already a decade ago to increase mating efficiency [31]. At the time of that study, the linkage of mating to white-opaque switching had not yet been uncovered and the mating experiments were performed with white cells. While overexpression of CPH1 may have enhanced the mating response of opaque cells that were already in the population, our results suggest that the increased mating efficiency might also have been caused by Cph1-induced switching of white cells to the opaque phase.
We observed that some of the kinases identified in our study efficiently stimulated white-opaque switching only when their expression was induced in the preculture, but not during growth on agar plates (STE11ΔN467, TPK1), while others (RAD53, TPK2) promoted switching under both conditions (see supplemental Table S2). Indeed, growth conditions affected the inducibility of white-opaque switching by overexpressed STE11ΔN467 and CPH1: Growth of the cells under static conditions in microtiter plates, a procedure that we used to streamline screening of the library, resulted in more efficient switching than the usual culturing in Erlenmeyer flasks or glass tubes with shaking. Under the latter conditions, the switching rates obtained after Tet-induced CPH1 expression were only slightly reduced, but no significant induction of switching by Ste11ΔN467 was observed, indicating that white cells could more easily be stimulated to switch when growing on the bottom of a microtiter plate (data not shown). A reduced growth rate, which facilitates Wor1 accumulation in a cell [24], or local buildup of higher CO2 levels, which promote white-opaque switching [25], may contribute to this effect. In this regard, it is interesting to note that increased CO2 concentrations were reported to induce white-opaque switching only when the cells were grown on agar plates, but not in liquid culture [25], and GlcNAc induces switching only in aged, but not in fresh cultures [26]. In addition, Tet-induced expression of Flo8, another transcription factor that was recently found to be involved in the regulation of white-opaque switching, induced switching only in the presence of elevated CO2 concentrations [29]. Evidently, the ability of a regulatory factor to induce switching depends on the environmental conditions, because these will affect the activity of additional regulators.
It has been proposed that the pheromone-induced biofilm formation response observed in white cells has evolved via the recruitment of components from the ancestral pheromone response pathway (all upstream components, from the pheromone receptor to the MAP kinases Cek1 and Cek2) and the filamentation pathway (the transcription factor Tec1) as well as target genes for biofilm formation [22]. Our results demonstrate that this signaling pathway can be used in a highly flexible way, depending not only on the cell type but also on the manner in which it is activated. The downstream transcription factor Cph1 mediates the mating response of opaque cells, but can also stimulate white cells to switch to the opaque phase to become mating-competent. In the latter case, one of the two MAP kinases, Cek1, is specifically recruited to transmit the signal from activated Ste11 to Cph1, whereas Cek1 and Cek2 have at least partially redundant functions in the pheromone-induced biofilm formation of white cells and in the mating response of opaque cells [23], [31]. CPH1 was only required for the induction of white-opaque switching by the artificially activated Ste11, but not for the induction of switching by the tested environmental conditions and for spontaneous switching, in agreement with findings by other researchers [23]. Of note, Cek1 is also known to be activated upon cell wall stress, for example after treatment with the cell wall disturbing agent tunicamycin, but Cph1 has not been implicated as a Cek1 downstream target under these conditions [38], [39], and we were unable to stimulate white-opaque switching with tunicamycin (unpublished results). In MTL heterozygous cells, the Cph1-dependent MAP kinase pathway induces filamentous growth under starvation conditions [40], [41]. Our results provide further evidence that certain transcription factors (Cph1, Czf1, Efg1, Flo8) are involved in the regulation of both filamentation and white-opaque switching.
It is conceivable that the N-terminally truncated, hyperactive form of Ste11, which induced CPH1 upregulation in white cells and switching to the opaque phase, reflects a normal function of the MAP kinase pathway when activated by an unknown physiological signal. In this respect, it is remarkable that white-opaque switching, which was thought to be restricted to MTL homozygous strains, has recently been observed in MTLa/α strains under specific growth conditions [42]. So far, we have not found environmental conditions that promote white-opaque switching in a Cph1-dependent fashion and such conditions might be encountered only in suitable host niches in vivo. It is even possible that the presence of pheromone is a prerequisite, and that additional signals are required to overcome the block in CPH1 upregulation that is seen upon pheromone treatment of white cells in vitro. This, in turn, raises the intriguing hypothesis that C. albicans white cells may have the ability to react to the presence of a potential mating partner (pheromone-secreting opaque cells) by switching to the opaque phase and thus become themselves mating-competent in an appropriate environment.
The C. albicans strains used in this study are listed in supplemental Table S3. All strains were stored as frozen stocks with 15% glycerol at −80°C. The strains were subcultured separately in the white and opaque phases at room temperature on agar plates containing Lee's medium, pH 6.8 [43], and 5 µg/ml phloxine B, which selectively stains opaque colonies pink [44]. Strains were routinely grown in YPD (10 g yeast extract, 20 g peptone, 20 g glucose) or SD (1.7 g yeast nitrogen base without amino acids [YNB; BIO 101, Vista, Calif.], 20 g glucose per liter) liquid medium at 30°C in a shaking incubator. To prepare solid media, 1.5% agar was added to the media before autoclaving. For selection of nourseothricin-resistant transformants, 200 µg/ml nourseothricin (Werner Bioagents, Jena, Germany) was added to YPD agar plates. To obtain nourseothricin-sensitive derivatives in which the SAT1 flipper cassette was excised by FLP-mediated recombination, transformants were grown overnight in YCB-BSA-YE medium (23.4 g yeast carbon base, 4 g bovine serum albumin, 2 g yeast extract per liter, pH 4.0) without selective pressure to induce the SAP2 promoter controlling caFLP expression. Alternatively, strains containing a SAT1 flipper cassette in which the caFLP gene is expressed from the MAL2 promoter (as in plasmids pOP4G4, pOP4R2, and pMTLΔ1) were grown overnight in YPM medium (10 g yeast extract, 20 g peptone, 20 g maltose per liter) instead of YCB-BSA-YE to induce the MAL2 promoter. Appropriate dilutions were plated on YPD agar plates containing 10 µg/ml nourseothricin and grown for 2 days at 30°C. Nourseothricin-sensitive clones were identified by their small colony size and confirmed by restreaking on YPD plates containing 200 µg/ml nourseothricin as described previously [45]. YCB-BSA-YE medium was also used to induce gene expression from the OPT3 promoter.
To induce gene expression from the Tet promoter, white cells of the strains containing the protein kinase library were grown for 24 h at 30°C in Lee's medium in 96-well microtiter plates in the absence or presence of 50 µg/ml doxycycline, diluted 10−5, and spread on Lee's agar plates with or without 50 µg/ml doxycycline. The plates were incubated for 7 days at room temperature and the number of white, opaque, and mixed white/opaque colonies was determined. Induction of white-opaque switching by environmental signals was performed as described in the legend to Fig. 9. Incubation under anaerobic conditions was performed in an anaerobic jar (Anaerocult, Merck KGaA, Darmstadt, Germany) that generates an oxygen-free milieu in a CO2 atmosphere (18% CO2) within one hour.
To generate a comprehensive library containing all known or putative protein kinases and kinase regulators of C. albicans, we searched the Candida genome database (CGD, http://www.candidagenome.org) for genes that were annotated with this function. These genes were amplified from genomic DNA of strain SC5314 by PCR with primers that introduced a SalI site in front of the start codon and a BglII site behind the stop codon (primer sequences are available upon request). For genes with internal SalI or BglII sites, primers containing compatible XhoI and/or BamHI sites were used. The PCR products were appropriately digested and cloned in place of the GFP reporter gene in the SalI/BglII-digested vector pNIM6 [28], which is identical to the originally developed Tet-On vector pNIM1 [46], except that it contains the TEF3 transcription termination sequence instead of the ACT1 terminator behind GFP. All cloned genes were completely sequenced to confirm their identity and to exclude PCR errors.
For the deletion of CEK1, CEK2, CPH1, CST5, STE11, TEC1, and WOR2, the upstream and downstream regions of the genes were amplified as SacI-SacII and XhoI-ApaI fragments, respectively, and cloned on both sides of the SAT1 flipper cassette in plasmid pSFS5, a derivative of plasmid pSFS2 in which the caFLP gene is placed under the control of the SAP2 promoter instead of the MAL2 promoter [47] to result in pCEK1M1, pCEK2M1, pCPH1M1, pCST5M1, pSTE11M1, pTEC1M1, and pWOR2M3, respectively. For complementation of the cek1Δ and cph1Δ mutants, the complete CEK1 and CPH1 coding regions and flanking sequences were cloned as SacI-SacII fragments and inserted in place of the upstream flanking region of pCEK1M1 and pCPH1M1, generating pCEK1K1 and pCPH1K1, respectively. The CPH1 and TEC1 coding regions were also amplified and inserted in plasmid pNIM6 to generate doxycycline-inducible expression cassettes, as described above for the protein kinase expression library. For deletion of the MTLa or MTLα locus of strain SC5314, ca. 0.8 kb of the common flanking regions were amplified as SacI-SacII and XhoI-ApaI fragments and cloned on both sides of the SAT1 flipper cassette in plasmid pSFS2 to generate pMTLΔ1. To express CPH1 from the OPT3 promoter, the CPH1 coding sequence was substituted for GFP in plasmid pOPT3G22 [33], resulting in pOPT3-CPH1. HA-tagged versions of CEK1, CEK2, and CPH1 were generated by amplifying the C-terminal parts of the genes with primers that introduced a BamHI site (encoding a Gly-Ser linker) instead of the stop codons. The PCR products were digested with BamHI and at internal or introduced SacI sites, fused to a PCR-amplified BamHI-SacII fragment containing the 3×HA-TACT1 sequences from pZCF36H2 [48], and substituted for the upstream flanking sequences in the corresponding deletion cassettes, resulting in pCEK1H1, pCEK2H1, and pCPH1H1.
C. albicans strains were transformed by electroporation [45] with the gel-purified inserts from the plasmids described above. The cassettes from the Tet-inducible protein kinase expression library were separated from the plasmid backbone by digestion with SacII/ApaI or SacII/KpnI (in some cases a partial digest was required). Gene deletion and reinsertion cassettes were excised from the corresponding plasmids by SacI/ApaI digestion. For insertion of POP4-GFP and POP4-RFP reporter fusions into various strain backgrounds, the ApaI-SacI fragments from pOP4G2 (with the caSAT1 marker alone) or from pOP4G4 and pOP4R2 (with the recyclable SAT1 flipper cassette) [18] were used. The correct integration of each construct was confirmed by Southern hybridization using the flanking sequences as probes. In each case, two independent series of strains were generated and used for further analysis.
Genomic DNA from C. albicans strains was isolated as described previously [45]. The DNA was digested with appropriate restriction enzymes, separated on a 1% agarose gel, transferred by vacuum blotting onto a nylon membrane, and fixed by UV crosslinking. Southern hybridization with enhanced chemiluminescence-labeled probes was performed with the Amersham ECL Direct Nucleic Acid Labelling and Detection System (GE Healthcare UK Limited, Little Chalfont Buckinghamshire, UK) according to the instructions of the manufacturer.
Overnight cultures of C. albicans strains in SD medium were diluted to 107 cells/ml in Lee's medium with or without 50 µg/ml doxycycline and 500 µl of these suspensions was added to each well of a 24-well polystyrene microtiter plate. After 24 h of incubation at 30°C, the wells were gently washed with PBS and imaged. Quantification of biofilm formation was performed as previously described [49], with some modifications. Three hundred microliters of the cell suspensions was added to each well of a 96-well polystyrene microtiter plate. A well containing Lee's medium without cells was included as reference. After 24 h of incubation at 30°C, the medium was aspirated and the wells were washed three times with 200 µl sterile PBS. Subsequently, 100 µl crystal violet solution (1%) was added to each well for 5 min. The wells were washed three times with sterile water and bound crystal violet was released by adding 200 µl of 33% acetic acid. The obtained solution was transferred to a new microtiter plate and the absorbance read at 595 nm.
Overnight cultures of C. albicans strains were diluted 10−2 in Lee's medium with 50 µg/ml doxycycline and grown for 18 h at 30°C in individual wells of a flat-bottomed 96-well polystyrene microtiter plate. Cells were harvested and total RNA was extracted by the hot acidic phenol method [50] combined with a purification step with the RNeasy mini kit (Qiagen, Hilden, Germany), and treated with Turbo DNA-free DNase (Ambion, Austin, TX). cDNA was synthesized using 500 ng of total RNA from each sample with the Superscript III Super Mix (Invitrogen, Karlsruhe, Germany). PCR was performed on a MyiQ Real-time PCR system using the iQ SYBR Green Supermix kit (Bio-Rad Laboratories, Hercules, CA). Relative mRNA levels, adjusted to ACT1 mRNA levels, were calculated using expression levels in the wild-type strain WO-1 (set to 1) as a reference. The primers used are listed in supplemental Table S4.
Overnight cultures of the strains were tenfold diluted in 10 ml fresh Lee's medium with and without doxycycline or α-pheromone (see below) and grown without shaking for 18 h at 30°C. The cells were pelleted by centrifugation, washed in 2.5 ml breaking buffer (100 mM Tris-HCl [pH 7.5], 200 mM NaCl, 20% glycerol, 5 mM EDTA), and resuspended in 500 µl breaking buffer supplemented with protease and phosphatase inhibitors (100 mM Tris-HCl [pH 7.5], 200 mM NaCl, 20% glycerol, 5 mM EDTA, 0.1% β-mercaptoethanol, cOmplete EDTA-free Protease Inhibitor Cocktail and PhosStop Phosphatase Inhibitor Cocktail [Roche Diagnostics GmbH, Mannheim, Germany]). An equal volume of 0.5-mm acid-washed beads was added to each tube. Cells were mechanically disrupted on a FastPrep-24 cell-homogenizer (MP Biomedicals, Solon, USA) for two 40-seconds intervals, with 1 min on ice between each cycle. Samples were centrifuged at 13,000 rpm for 10 min at 4°C, the supernatant was collected, and the protein concentration was quantified using the Bradford protein assay. Extracts were heated for 5 min at 95°C, and equal amounts of protein of each sample were separated on an SDS-10% polyacrylamide gel. Separated proteins were transferred onto a nitrocellulose membrane with a Mini-Protean System (Bio-Rad, Munich, Germany) and stained with Ponceau S to control for equal loading. Membranes were blocked in 5% milk in PBS at room temperature for 1 hour and subsequently incubated overnight at 4°C with rat monoclonal anti-HA-Peroxidase antibody (Roche Diagnostics GmbH, Mannheim, Germany). Membranes were washed in PBS containing 0.1% Tween-20 and signals detected with ECL chemiluminescence detection system (GE Healthcare Bio-Sciences GmbH, Munich, Germany). For phosphatase treatment, cell extracts were washed in Amicon-ultra 10 K columns (Millipore Corporation, Billerica, USA) with three volumes of breaking buffer without EDTA and phosphatase inhibitors, and subsequently washed with two volumes of phosphatase buffer supplemented with 1 mM MnCl2. Washed cell extracts were incubated with λ Protein Phosphatase (New England Biolabs, Ipswich, USA) at 30°C for 30 min.
To investigate whether white-opaque switching could be induced by pheromone, white cells of MTLa/Δ derivatives of strain SC5314 were grown overnight in SD medium at room temperature, washed with water, and resuspended in Lee's medium or Spider medium to a concentration of 107 cells/ml. α-pheromone (Seqlab, Göttingen, Germany) was added at a concentration of 10 µg/ml, and the cultures were incubated at room temperature for 24 h. An appropriate dilution of the cultures was spread on Lee's agar plates with phloxin B and incubated for 7 days at room temperature to determine the frequency of opaque colonies. The experiment was repeated several times using additional additives and conditions along the pheromone treatment, including addition of 1 µg/ml pepstatin A (to prevent pheromone degradation by secreted aspartic proteases) and/or 10% FCS, incubation in the presence of 1% CO2, incubation at 30°C or 37°C, and elongation of the treatment time to 48 h. In addition, the cells were also grown in the absence or presence of α-pheromone on Lee's agar plates without or with 1.2 M sorbitol in normal air or in a 1% CO2 atmosphere.
To test the effect of a-pheromone on the MTLα/α strain WO-1, opaque cells of MTLa/Δ and MTLα/Δ derivatives of strain SC5314 were mixed (to serve as a source of both a- and α-pheromone) and incubated with white cells of nourseothricin-resistant derivatives of strain WO-1 (the wild-type strains WOP4G2A and -B and the tec1Δmutants WTEC1M3A and -B) to a final concentration of 107 cells/ml in Lee's medium. The proportions of cells in the mixture were 50% white test cells, 25% opaque a-cells, and 25% opaque α-cells. The cell suspensions were incubated at room temperature for 24 h, diluted, and spread on Lee's agar plates with phloxin B and 100 µg/ml nourseothricin, on which only the nourseothricin-resistant test strains could grow. The plates were incubated for 7 days at room temperature and the presence of opaque colonies was recorded. In control experiments, we incubated RFP-expressing MTLa/Δ and MTLα/Δ opaque cells (strains SCMTLαM2BOP4R22 and SCMTLaM2BOP4R22) together with GFP-expressing opaque cells of strain WO-1 (strains WOP4G2A and -B). Microscopic inspection of the cells after 6 h of incubation showed that both GFP- and RFP-expressing cells formed shmoos, demonstrating that sufficient a-pheromone was produced by the mixture to induce the mating response in the MTLα/α strain WO-1.
|
10.1371/journal.pmed.1002055 | South Asia as a Reservoir for the Global Spread of Ciprofloxacin-Resistant Shigella sonnei: A Cross-Sectional Study | Antimicrobial resistance is a major issue in the Shigellae, particularly as a specific multidrug-resistant (MDR) lineage of Shigella sonnei (lineage III) is becoming globally dominant. Ciprofloxacin is a recommended treatment for Shigella infections. However, ciprofloxacin-resistant S. sonnei are being increasingly isolated in Asia and sporadically reported on other continents. We hypothesized that Asia is a primary hub for the recent international spread of ciprofloxacin-resistant S. sonnei.
We performed whole-genome sequencing on a collection of 60 contemporaneous ciprofloxacin-resistant S. sonnei isolated in four countries within Asia (Vietnam, n = 11; Bhutan, n = 12; Thailand, n = 1; Cambodia, n = 1) and two outside of Asia (Australia, n = 19; Ireland, n = 16). We reconstructed the recent evolutionary history of these organisms and combined these data with their geographical location of isolation. Placing these sequences into a global phylogeny, we found that all ciprofloxacin-resistant S. sonnei formed a single clade within a Central Asian expansion of lineage III. Furthermore, our data show that resistance to ciprofloxacin within S. sonnei may be globally attributed to a single clonal emergence event, encompassing sequential gyrA-S83L, parC-S80I, and gyrA-D87G mutations. Geographical data predict that South Asia is the likely primary source of these organisms, which are being regularly exported across Asia and intercontinentally into Australia, the United States and Europe. Our analysis was limited by the number of S. sonnei sequences available from diverse geographical areas and time periods, and we cannot discount the potential existence of other unsampled reservoir populations of antimicrobial-resistant S. sonnei.
This study suggests that a single clone, which is widespread in South Asia, is likely driving the current intercontinental surge of ciprofloxacin-resistant S. sonnei and is capable of establishing endemic transmission in new locations. Despite being limited in geographical scope, our work has major implications for understanding the international transfer of antimicrobial-resistant pathogens, with S. sonnei acting as a tractable model for studying how antimicrobial-resistant Gram-negative bacteria spread globally.
| Antimicrobial resistance is a major issue in Shigella, and ciprofloxacin is a recommended treatment for Shigella infections.
Ciprofloxacin-resistant Shigella sonnei are being increasingly isolated globally.
We performed genome sequencing on 60 ciprofloxacin-resistant S. sonnei isolated in six countries within and outside of Asia.
Placing these genome sequences in context with other strains, we found that all ciprofloxacin-resistant S. sonnei formed a single clade, with South Asia as the likely primary source of these organisms.
These organisms are also being regularly exported across Asia and intercontinentally into Australia, the United States, and Europe.
The number of S. sonnei sequences available from other locations limited our analysis.
Our work suggests that a single clone, which is widespread in South Asia, is driving the current global surge of ciprofloxacin-resistant S. sonnei.
We argue that S. sonnei acts as a tractable model for studying how antimicrobial-resistant, Gram-negative bacteria spread globally.
| Diarrheal disease is the second most common cause of mortality in children under the age of 5 y worldwide, equating to approximately 800,000 deaths per year [1]. The recent Global Enteric Multicentre Study (GEMS), a large, prospective, case-control study focusing on mild and severe paediatric diarrheal illnesses in sub-Saharan Africa and South Asia, found that Shigella (a genus of Gram-negative enteric bacteria) were amongst the top four most prevalent diarrhoeal pathogens in these settings [2]. The most recent estimates suggest that Shigella infections account for around 125 million cases of diarrhoea annually, with the majority occurring in children in low-income countries [3]. There are four Shigella species (dysenteriae, boydii, flexneri, and sonnei), but the overwhelming majority of the current global burden is presently caused by S. sonnei and S. flexneri. Present-day international epidemiology of the various Shigella species is particularly intriguing, as S. sonnei is replacing S. flexneri as the most common cause of shigellosis worldwide; this pattern is accentuated in regions undergoing rapid economic development [4,5], where S. flexneri dominated as recently as a decade ago.
Shigella infections are characterised by the invasion and disruption of the epithelial cells lining the gastrointestinal mucosa, resulting in mucous and/or bloody diarrhoeal discharge. Although shigellosis is typically self-limiting, antimicrobial treatment is used to prevent complications, reduce dysenteric discharge, and curb post-symptomatic faecal shedding [6,7]. Consequently, resistance to antimicrobials restricts treatment options, placing vulnerable individuals suffering from shigellosis at increased risk of complications and increasing the likelihood of protracted faecal shedding. One of the current recommended first-line treatments for shigellosis is the fluoroquinolone ciprofloxacin [8]. The fluoroquinolones target the DNA gyrase, a type II topoisomerase that is essential for bacterial DNA replication and transcription [9].
Antimicrobial resistance is an emerging global issue in S. sonnei, with a specific multidrug-resistant (MDR) lineage (III) now dominating internationally. Furthermore, organisms belonging to lineage III appear to be highly proficient at acquiring resistance to additional antimicrobials (including third-generation cephalosporins) when they are introduced into new locations [10]. However, given their common usage and broad spectrum of activity, resistance against the fluoroquinolones is the most concerning. Since the first isolation of S. sonnei with reduced susceptibility to ciprofloxacin in Japan in 1993 [11], ciprofloxacin-resistant S. sonnei have been increasingly reported throughout Asia [12–14]. Furthermore, public health laboratories in several non-Asian countries with low incidences of shigellosis have reported the isolation of ciprofloxacin-resistant S. sonnei, often from individuals reporting recent travel to locations with a high risk of shigellosis [15–17].
Whole-genome sequencing has proven to be the gold standard for tracking the international dissemination of clonal bacterial pathogens [18,19], and we have previously exploited this method to study the phylogenetic structure and spread of S. sonnei at both national and intercontinental levels [10,20]. Hypothesising that Asia was a hub for the recent international spread of ciprofloxacin-resistant S. sonnei, we performed whole-genome sequencing and phylogenetic characterisation of a collection of ciprofloxacin-resistant S. sonnei isolated from within and outside Asia, aiming to explore the origins of this growing international epidemic.
S. sonnei isolates from Bhutan, Thailand, and Vietnam were collected during diarrheal surveillance studies [14]. IRB approval was granted for these studies (including organism characterization) from the Research Ethics Board of Health (REBH), Ministry of Health, Bhutan (Bhutan study), the Walter Reed Army Institute of Research (WRAIR) Institutional Review Board, USA (Bhutan and Thailand studies), the Oxford Tropical Research Ethics Committee (OxTREC), UK, and the Hospital for Tropical Diseases Ho Chi Minh City, Vietnam (Vietnam study). Written (Vietnam) or oral (Thailand and Bhutan) consent was obtained from a parent or guardian at the time of enrolment into the study. The target patient groups for these studies were generally hospitalised children aged less than 5 y residing in close proximity to the study centres. S. sonnei isolates from Cambodia were collected at the Angkor Hospital for Children (AHC) in Siem Reap province from the routine diagnostic laboratory; no patient data were collected, and these organisms were not subject to local or international IRB approval. The ciprofloxacin-resistant S. sonnei from countries outside Asia were collected and characterized by the National Salmonella, Shigella, and Listeria monocytogenes Reference Laboratory, Galway, Ireland, and the Microbiological Diagnostic Unit Public Health Laboratory, Melbourne, Australia. These isolates were generally, but not exclusively, obtained from patients reporting recent travel to countries with a high incidence of shigellosis in Asia (Table 1). As these isolates were from anonymous sources and collected at local public health laboratories, these were not subject to IRB approval, and informed consent was not required.
Aiming to investigate the current international upsurge in ciprofloxacin-resistant S. sonnei in detail, we gathered a collection of 60 contemporary ciprofloxacin-resistant S. sonnei from six countries for whole-genome sequencing. The isolates originated from Asian countries with a high incidence of shigellosis (Vietnam, n = 11; Bhutan, n = 12; Thailand, n = 1; Cambodia, n = 1), as well as countries with a low incidence of shigellosis (Australia, n = 19; Ireland, n = 16). Twelve additional ciprofloxacin-susceptible S. sonnei sequences from these settings were also included for phylogenetic context. All strains were isolated independently between 2010 and 2015; details of the isolates used in this study are shown in Table 1.
Susceptibility to ciprofloxacin was determined by disk diffusion, E-test, agar dilution, or broth microdilution, depending on the collaborating institution, and susceptibility breakpoints were interpreted according to the European Committee on Antimicrobial Susceptibility Testing (http://www.eucast.org/clinical_breakpoints). Namely, resistance was determined as strains with a zone of inhibition <19 mm (5 μg disc) and/or a minimum inhibitory concentration (MIC) >1 μg/ml against ciprofloxacin; the various location-specific methods and resulting data are described in Table 1.
All isolated S. sonnei were subcultured and subjected to DNA extraction prior to whole-genome sequencing on various Illumina platforms to produce pair-ended short read sequences; the specific sequencing system and the resulting public database numbers are shown in Table 1. We additionally included 14 S. sonnei sequences from the NCBI database under the Bioproject accession number PRJNA218110. These organisms were isolated in the US and deposited as part of the GenomeTrackr Project. All sequences were mapped to the S. sonnei Ss046 reference sequence (Accession number: NC_007384) using SMALT (version 0.7.4), and SNPs were called against the reference and filtered using SAMtools [21]. To contextualize all ciprofloxacin-resistant S. sonnei within the global phylogeny, we appended our collection to include 133 publicly available sequences from a previous global analysis (accession ERP000182) [20]. Previously characterized mobile genetic elements and putative recombination (predicted using Gubbins) were removed [20], resulting in a gap-free alignment of 211 non-duplicate pseudo-whole genome sequences of 4,738 SNPs. A whole-genome phylogeny was inferred from this alignment using RAxML v8.1.3 under the GTRGAMMA substitution model, and sufficient bootstrap replicates were determined automatically using the extended majority rule (MRE) bootstrap convergence criterion. Bootstrap values >75% signify strong statistical support for a node within a maximum likelihood phylogeny, thus indicating that all isolates falling within that lineage are highly likely to be linked in evolutionary history at a more recent time than those falling outside of that lineage. In order to obtain a refined phylogenetic structure of the Central Asia clade, we applied the aforementioned approach to a set of 97 S. sonnei sequences (86 novel sequences and 11 historical sequences) belonging to this clade. This resulted in an alignment of 1,121 SNPs, which was used for phylogenetic inference. De novo assemblies were generated for each read set using Velvet and VelvetOptimiser, and read sets were mapped back to each assembly [22]. A manually curated database based on ResFinder [23] was mapped against each assembly to identify mobile resistance genetic determinants in all ciprofloxacin-resistant strains.
We constructed a whole genome phylogeny of S. sonnei, incorporating sequences from 133 globally representative isolates and 86 novel isolates from Vietnam, Cambodia, Thailand, Bhutan, Australia, Ireland, and the US. The novel sequences included 60 from ciprofloxacin-resistant (MIC >1 μg/ml or zone of inhibition <19 mm) organisms and 26 from ciprofloxacin-susceptible organisms (or of unknown ciprofloxacin susceptibility isolated in the US). The related metadata for the bacterial isolates are shown in S1 Table. The overall tree topology reflected the previously described global phylogenetic structure [20], confirming the presence of four distinct lineages (I, II, III, and IV); lineage III was the most commonly represented and the most widely geographically distributed (Fig 1A). All ciprofloxacin-resistant S. sonnei formed a single, well-supported monophyletic clade within the Central Asian expansion of lineage III (Central Asia III), an MDR group that is closely related but distinct from the Global III clade (Fig 1A and 1B).
We next performed a more detailed phylogenetic reconstruction of the Central Asia III clade, incorporating sequence data from the 60 phenotypically ciprofloxacin-resistant isolates and 26 others (ciprofloxacin-susceptible or of unknown ciprofloxacin susceptibility), along with 11 historical Central Asia III sequences sourced from our previous global study (Fig 1B) [20]. The majority of the Central Asia III isolates carried more than three antimicrobial resistance genes, encoding resistance to a wide range of first-line antimicrobials including tetracycline (tetA), streptomycin (strAB), and co-trimoxazole (dfrA1 and sul2), as previously described [20]. No plasmid-mediated quinolone resistance genes (qnr, qepA, oxqAB, aac(6’)lb-cr) were detected in the sequences of any of the ciprofloxacin-resistant S. sonnei. We additionally examined the genome sequences for mutations in the quinolone resistance-determining region (QRDR) within the DNA gyrase gene (gyrA) and the topoisomerase IV gene (parC), the regions encoding the target residues for fluoroquinolones. Overlaying these mutations on the phylogenetic tree indicated that the gyrA-S83L mutation, the first sequential mutation that confers reduced susceptibility against fluoroquinolones, has arisen independently within the Central Asia III clade on at least four separate occasions (Fig 1B). Amongst the isolates examined here for the first time, extensive resistance to ciprofloxacin can be attributed to a single clonal emergence event, via the sequential accumulation of gyrA-S83L followed by parC-S80I and gyrA-D87G, except for a single outlier isolated in Australia (Fig 1B). Strong bootstrap support at each of these nodes suggests that these mutations were sequential lineage-defining events, with the final gyrA-D87G mutation preceding the expansion and intercontinental dissemination of the Central Asia III clade. These three QRDR mutations were also shared by ten phenotypically uncharacterized S. sonnei from the US, thus providing genotypic evidence for ciprofloxacin resistance. The single outlier isolate shares the gyrA-S83L and parC-S80I QRDR mutations of the other ciprofloxacin-resistant isolates, but harbours gyrA-D87N rather than a gyrA-D87G, and is within a closely related out-group of the major ciprofloxacin-resistant clone (Fig 1B).
We additionally mapped the country of isolation and patient travel history onto the Central Asia III phylogeny to investigate the geographical structure of the clade (Fig 1B). For the ciprofloxacin-resistant S. sonnei isolated from countries with a low incidence of shigellosis (Ireland, Australia, and US), and for which data on recent travel history was confirmed (27/45; 60%), India was the most commonly reported travel destination (21/27; 78%). The majority of the isolates associated with travel to India clustered closely with strains isolated in neighbouring Bhutan. These data suggest that South Asia is a primary source of ciprofloxacin-resistant S. sonnei that have increasingly been isolated both within and outside of Asia in recent years. Furthermore, greater genetic diversity was observed within the South Asian S. sonnei than within the other sampled countries (Fig 1B), suggesting that this region currently acts as the most likely geographical source population.
Our data also show evidence of regional diversification of ciprofloxacin-resistant S. sonnei within Asia. The phylogenetic structure is highly suggestive of a clonal expansion of ciprofloxacin-resistant S. sonnei in Southeast Asia, specifically within Vietnam, as indicated by a long branch with 100% bootstrap support (Fig 1B). We additionally noted that S. sonnei nested within this clonal expansion were also isolated from travellers returning from countries including Cambodia and Thailand, indicating that isolates from this lineage have spread widely across Southeast Asia, as well as having been introduced into Australia on at least five separate occasions. An additional well-supported subpopulation of ciprofloxacin-resistant S. sonnei, isolated in Ireland (five individuals with no recent history of travel and one individual returning from Germany) and the US, likely represents an extended chain of local transmission within Europe and the US (Fig 1B). Alternatively, although less likely, this subpopulation could represent multiple importations and short-term local transmission of S. sonnei strains originating from an unsampled source population. Whilst it was not possible to identify the geographical source or extent of local transmission definitively, the isolates most closely related to this European/US subpopulation originated in India and Bhutan, again suggesting South Asia was the most likely origin of this subpopulation. These two expansion events in Southeast Asia and Europe/US indicate that this clone of ciprofloxacin-resistant S. sonnei is also capable of sustained circulation upon introduction into new locations.
Here we provide direct evidence for the ongoing global expansion of S. sonnei exhibiting new and clinically relevant antimicrobial resistance profiles. What is more, as we can use phylogeography in high resolution, we can link the reservoir of these organisms to a specific region. Therefore, this study has significant implications for understanding the international trafficking of antimicrobial-resistant bacterial pathogens from Asia. Furthermore, we suggest that, as a single-serotype, human-adapted pathogen with a clonal population structure, S. sonnei serves as a tractable model for understanding how Gram-negative antimicrobial resistant pathogens are being regularly mobilised around the globe.
To our knowledge, this is the first study that has used whole-genome sequencing to examine the emergence and global spread of ciprofloxacin-resistant S. sonnei. Our data show that all sequenced extant ciprofloxacin-resistant S. sonnei, though sourced from disparate geographical locations, belonged to a single clonal expansion of lineage III, with South Asia being the most likely hub for its origin and spread. Our findings support previous hypotheses suggesting that ciprofloxacin-resistant S. sonnei in industrialised countries is being imported from South Asia [15,16]. A recent estimation of worldwide antimicrobial usage reported that India was the largest consumer of antimicrobials in 2010 [24]. Additionally, the fluoroquinolones are ranked as the most common antimicrobial prescribed for acute enteric diseases in India and Bangladesh [25,26]. The intensive use of fluoroquinolones in a region where there are foci of high population density and inconsistent access to good sanitation is likely to have contributed to emergence of ciprofloxacin-resistant enteric bacteria, such as S. sonnei and Salmonella Typhi, on the Indian subcontinent [19]. Global dissemination of these organisms is likely facilitated by the volume of travel between these regions and other areas of the world.
Our new data highlight the limitations of current typing protocols for tracking S. sonnei. It had been previously observed that some of the ciprofloxacin-resistant S. sonnei isolates in this study (originating from Bhutan and Ireland) shared a similar XbaI pulsed field gel electrophoresis (PFGE) pattern [14,15]. This pulsotype has been observed previously in India and Bangladesh [12,13,27–29], as well as in Canada [30], Belgium [31], and Japan [32], where the association with ciprofloxacin resistance was inconsistent. However, PFGE in this context did not offer sufficient granularity to link all of the isolates or provide sufficient resolution into the regional evolution of S. sonnei. Our phylogenetic analyses show that this pulsotype is associated with a phylogenetic lineage, supporting the notion that this pulsotype actually represents a widespread and pervasive subclade of Central Asia III.
This work has limitations. First, the lack of historical organisms from South Asia restricts our inference to only the current situation. Furthermore, additional contemporary organisms from other settings would have improved our understanding of the current geographical spread of this clonal group. Although the data included in this analysis was generated from organisms sampled in diverse geographic locations, the majority of sequences were retrieved from South Asia, Southeast Asia, Europe, Australia, and the Americas. It is possible, albeit unlikely, that our study might have overlooked an undersampled population that may ultimately act as an additional reservoir. This issue is inherent to all phylogeographical investigations and can potentially be overcome by global-scale prospective sampling to capture and characterize maximal global diversity. Notwithstanding these limitations, whole-genome sequencing of these geographically disparate organisms, together with plausible epidemiological links, has provided data at the highest resolution for deciphering the emergence and international spread of ciprofloxacin-resistant S. sonnei. Future studies interrogating extensive spatial and temporal collections of ciprofloxacin-resistant S. sonnei, as well as the S. sonnei diversity specific to South Asia prior to and during the emergence of antimicrobial resistance, are essential to further elucidate the origins and epidemiological dynamics of these populations. These supplementary investigations will greatly aid our efforts in controlling the spread of the current ciprofloxacin-resistant clone and to prevent future emergent antimicrobial-resistant bacterial populations.
In conclusion, the international surge of ciprofloxacin-resistant S. sonnei clone poses a substantial global health challenge, and our data show this threat is not only manifested in sporadic cases from returning travellers but also the establishment of endemic transmission in new settings. The latter is already evident in high shigellosis incidence areas such as Southeast Asia. Therefore, integrative efforts from both the research community and public health authorities should be prioritised to track, monitor, and prevent the international spread of this key enteric pathogen.
|
10.1371/journal.pntd.0004968 | An Ecological Assessment of the Pandemic Threat of Zika Virus | The current outbreak of Zika virus poses a severe threat to human health. While the range of the virus has been cataloged growing slowly over the last 50 years, the recent explosive expansion in the Americas indicates that the full potential distribution of Zika remains uncertain. Moreover, many studies rely on its similarity to dengue fever, a phylogenetically closely related disease of unknown ecological comparability. Here we compile a comprehensive spatially-explicit occurrence dataset from Zika viral surveillance and serological surveys based in its native range, and construct ecological niche models to test basic hypotheses about its spread and potential establishment. The hypothesis that the outbreak of cases in Mexico and North America are anomalous and outside the native ecological niche of the disease, and may be linked to either genetic shifts between strains, or El Nino or similar climatic events, remains plausible at this time. Comparison of the Zika niche against the known distribution of dengue fever suggests that Zika is more constrained by the seasonality of precipitation and diurnal temperature fluctuations, likely confining autochthonous non-sexual transmission to the tropics without significant evolutionary change. Projecting the range of the diseases in conjunction with three major vector species (Aedes africanus, Ae. aegypti, and Ae. albopictus) that transmit the pathogens, under climate change, suggests that Zika has potential for northward expansion; but, based on current knowledge, our models indicate Zika is unlikely to fill the full range its vectors occupy, and public fear of a vector-borne Zika epidemic in the mainland United States is potentially informed by biased or limited scientific knowledge. With recent sexual transmission of the virus globally, we caution that our results only apply to the vector-borne transmission route of the pathogen, and while the threat of a mosquito-carried Zika pandemic may be overstated in the media, other transmission modes of the virus may emerge and facilitate naturalization worldwide.
| A combination of media attention and the declaration of a World Health Organization state of emergency have made the pandemic expansion of Zika virus a topic of great public concern. Understanding the threat North America faces from the still-expanding viral range requires an understanding of the historical range and ecology of the disease, a topic currently difficult to study due to incomplete occurrence data. We compile the most comprehensive geospatial dataset of Zika occurrences in its native range, beginning with its discovery in 1947, and build bioclimatic models that set an outer bound on where the virus is likely to persist. Our results suggest Zika is likely far more constrained than the closely-related dengue fever, on which many projections have been based. While Zika poses a serious threat in current outbreak regions and is clearly a high-priority neglected tropical disease, our models suggest that even under an extreme climate change scenario for 2050, the disease is unlikely to become cosmopolitan in most temperate regions as a vector-borne disease, a discrepant finding from the results of non-ensemble modeling methods. Despite that, sexual transmission remains a serious public health concern, and a route by which Zika could become a severe public health emergency in temperate zones, including in the United States.
| Following a twenty-fold upsurge in microcephalic newborns in Brazil linked to Zika virus (ZIKV), the World Health Organization has declared an international health emergency. [1] Despite being profiled for the first time in 1947. [2] Zika remained poorly characterized at a global scale until the last six months. Thus, the present pandemic expansion in the Americas poses a threat of currently unknown magnitude. Closely related to dengue fever, Zika conventionally presents as a mild infection, with 80% of cases estimated to be asymptomatic. [3] The cryptic nature of infection has resulted in sporadic documentation of the disease and rarely includes spatially explicit information beyond the regional scale. [1, 4–6] This greatly limits the confidence with which statistical inferences can be made about the expansion of the virus. With an estimated 440,000–1,300,000 cases in Brazil in 2015, [3] and continuing emergence of new cases in Central America and, most recently, the United States, assessing the full pandemic potential of the virus is an urgent task with major ramifications for global health policy.
Current evidence portrays the global spread of ZIKV as a basic diffusion process facilitated by human and mosquito movement, a hypothesis supported by the frequency of infected traveler case studies in the Zika literature. [7–10] Tracing phylogenetic and epidemiological data has revealed the expansion of ZIKV has occurred in a stepwise process through the South Pacific, moving the disease from Southeast Asia into French Polynesia and the Philippines, and subsequently to Easter Island. [1, 4–6] Based on phylogenetic reconstruction, ZIKV is assumed to have dispersed into South America as recently as three years ago from the last of those locations, [11] and the virus is presumed to be at a biogeographic disequilibrium in the Americas. With cases in the ongoing outbreak in Colombia, El Salvador, Guatemala, Paraguay, and Venezuela, and by November of last year, as far north as Mexico and Puerto Rico, the full potential distribution of the disease remains unknown. Moreover, several alternative explanations for the disease’s expansion remain overlooked; most notably, the role of climate change in Zika’s expansion has not yet been thoroughly investigated. [12]
We present three competing hypotheses that describe the path of expansion that Zika could take, based on evaluations of the ecological niche of the virus within and outside of its vectors. First, if Zika has no additional climatic constraints relative to those of its vectors, future range expansions should match mosquito ranges. Second, if Zika has a transmission niche that is constrained by climatic factors within the ranges of its mosquito vectors, its range may be much more limited—with, as we show below, possible confinement to the tropics—and cases in North America could be driven by human dispersal or extreme episodic weather events. Finally, it is possible that the expansion of Zika into North America may be a steady range expansion beyond the known niche in its native range, facilitated by climatic shifts or by genetic shifts in the virus or vectors. To test these hypotheses, we present a spatially explicit database of Zika occurrences from the literature and an ensemble of ecological niche models [13] using that data to map the potential distribution of the virus.
Occurrence data for Zika virus was compiled from the literature from studies dating as far back as the original discovery of the virus in Zika Forest, Uganda in 1947. While the asymptomatic nature of the virus limits the total availability of data, lack of evidence for spatial patterns in symptoms in the native range suggest this is an unlikely cause of spatial bias (and instead, merely limits total dataset size). Special attention was paid to correctly attributing cases of travelers to the true source of infection. Locality data was extracted from a combination of clinical cases and seropositivity surveys in humans and mosquitoes, and georeferenced using a combination of Google Maps for hospitals and the Tulane University GEOLocate web platform for the remainder, [14] which allows for the attribution of an uncertainty radius to points only identified to a regional level. Four points were georeferenced in the New World but excluded from niche models because a limited sample as small as four points was likely to significantly bias predictions (compared to the necessary number of pseudoabsences in the same region). Thus, sixty points from the Old World were used in the final models presented in our paper after eliminating data from the current outbreak in the Americas. All points included in our dataset had an outer-bound of at most 65 km of uncertainty, with most substantially less. Constraining datasets based on an uncertainty threshold will become more statistically feasible in future studies once more survey data become available. In the present study, we deemed that the additional information gained from each point outweighed the potential impact of the uncertainty on model performance (S1 Table). We note that for similar reasons, we did not subsample our dataset for spatial thinning in our main models, as software packages like spThin allow, [15] due to information-accuracy tradeoffs; and the strong final performance of models (and the correspondence of our predictions for dengue and Aedes species to published “gold standard” niche models) speaks to the appropriateness of the underlying data and variables. Sensitivity analyses in the literature unequivocally suggest that accuracy of the modeling methods we employ plateaus at or near 50 points, justifying the use of a dataset of this size. [16–18]
Occurrence data for the other species included in our study were compiled from the literature. For Aedes africanus, we used a dataset of 99 points downloaded from the Global Biodiversity Informatics Facility (www.gbif.org). GBIF’s coverage of Aedes aegypti and Aedes albopictus was deemed to be lacking, so occurrences for those species were taken from the previously published work of Kraemer et al. [19–20] Finally, Messina et al.’s database was used for dengue, [21] as it has been previously published and used with great success to generate a global distribution model. [22] Both of these datasets were reduced down to point-only data (i.e., polygons of occurrence were excluded), leaving 5,216 points for dengue and 13,992 and 17,280 points for Ae. aegypti and Ae. albopictus respectively.
A number of other Zika vectors are known from previous reports, including at least a dozen Aedes species, as well as Anopheles coustani, Culex perfuscus, and Mansonia uniformis. [23–24] While we do not include these vectors in this study in order to keep focus on the most likely globally-cosmopolitan Aedes vectors, we note these species could be important in regional patterns of establishment. These species lack the globally comprehensive datasets that dominant arbovirus-vectoring Aedes species have, and require future attention by similarly-dedicated researchers.
Due to the potentially transient nature of the New World distribution of Zika virus, our model uses presence and 1000 randomly selected pseudo-absence points from the Eurasian, African, and Australian regions where the virus is established. We used the WorldClim data set BIOCLIM at 2.5 arcminute resolution, an aggregated dataset across values from 1950 to 2000, to provide all but one of our climate variables. [25] The BIOCLIM features 19 variables (BIO1-BIO19) that summarize trends and extremes in temperature and precipitation at a global scale. Given the relevance of the normalized difference vegetation index (NDVI) in previous studies of dengue and as a predictor of vector mosquito distributions, [26] we downloaded monthly average NDVI layers for each month in 2014 from the NASA Earth Observations TERRA/MODIS data portal, [27] at a resolution of 0.25 degrees to maintain compatibility with the BIOCLIM layers (0.25 degrees is equivalent to 15 arcminutes). The twelve monthly layers were averaged to provide a single mean NDVI layer. Due to the absence of NDVI data at the necessary resolution associated with many of the historical records (especially prior to 1992), the use of a recent mean NDVI layer was deemed the most pragmatic method of including vegetation in our models. We also make the simplifying assumption that areas of prior presence correspond to areas of current presence, an assumption that allows the use of current NDVI and is relatively standard for the niche modeling literature.
Species distribution models were executed using the BIOMOD2 package in R 3.1.1, which produces ensemble species distribution models using ten different methods: general linear models (GLM), general boosted models or boosted regression trees (GBM), general additive models (GAM), classification tree analysis (CTA), artificial neural networks (ANN), surface range envelope (SRE), flexible discriminant analysis (FDA), multiple adaptive regression splines (MARS), random forests (RF), and maximum entropy (MAXENT). [28] The BIOMOD algorithm runs a series of distribution models using training data, each of which is subsequently weighted and stacked across methods based on relative predictive performance with test data. As Thuiller et al. note, if a single modeling method is consistently most accurate, use of that method should be favored over ensemble approaches, [28] but in our study model performance varied, making ensemble approaches informed by degree-of-belief in a given model the most powerful option available. With recent publication of two Zika niche modeling papers using MAXENT and boosted regression trees, respectively, [29–30] differences between these two modeling methods may be responsible for differences in predictions–an issue that makes ensemble models particularly robust to idiosyncrasies of any individual methods.
Models were run individually for Zika (ZIKV), dengue (DENV), Ae. aegypti, Ae. albopictus, and Ae. africanus. For Zika, models trained on Old World environmental data (from Europe, Africa, Asia and Australia) were used to establish the potential distribution of the virus in the Americas under climatic conditions captured by WorldClim data, which are an aggregate of data between 1950 and 2000 (appropriately matching the date range of historical Zika occurrence data), and represent an expected range of variability that does not incorporate anomalous events like 2015 El Niño Southern Oscillation. Extrapolation between continents is a procedure with the potential for error: if novel environments exist in the New World with incomparable covariance structure between climate variables, predictive accuracy is likely to decline. While using only Old World data could potentially bias our models towards a subset of the niche, this can be readily tested for, by comparing models that include or exclude South American occurrence data.
To address colinearity in the environmental variable set, we produced a correlation matrix for our 20 variables, and identified each pair with a correlation coefficient > 0.8. For each species, we ran a single ensemble model with all ten methods and averaged the variable importance for our 20 predictors across the methods (S2–S6 Tables). In each pair we identified the variable with the greater contribution, and we produced species-specific reduced variable sets used in the final published models by eliminating any covariates that universally performed more poorly than their pair-mate. Based on this criterion, we excluded the following variables for each species to reduce colinearity:
The AUC of every model run with reduced variable sets is presented in S7 Table. We found no significant correlation between NDVI and any individual BIOCLIM variable, so NDVI was included in every model of current distributions. We ran five iterations of each reduced variable set model and eliminated any prediction methods from the ensemble with an AUC of lower than 0.95, so that the final model had only included the best predicting models. This greatly limited the models available for ZIKV and DENV, so a cutoff of 0.9 was applied in those cases, to keep the ensemble approach constant across datasets. The final models were run with the following methods with ten iterations using an 80/20 training-test split in the final presentation:
The importance of variables of the reduced model set for each are presented in S8–S13 Tables, and the final ensemble models are projected from the BIOMOD output in S1–S5 Figs.
To assess the transferability of our Zika model across environmental space, we conducted a geographic cross validation (GCV) between African and Asian datasets (an analysis we did not repeat for Aedes species or dengue, given the far greater sample size and geographic coverage of those species, and the publication of more intensive niche modeling efforts by experts for those systems). While under normal circumstances, a model would be trained on New World data and projected onto the Old World to cross-validate results, the lack of data prior to the current outbreak makes such a direct comparison infeasible. However, given the evidence for separate Asian and African strains, a cross-validation between the two was supported, and models trained on those two continents were projected globally to test the performance of the model across geographic regions, and evaluate how sensitive our projections in the Americas are to the environmental covariates sampled. The clustering of points in western India narrows the environmental range sampled by presences, potentially limiting the apparent transferability of the Asian sub-model. In contrast, the African sub-model performs well in new regions, and corresponds well to the global model.
The potential contribution of climate change to Zika’s current expansion, and the outer bounds of transmission under future expansion, are largely unaddressed. While these have not been the subject of any concerted speculation, Shapshak et al. [31] point out that the majority of arboviruses are potentially implicated in the climate change-driven expansion of global disease burden, with a shared set of drivers that quite probably extends to Zika as well. Consequently this analysis serves two purposes; to address the potential expansion and thereby assist public health planning, and to test whether even a liberal post-climate-change interpretation of range margins matches the predictions of Messina et al. [29] and Samy et al. [30] that we consider limited in specificity and potentially over-predictive. To project the distribution of the species under a worst-case scenario for climate change, we reran each model with the previously chosen method and variable sets but excluded NDVI, as future values could not be simulated effectively. BIOCLIM forecasts were taken from WorldClim using the Hadley Centre Global Environmental Model v. 2 Earth System climate forecast (HadGEM2-ES) predictions for representative climate pathway 8.5 (RCP85), which, within that model, represents a worst-case scenario for carbon emissions and climate warming. [32] All five species’ models were retrained on current climate data and projected onto forecasts for the year 2050. While we could have also included milder climate change forecasts and scenarios in our analysis, public concern over the future spread of Zika make the worst case scenario the most relevant question of interest for public health research (and intermediate scenarios would fall between current ranges and the worst case scenario we project).
To compare the niche of dengue and Zika and thereby address whether dengue models can be appropriately used to forecast the Zika pandemic, we used the R package ecospat, which uses principal component analysis to define the position of species’ ecological niche relative to background environmental variation. [33–34] The ecospat analysis was run using the full 64 point database and the full extent of global environmental data, because, while the niche of Zika in the Americas is uncertain, dengue is well established, and the analysis was most appropriately done with global coverage. Niche similarity tests were run with 500 iterations and using the entire set of 20 environmental variables (BIOCLIM + NDVI).
Our study is centered on the assumption that incorrect predictions at the country level can have drastic consequences for the misinterpretation of science. As a final precautionary analysis, we supplemented the data published in the Messina et al. study [30] to our own for a final re-analysis. Broennimann & Guisan [35] recommend the pooling of data from native and invasive ranges for ecological niche modeling during the course of a biological invasion, an approach we adopt in this final analysis. The Messina data is heavily clustered in Brazil, with a high degree of aggregation, and especially compared against our less-aggregated, smaller dataset this made the combination of datasets potentially inaccurate. To address this problem, the 390 pooled points were reduced down to 242 points using the package spThin, [15] with a 40km buffer between points (the width of an average grid cell for our environmental data). Models were rerun using the same variable and model set as for the primary Zika model and the results of the analysis are included in the supplementary information as S6 Fig and, with a threshold applied based on the true skill statistic, S7 Fig. The final model performs poorer than our main ensemble (weighted model: AUC = 0.970), and while it more appropriately predicts presences in southern Brazil, it does a far poorer job in the rest of the world, once again most likely due to the relative balance of points even after thinning the dataset.
Our final Zika model combines seven methods with a variable set chosen from bioclimatic variables and a vegetation index to minimize predictor covariance. The ensemble model performs very well (AUC = 0.993; Fig 1), to a degree that resembles overfitting but is in fact driven by the strength of the ensemble modeling approach (which preferentially weights the best models across iterations, minimizing the error associated with any given high-performing iteration). The model strongly matches most occurrences including the hotspots of Brazilian microcephaly. It also predicts additional regions where Zika is so far unrecorded, but where further inquiry may be desired (in particular, Southern Sudan and the northern coast of Australia). Our model indicates that certain occurrences, like the 1954 report from Egypt and almost all North American cases, are likely outside the stable transmission niche (i.e., persistent over time) of the virus (sensu [36]). Moreover, we note that visual presentation of cases–or, of ecological niche models–at the country level may make the range of the virus appear far larger than our models suggest (see Fig 1).
Given the public health crisis posed by Zika, and the potential costs associated with underpredicting the extent of the current outbreak, we pay special attention to evaluating the sensitivity of our models to variations in our preliminary dataset. Historical geographical data on cases in the Americas are lacking, given the recent introduction of the virus, and the routes and drivers of transmission involved in that outbreak are uncertain, preventing meaningful cross-validation of models of the current outbreak with our Old World model. However, it is worth noting that recent phylogenetic work suggests a deep phylogenetic division between African and Asian strains, the latter of which as a monophyletic group include the entire radiation through French Polynesia into current outbreak areas; [11, 37] to address the potential evidence that African and Asian strains of the virus may be ecologically distinct, we present models trained on each continent and projected globally as a basic sensitivity analysis (Fig 2).
The two models cross-validate weakly compared to the performance of the global model; driven by both the 50% reduction in sample size and the higher degree of aggregation of Asian occurrences, the two projected distributions are dramatically different. Despite the over-prediction of the Asian model in Africa and the possible overfitting of the African model, we emphasize that neither extreme scenario predicts any substantially greater range in North America than our main ensemble model. Moreover, our Asian model underpredicts but does predict two major hotspots of occurrence in Brazil, the Ceara/Rio Grande do Norte region and Roraima, both of which spatially correspond to hotspots of Zika according to the recent Faria et al. publication in Science, [11] adding further support to the model. Finally, despite low transferability between continents, both sub-models are well matched by our aggregated model in their native range, further supporting the accuracy and predictive power of our global projection.
Recently published work by Bogoch et al. [38] uses an ecological niche model for dengue as a proxy for the potential full distribution of ZIKV in the Americas, presenting findings in terms of potential seasonal vs. full-year transmission zones. While that approach has been effectively validated for dengue transmission in mosquitoes, using a model of one disease to represent the potential distribution of another emerging pathogen is only a placeholder, and is particularly concerning given the lack of evidence in our models that ZIKV and dengue have a similar niche breadth. [39] Comparing our niche models for dengue and ZIKV reveals that the two niches are significantly different (Schoener’s D = 0.176; p < 0.01; Fig 3). While the two occupy a similar region of global climate space, Zika is more strictly tropical than dengue, occupying regions with higher diurnal temperature fluctuations and seasonality of precipitation (Fig 3A).
Projecting niche models to the year 2050 suggests that expansion of Zika’s niche outside the tropics is an unlikely scenario, independent of vector availability (Fig 4). However, significant westward expansion in South America and eastward expansion in Africa implies that Zika may continue to emerge in the tropics. Moreover, our future projections for dengue (which strongly agree with previously published ones [40]) show an expansion out of the tropics that is not shared with Zika (Fig 4). These results call into question the applicability of dengue niche models used to project a significant future range for Zika in North America. [38]
Finally, we add a last layer of validation in the form of an analysis aggregating our and Messina et al.’s data, and include the results of an updated ensemble model in Fig 5 (as well as S6 and S7). Even with spatial thinning, that updated model is still heavily biased in favor of the South American occurrence data, which it predicts excellently, compared to a weaker fit in Africa and Asia. That accompanying loss of specificity is partly responsible for a lower AUC than our main model (AUC = 0.970) and the low TSS-based threshold (271, from 0 to 1000) that produces the substantially-greater predicted range shown in S7 Fig. The model does predict the current outbreak more effectively than ours, in particular better encompassing the southern half of Brazil where a surprising number of cases are clustered. But those southward expansions are accompanied by far less expansion above the equator in the Americas, and once again with the exception of the southernmost tip of Florida, there is no substantial predicted range in the United States, even along the Gulf Coast. If model discrepancies are attributed to evolutionary change and not to differences in model methods and specificity, those evolutionary changes seem to have done little to expand the North American niche of the virus (S8 Fig).
Ecological niche modeling has become one of the most generalized and useful parts of the streamlined response process for emerging infections. Recently published ecological niche models for Zika using MAXENT [30] and boosted regression trees [29] have resulted in somewhat conflicting results. Samy et al., using data exclusively from the range of the current outbreak, project autochthonous transmission in the southeastern United States, and potentially throughout the U.S. following regional outbreaks introduced by travelers. Their analysis incorporates socioeconomic factors into prediction, a valuable extra dimension we did not incorporate into our analysis; but the prediction of regions throughout the United States and most of the European continent as suitable based on only these criteria (i.e. despite lacking available vectors) seems uninformative except for the prediction of sexual outbreaks. Samy et al., however, conclude: “In Western Europe, ZIKV transmission risk is enhanced by travel times and connectivity to known transmission areas; as such, isolated autochthonous cases may occur at least seasonally when competent vector species are present.” [30] Messina et al. have a similar finding, based on a primarily ecological approach applied to 323 occurrences mostly from the New World; they map out most countries in the world as highly suitable, including the United States, with the conclusion that 2.17 billion people live in countries within Zika’s potential expanse. [29] These studies, being contemporaneous, do not refer to each other, and their conflicting results could render Zika forecasts unclear to the media and policymakers.
Interpreting conflicts between these models and those published here requires acknowledging three fundamental problems. First, differences in virulence between American and Asian strains of the virus may have changed the range limits. The niche of the vector-borne disease is manifest in its transmission and prevalence in mosquitoes (as well as humans and reservoirs), and increases in virulence could change the threshold of habitat suitability manifest in range limits. Without comparative work using updated data in Samy et al. and Messina et al.’s papers, equal support exists for our differences being attributable to methodological discrepancies or to a difference between Asian and American strains. But in the preliminary analysis we present in the supplementary information, incorporating data from the New World does not substantially expand projections in the United States (though a greater region of Brazil is predicted); and we believe a combination of evolutionary shifts and methodological differences is likely the most parsimonious explanation for differing results.
Second, we acknowledge the untested possibility that Zika has been expanding in its range since discovery in the 1940s (though, the virus was soon recorded in Borneo and Vietnam in the 1950s [23]), which would also decrease both the accuracy of our models in that region, and their power in the New World compared to the models published in the other two studies. Testing that possibility using our data broken down by time periods would be strongly statistically biased by the non-random element of viral discovery in different tropical countries, a factor for which it would be nearly impossible to control. Phylogenetic evidence has placed the introduction in the Americas within the last decade [11], but the age of divergence between Zika and closely related viruses like Japanese and St. Louis Encephalitis Viruses is less certain. Improving phylogenetic evidence based on updated Old World genomes in the coming years is a far more appropriate methodology for testing different biogeographic theories within that region.
Third and finally, we acknowledge the possibility that dispersal limitations have changed between the Old and New World, in such a way that the present expansion of Zika is not the emergence of novel niche space but the manifestation of hidden plasticity. This possibility is troubling from a public health perspective: if Zika’s niche is simply more expansive than current data/models capture, its geographic expansion could progress much further than we predict. This problem is fundamental to all predictive models applied to biological invasions, but Broenniman & Guisan [35] suggest that combining data from the native and invasive range maximizes the utility of ENMs in these scenarios. In our combined model we find evidence for subtle differences, especially in South America, but our findings remain sound with respect to the boundaries of transmission in North America. In any niche modeling study, there is always the possibility for error by omission; but we find no evidence that this has occurred in our study.
The dynamics of arboviruses at the range margins of their vectors are complex. In the case of dengue, the distribution of the virus in the United States (and elsewhere in temperate regions) remains more constrained than the range of its vectors. Our paper tests and rejects the hypothesis that predictions of Zika will occupy the entire niche of Aedes populations in North America, disagreeing with the two recently published niche model studies. Our models imply a similar constraint on Zika transmission to that of dengue if not a more pronounced one, and owing to the complexities surrounding transmission dynamics at the edges of suitable ranges, [41] the potential existence of Zika in even the southernmost parts of Florida [42] may not sustain autochthonous Zika transmission indefinitely. Making more specific predictions within Florida can be done through ecological niche models, but is likely more appropriately achieved through conventional epidemiological models that explicitly model vector abundance, biting rates and phenology.
Our models find an ecological nonequivalence of Zika and dengue, and suggest that the niche of the virus in both Africa and Asia is far narrower than what other models project based on current outbreak data or based on knowledge of dengue’s spread. We reject our first hypothesis, but based on the occurrence of Zika cases outside our predicted suitable range for the virus, we cannot eliminate our second hypothesis that the 2016 Zika outbreak may be in ephemeral, rather than stable, parts of the Zika transmission niche due to episodic climatic conditions. Specifically, El Niño Southern Oscillation (ENSO) events drive outbreaks of dengue in the Americas and in Southeast Asia, [43] and Paz et al. [12] have conjectured that the 2015 ENSO event could have contributed to the severity of the ZIKV outbreak in North and Central America (in response to Bogoch et al. [38]). While wind-dispersed mosquitoes carrying infections can be responsible for the introduction of diseases to new regions, [44] reported cases in the United States have all been contracted sexually or while traveling abroad to regions with endemic outbreaks, further supporting the tropical constraint hypothesis. However, in the second hypothesis scenario, the rapid expansion during the current outbreak beyond the boundaries of the stable transmission niche is unlikely to be followed by naturalization of the pathogen in the United States in the future, except perhaps in the southernmost tip of Florida. While ecological niche models relate occurrence to climate, drivers of disease may operate at the temporal scale of weather, and we suggest further analyses of a different methodology are necessary to confirm or reject the potential contribution of El Nino or anomalous storms to Zika’s expansion.
In the case of our third hypothesis, if alternative modeling efforts based on data from the Americas are evidence that the niche of the American strain of the virus has broadened, it is possible that mutations allowing increased virulence or changing transmission dynamics have occurred (and that weather events have not driven the severity of the current outbreak). From the results of our supplementary analysis using aggregated global data, we continue to treat the third hypothesis as a hypothesis for which there may be weak evidence. But we suggest it cannot be rejected or accepted confidently unless alternative hypotheses are eliminated and more evidence is collected–in particular, empirical data demonstrating or failing to find differences in transmission dynamics or virulence between the native Asian virus and its invasive descendant (rather than global comparisons and cross-validations of different ecological niche models).
Our models nevertheless suggest it could be premature to expect Zika naturalization as a widespread eventuality in North America, as other models have forecasted. Without more definitive information on the basic biology of Zika, however, the confidence with which niche models can forecast pandemics is limited. In particular, we also draw attention to recent evidence suggesting Zika persistence may depend on wildlife reservoirs in addition to human hosts and mosquitoes. Primates have been suggested as the primary candidate clade because the Zika flavivirus was first isolated in a rhesus macaque in the Zika Forest in Uganda. But as rhesus macaques do not occur on the African continent, and were captive there for inoculation experiments, the primate reservoir hypothesis remains unsupported. A 2015 case of an Australian presumed to have contracted Zika from a monkey bite while traveling in Indonesia, however, indicates that primates may transmit the virus directly. [9] Additionally, antibodies against Zika have been observed in several rodent and livestock species in Pakistan, [45] as well as several large mammal species, including orangutans, zebras, and elephants. [46] The potential for any North American wildlife species to play host to Zika is, at the present time, entirely unknown, and the emergence of novel amplification hosts (which may allow the virus to proliferate above the host density threshold in vectors in regions otherwise unsuitable for sustained transmission) could potentially expand the suitable range margins of Zika infection on a global scale.
From the results of our model we find strong evidence for the hypothesis that the global threat of a specifically vector-borne Zika pandemic, though devastating, may be most acute in the tropics; and we find that the evidence of future North American transmission in the literature is not unequivocal. However, we concur with the scientific majority that sexual transmission of Zika infections may still facilitate a significant outbreak in the United States and other previously unsuitable regions, particularly under evolutionary processes that select for the most directly transmissible strains of pathogens. [47] A case of sexual transmission in Texas has been suspected in the 2016 outbreak, and two previous reports of likely sexual transmission of ZIKV occurred in 2011 and 2015. [5, 48] Even if the Zika cases in the United States represent a rare spillover outside of the mosquito-borne viral niche, sexual transmission could create a new, unbounded niche in which the virus could spread. We draw attention to the potential parallels with simian and human immunodeficiency virus (SIV/HIV), for which a sexually transmitted pandemic has overshadowed the zoonotic origin of the disease. [49] With Zika’s asymptomatic presentation and the overall confusion surrounding its basic biology and transmission modes, we caution that its potential for severe sexually-transmitted outbreaks cannot be overlooked in the coming months.
To address the broader community of modelers and ecologists involved in the Zika intervention, we conclude with a final cautionary note. The consequences of under-predicting an outbreak’s potential distribution are obvious and our results are phrased cautiously as a result. But there are also economic and social consequences to over-predicting the potential distribution, especially in the United States. The response to Zika is necessarily political and consequently involves the division of resources between domestic preparedness and international relief; while new tools are being developed to help allocate funds efficiently based on epidemiological principles (we particularly highlight the work of Alfaro-Murillo et al. [50]), global overestimation of the virus’s trajectory could vastly reduce the power of those methods.
Models like those of Messina et al. and Samy et al. that predict substantial Zika expansion in the United States, and in the case of the former suggest Zika could threaten up to 2.17 billion people, contribute (independent of accuracy) to fear of an American pandemic. This prediction necessarily diverts funding away from relief efforts in Brazil and other affected countries in Latin America, increasing the probability of traveler infections feeding sexual outbreaks in the U.S.; and further reduces the credibility and impact of the American foreign response to Zika by mobilizing potentially-unnecessary domestic responses. At the time of writing, the Zika Vector Control Act passed by the U.S. House of Representatives weakens permit requirements for spraying pesticides near bodies of water without reallocating any funding for Zika interventions; and preventative efforts in New York City alone will cost $21 million to trap mosquitoes and hire epidemiological experts, with other cities outside our predicted range investing in preparation and vector control to similar degrees. Voices of scientific authority contributing to fear in the United States can substantially impact the political response to Zika, and it serves future modeling efforts to be as accurate, cautious, and objective as possible in the information and statistics that underpin media and policy conversations. But even more importantly, scientific teams with different approaches and data must work collaboratively to interpret the discrepancies between their results and to build an unbiased scientific consensus that is accessible to the public.
|
10.1371/journal.pntd.0001456 | Negative Regulation of Schistosoma japonicum Egg-Induced Liver Fibrosis by Natural Killer Cells | The role of natural killer (NK) cells in infection-induced liver fibrosis remains obscure. In this study, we elucidated the effect of NK cells on Schistosoma japonicum (S. japonicum) egg-induced liver fibrosis. Liver fibrosis was induced by infecting C57BL/6 mice with 18–20 cercariae of S. japonicum. Anti-ASGM1 antibody was used to deplete NK cells. Toll-like receptor 3 ligand, polyinosinic-polycytidylic acid (poly I∶C) was used to enhance the activation of NK cells. Results showed that NK cells were accumulated and activated after S. japonicum infection, as evidenced by the elevation of CD69 expression and IFN-γ production. Depletion of NK cells markedly enhanced S. japonicum egg-induced liver fibrosis. Administration of poly I∶C further activated NK cells to produce IFN-γ and attenuated S. japonicum egg-induced liver fibrosis. The observed protective effect of poly I∶C on liver fibrosis was diminished through depletion of NK cells. Disruption of IFN-γ gene enhanced liver fibrosis and partially abolished the suppression of liver fibrosis by poly I∶C. Moreover, expression of retinoic acid early inducible 1 (RAE 1), the NKG2D ligand, was detectable at high levels on activated hepatic stellate cells derived from S. japonicum-infected mice, which made them more susceptible to hepatic NK cell killing. In conclusion, our findings suggest that the activated NK cells in the liver after S. japonicum infection negatively regulate egg-induced liver fibrosis via producing IFN-γ, and killing activated stellate cells.
| Schistosomiasis continues to be a major public health problem in the developing world. Parasite egg-induced liver fibrosis is the principal cause of morbidity and mortality in human infected with schistosoma. Thus, elucidating the mechanisms that restrict tissue fibrosis may lead to more effective strategies for immunological intervention in this and a variety of chronic diseases. NK cells have been demonstrated to play an important role in suppressing carbon tetrachloride (CCl4)-induced liver fibrosis. However, little is known about the role of NK cells in an infection-based model of fibrosis. In the current study, we determined, for the first time, the role of NK cells in S. japonicum egg-induced liver fibrosis. Our findings suggest that the activated NK cells in the liver after S. japonicum infection negatively regulate egg-induced liver fibrosis via producing IFN-γ, and killing activated stellate cells. These results further our understanding of the innate immune cells that regulate the development of S. japonicum-induced fibrosis and aid in the development of potential strategies to enhance immunity against this and other chronic inflammatory diseases of the liver where fibrosis is a common feature.
| The development of fibrous tissue is a common scarring response to chronic and debilitating illnesses, including several autoimmune, allergic, and infectious diseases [1]. In schistosomiasis, egg-induced hepatic fibrosis can lead to portal hypertension, which causes much of the morbidity and mortality associated with this disease. Murine models of schistosome infection indicate that most pathology is attributable to type 2 T-helper cell (Th2)-mediated granulomatous response against schistosome eggs and egg antigens [2]. These Th2 cytokines, in particular IL-13 and its receptor (IL-13Rα), play a pivotal role in the development of granuloma-associated fibrosis [3], [4]. In addition, anti-IL-4 or anti-IL-5 treatment markedly decreases schistosomiasis-induced fibrosis [5], [6]. IFN-γ has major roles in downregulating Th2 cell expansion [7] and selectively promoting Th1 cell differentiation [8]. Moreover, IFN-γ has been shown to have potent antifibrogenic effect [9]–[11], and its antifibrogenic effect is believed to be mediated via inhibiting hepatic stellate cell (HSC) activation and TGF-β signaling [12], [13]. In Schistosoma mansoni (S. mansoni)-infected mice, IFN-γ treatment leads to significant reduction of hepatic fibrosis [14], [15].
Natural killer (NK) cells represent a major source of IFN-γ [16]–[18]. The antifibrogenic roles of these cells have been studied in two artificial models of liver fibrosis induced by a 3,5-diethoxycarbonyl-1,4-dihydrocollidine diet (DDC) and carbon tetrachloride (CCl4). The authors showed that hepatic NK cells killed activated HSCs via retinoic acid early inducible 1 (RAE1)/NKG2D dependent and tumor necrosis factor-related apoptosis-inducing ligand (TRAIL)-dependent mechanisms, thereby inhibited liver fibrosis [19]. Little is known about the roles of NK cells in an infection-based model of fibrosis. NK cells have been demonstrated within schistosome egg granulomas [20]. However, the roles of NK cells in the development and progression of schistosome egg-induced liver fibrosis remain unknown. Here, in S. japonicum-infection model, we found that hepatic NK cells markedly ameliorated egg-induced liver fibrosis. NK cells played their critical roles partly via an IFN-γ-dependent mechanism. We also presented evidence that NK cells might kill activated HSCs to attenuate fibrosis via NKG2D-RAE1 recognition.
All mice were maintained in a specific pathogen-free microenvironment, and received care in compliance with the guidelines outlined in the Guide for the Care and Use of Laboratory Animals. All work was conducted with the approval of the Anhui Experimental Animal Training Base.
Six-week old female C57BL/6 mice were purchased from Experimental Animal Center, Chinese Science Academy (Shanghai, China). Interferon-gamma knockout (IFN-γ−/−) mice were purchased from model animal research central of Nanjing University. For infection, mice were anesthetized and percutaneously exposed to 18–20 cercariae of S. japonicum (strain from Jiangxi Province, China) that were obtained from infected Oncomelania hupensis snails.
To deplete NK cells, mice were injected intravenously with anti-ASGM1 antibody (Ab) (Wako Co., Tokyo, Japan). After 24 hours, depletion of NK cells was confirmed by flow cytometry. To chronically deplete NK cells, mice were treated with anti-ASGM1 Ab every 5 days from week 5 after infection for 3 or 5 weeks.
Serum samples from individual mice were obtained at week 8 and week 10 post-infection. Liver injury was assessed by measuring serum alanine aminotransferase (ALT) activities using commercially available kit (Rong Sheng, Shanghai, China).
Poly I∶C (Sigma Chemical Co., St Louis, MO) was dissolved in the pyrogen-free saline. Mice were injected intraperitoneally with poly I∶C (0.5 µg/g) every 3 days since week 5 post-infection. Control infected mice received saline injection.
Liver tissues were fixed in 10% buffered formalin and embedded in paraffin. Tissue sections were affixed to sides, deparaffinized, and stained with Masson trichrome for collagen deposition. Immunostaining for α-smooth muscle actin (α-SMA) was performed using a monoclonal α-SMA primary Ab (clone 1A4; Dako, Carpinteria, CA), and a horseradish peroxidase-labeled secondary Ab. Six to ten images per mouse liver were photographed using an inverted microscope (Nikon 80I, Japan) and then digitized and analyzed on Image-Pro Plus software.
Liver tissues were homogenized in RIPA lysis buffer (Solarbio, China) added with 1 mM PMSF. Western blot analyses were performed as described previously [21]. Briefly, proteins were separated by 10% SDS-polyacrylamide gel electrophoresis, transferred onto nitrocellulose membranes, and blotted with primary Abs. After wash, membranes were incubated with horseradish peroxidase-conjugated secondary Abs. Specific binding was visualized by ECL reaction (Pierce).
HSCs were isolated using two-step collagenase perfusion method as described [22]. The viability of the isolated cells was determined to be 98% using trypan blue staining. The purity of the cells was assessed visually by light microscopy examination of typical lipid droplet appearance, and vitamin A autofluorescence was more than 90%.
Liver mononuclear cells (MNCs) were isolated essentially as described previously [17]. To culture liver MNCs in vitro, cell pellets were resuspended in RPMI 1640 medium containing penicillin, streptomycin, and 10% FBS, and then plated onto 24-well plates at a density of 5×106 cells per well in 1 mL culture medium with or without poly I∶C treatment for 48 hours. The supernatants were collected for IFN-γ measurement by ELISA (R&D system).
The RNA of HSCs was extracted using RNAprep pure Micro Kit (Tiangen Biotech CO., LTD.). Quantitative PCR was performed using a sequence detector (ABI-Prism 7500; Applied Biosystems) and a SYBR Premix Ex Taq (Takara), according to the manufacturer's instructions. Primers for RAE1α, RAE1β, RAE1γ, RAE1δ, RAE1ε, H60, and Mult1 were used as reported previously [19]. Other primer sequences used are as follows: β-actin, sense, 5′-TGG AAT CCT GTG GCA TCC ATG AAA-3′, antisense, 5′-TAA AAC GCA GCT CAGTAA CAG TCC G-3′; IL-4, sense, 5′-TCA TGG AGC TGC AGA GAC TCT T-3′, antisense, 5′-CAT TCA TGG TGC AGC TTA TCG A-3′; IL-5, sense, 5′-CTC TGT TGA CAA GCA ATG AGA CG-3′, antisense, 5′-TCT TCA GTA TGT CTA GCC CCT G-3′; IL-13, sense, 5′- CCT GGC TCT TGC TTG CCT T-3′, antisense, 5′-GGT CTT GTG TGA TGT TGC TCA-3′. For analysis, all expression levels of target genes were normalized to the housekeeping gene β-actin (ΔCt). Gene expression values were then calculated based on the ΔΔCt method as mentioned previously [17].
Liver NK cells were separated by positive magnetic cell sorting using anti-DX5 mAb according to the manufacturer's protocol (Miltenyi Biotec, Auburn, CA) from S. japonicum-infected mice. Approximately 90% of the magnetic cell sorting-purified cells were DX5+.
The monoclonal Abs used for flow cytometry in this study included Cy5-anti-CD3e, phycoerythrin (PE)-anti-CD69, fluorescein isothiocyanate (FITC)-anti-NK1.1, PE-anti-IFN-γ, PE Rat immunoglobulin G2a (IgG2a) isotype control (BD PharMingen). For intracellular cytokine staining, liver MNCs were incubated for 4 hours in the presence of ionomycin (1 µg/ml, Sigma), PMA (20 ng/ml, Sigma), and brefeldin A (1 µg/ml, BD Pharmingen). After staining with FITC-anti-NK1.1 mAb and Cy-5-anti-CD3e mAb, cells were fixed, permeabilized, and stained with PE-anti-IFN-γ mAb using a Cytofix/Cytoperm plus kit (BD PharMingen). The stained cells were analyzed using a flow cytometer (FACScalibur; Becton Dickinson, Franklin Lakes, NJ), and data were analyzed with WinMDI2.9 software.
NK cell-mediated cytotoxicity against primary HSCs was measured by ToxiLight BioAssay Kit (Lonza, Rockland, Inc). Briefly, freshly isolated HSCs were plated onto round-bottom, 96-well plates. NK cells were subsequently added as effector cells. The release of adenylate kinase from damaged cells was measured from the culture medium by using the ToxiLight kit. Specific lysis was calculated according to the formula [(test release−spontaneous release)/(maximum release−spontaneous release)]×100.
Data were analysed using SPSS (v11.0) and GraphPad Prism (v5). To compare values obtained from multiple groups, 1-factor analysis of variance (ANOVA) was used, followed by Tukey's post hoc test. To compare values obtained from two groups, the Student t test was performed. All data were shown as mean ± standard error of the mean (SEM). P value≤0.05 was considered to be statistically significant.
To investigate whether NK cells were involved in S. japonicum infection-induced liver fibrosis, we first determined the activation of NK cells in the liver post-infection. As shown in Figure 1A, the percentage of NK cells among hepatic MNCs significantly increased at week 3 post-infection, and then diminished to baseline and stabilized between the forth and the tenth week of infection. The absolute number of hepatic NK cells increased dramatically during infection, which peaked at week 3 post-infection (Figure 1B). Furthermore, the percentage of CD69 positive NK cells was significantly increased from week 3 to 10 (Figure 1C).
IFN-γ production of NK cells was also measured. As shown in Figure 1D, at week 6 and week 8 post-infection, the number of NK cells (CD3-NK1.1+) stained positively for IFN-γ was increased. Since NKT cells can also produce IFN-γ, we examined IFN-γ production of these cells. The increase in the number of NKT cells (CD3+NK1.1+) stained positively for IFN-γ after infection was very slight. Moreover, IFN-γ secretion by hepatic MNCs from anti-ASGM1 Ab pretreated mice was strongly reduced compared to that from control IgG pretreated mice (Figure 1 E). These results suggested that IFN-γ was mainly produced by liver NK cells.
To investigate the role of NK cells in S. japonicum egg-induced liver fibrosis, anti-ASGM1 Ab was used to deplete NK cells for 3 or 5 weeks. The depletion effect was confirmed by flow cytometry (Figure 2A). At week 8 and week 10 post-infection, the liver injury was evaluated by examining serum ALT activities. As shown in Figure 2B, depletion of NK cells did not significantly affect ALT levels. Meanwhile, liver fibrosis was monitored by Masson trichrome staining for collagen deposition (blue staining) and immunohistochemical staining for HSC activation (α-SMA-positive staining). Results showed that depletion of NK cells significantly enhanced S. japonicum egg-induced collagen deposition (Figure 2C) and elevation of α-SMA+ cells (Figure 2D). The inhibitory effect of NK cells on HSC activation was also confirmed by Western blot analyses. As shown in Figure 2E, α-SMA protein expression was increased in anti-ASGM1-treated mice compared with IgG-treated mice.
Poly I∶C could activate NK cells and induce accumulation of NK cells in the liver but did not induce remarkable elevation of ALT in B6 mice [17], [21]. Therefore, we treated infected mice with poly I∶C to study the effect of NK cell activation on liver fibrosis. First, the effect of poly I∶C injection on NK cell activation was examined. As shown in Figure 3(A–C), the percentage and absolute number of NK cells in the liver post-infection were markedly elevated in poly I∶C-treated mice compared with those in saline-treated mice. And poly I∶C injection also increased CD69 expression on NK cells (Figure 3D). Notably, the percentage of IFN-γ positive NK cells was notably increased in poly I∶C-treated group (Figure 3E). Further analyses of IFN-γ concentration in the culture supernatants of hepatic MNCs derived from S. japonicum infected-mice showed that stimulation with poly I∶C promoted hepatic MNCs to produce IFN-γ, but not MNCs without NK cells, which further suggested that NK cells were mainly responsible for IFN-γ production (Figure 3F).
Next, the role of poly I∶C on S. japonicum egg-induced fibrosis was examined. As shown in Figure 4, poly I∶C treatment significantly reduced S. japonicum egg-induced collagen deposition and HSC activation (α-SMA+ cells). However, the inhibitory effect of poly I∶C was abolished in anti-ASGM1-treated mice, which suggested that poly I∶C suppression of liver fibrosis was dependent on NK cells.
To examine whether NK cells negatively regulated S. japonicum egg-induced liver fibrosis via production of IFN-γ, we compared liver fibrosis between wild-type and IFN-γ−/− mice. As shown in Figure 5A, S. japonicum egg-induced liver fibrosis in IFN-γ−/− mice was much more severe than that in wild-type mice (14.65%±4.21 versus 8.76%±1.83, P = 0.04). Consonantly, there were more α-SMA positive HSCs in IFN-γ−/− mice than those in wild-type mice (10.60%±2.85 versus 6.10%±1.24, P = 0.02) (Figure 5B). These findings suggested that IFN-γ played negative roles in egg-induced liver fibrosis. Moreover, IFN-γ was also involved in poly I∶C-mediated suppression of liver fibrosis. Results showed that the suppression of collagen deposition and HSC activation by poly I∶C in IFN-γ−/− mice was reduced compared with that in wild-type mice (Figure 5A, B). Furthermore, we examined the Th2 cytokine expression in the liver of IFN-γ−/− mice. As shown in Figure 5C, IL-4, IL-5 and IL-13 mRNA expression was not different in IFN-γ−/− mice compared with wild mice.
Because activated HSCs were more susceptible to hepatic NK cell killing via RAE1/NKG2D recognition [19], [22], we attempted to determine whether NKG2D-ligand interactions were also involved in the suppression of S. japonicum egg-induced fibrosis by NK cells. First, the identity of activated HSCs isolated from S. japonicum-infected mice was confirmed by the expression of α-SMA (Figure 6A). Next, the expression of NK receptor activating ligands including RAE1, histocompatibility 60 (H60), and UL16-bingding protein like transcript 1(Mult-1) on HSCs was examined. As shown in Figure 6B, the expression of RAE1α, β, γ, and ε was dramatically increased on these activated HSCs at week 6 and week 8 post-infection, while the expression of RAE1δ, H60, and Mult-1 was not induced. To further determine whether NK cells in S. japonicum-infected mice kill activated HSCs, in vitro cell-mediated cytotoxicity experiments were performed. As shown in Figure 6C, liver NK cells produced about 30% cytotoxicity against activated HSCs from S. japonicum-infected mice, but less than 10% cytotoxicity against quiescent HSCs from normal mice. Furthermore, NKG2D blocking mAb diminished the cytotoxicity of NK cells against activated HSCs (Figure 6D). These data suggested that NK cells might inhibit S. japonicum egg-induced liver fibrosis via killing activated HSCs.
Infection with the helminth parasite schistosoma accounts for a significant portion of liver fibrosis cases reported each year in humans. Both T cells and B cells have been implicated in regulating S. mansoni egg-induced liver fibrosis. S. mansoni egg deposition leads to the development of Th2 response, which promotes the development of liver fibrosis [4], [23], [24]. B cells promote Th2-type response to the S. mansoni eggs, thus B-cell-deficient mice display an increased hepatic fibrosis at 10 weeks post-infection [25], [26]. NK cells are an important population of innate immune cells in the liver, comprising 30%–40% and 10%–20% of total intrahepatic lymphocytes in humans and mice, respectively [27], [28]. However, the roles of NK cells in schistosoma egg-induced liver fibrosis remain obscure. Treatment with anti-NK1.1 Abs to deplete NK1.1+ cells, which include NK cells (CD3-NK1.1+) and NKT (CD3+NK1.1+) cells, enhanced S. mansoni-induced liver fibrosis, thus suggesting that NK cells may be involved in negative regulation of schistosoma egg-induced liver fibrosis [29].
In this study, we provided several lines of evidence suggesting that NK cells negatively regulated S. japonicum-induced liver fibrosis. First, the absolute number of hepatic NK cells dramatically increased in the liver after S. japonicum infection. And these NK cells were activated as demonstrated by the upregulation of activation marker CD69 and production of IFN-γ (Figure 1). Second, depletion of NK cells using anti-ASGM1 Ab accelerated liver fibrosis (Figure 2). Third, injection of poly I∶C enhanced the accumulation of NK cells in the liver and activated NK cells to secrete IFN-γ (Figure 4), attenuating S. japonicum egg-induced liver fibrosis (Figure 3). The change in liver fibrosis was not due to differences in the intensity of infection or egg burden because worm pairs, total worms, eggs entrapped in the livers, and granuloma size of all groups in an experiment were identical (data not shown).
Currently, the underlying mechanisms by which NK cells are activated after S. japonicum infection are not clear. Our results showed that NK cells were activated before S. japonicum worms lay eggs (about 24 days post-infection), because the expression of CD69 on NK cells had been upregulated 3 weeks post-infection. The upregulation of CD69 was enhanced after worms lay eggs (Figure 1). These data suggest that the antigens or secretion of both worms and eggs may be involved in the activation of NK cells. Although NK cells express many activating receptors and toll-like receptors [16], [30], which lead to direct NK cell activation when encountering their ligands. Activation of NK cells by most pathogens seems to be indirect and results from signals provided by accessory cells, such as monocytes, macrophages and dendritic cells [31], [32]. Further studies are needed to determine whether NK cell activation during S. japonicum infection requires the presence of accessory cells and how these cells interact.
To investigate the underlying mechanisms by which NK1.1+ cells inhibited S. mansoni-induced liver fibrosis, Asseman et al. examined hepatic mRNA expression of IFN-γ as well as serum IFN-γ levels. They did not observe variation between control Ab-treated mice and anti-NKl.l Ab-treated mice. So they did not perform other experiments to study the role of IFN-γ in S. mansoni-induced liver fibrosis [29]. We also examined serum level of IFN-γ in S. japonicum-infected mice, but the level was too low to detect. Then we examined IFN-γ production from their hepatic mononuclear cell cultures, and found that IFN-γ secretion by hepatic MNCs from anti-ASGM1 Ab pretreated mice was strongly reduced compared to that from control IgG pretreated mice (Figure 1E). These data suggested that NK cells were the primary source of IFN-γ after S. japonicum infection. Furthermore, disruption of the IFN-γ gene accelerated S. japonicum egg-induced liver fibrosis and reduced poly I∶C-mediated suppression of liver fibrosis (Figure 5). Therefore, it is postulated that the production of IFN-γ by NK cells after S. japonicum infection was an important mechanism responsible for NK cell-mediated suppression of liver fibrosis. It is reported that IFN-γ also has high anti-fibrogenic activities and immune protection in human schistosomiasis [33], [34].
It is well known that IFN-γ promotes Th1 cell differentiation while dampening Th2 cell expansion [7], [8]. Thus there might be more Th2 cells in IFN-γ−/− mice. However, despite the absence of endogenous IFN-γ, Th2 cytokine (IL-4, IL-5, and IL-13) production in response to soluble egg antigen (SEA) stimulation by splenocytes was not significantly altered in IFN-γ−/− mice compared to wild mice [35], [36]. And we also found that IL-4, IL-5, and IL-13 mRNA levels in the liver of IFN-γ−/− mice were similar to those of wild mice (Figure 5C). So the increased fibrosis in IFN-γ−/− mice may be not because that IFN-γ−/− mice had higher levels of Th2 cytokines. According to other reports, IFN-γ suppression of liver fibrosis was mainly mediated through inducing HSC apoptosis and cell cycle arrest [37].
Although hepatic invariant NKT (iNKT) cells could also produce IFN-γ after transferring live S. mansoni eggs into the caecal vein of mice, the contribution of iNKT cells in IL-4 production by liver monocytes was total, whereas that in IFN-γ production was partial and minor [38]. Our data also demonstrated that the contribution of NKT cells in IFN-γ production was less than NK cells (Figure 1). In addition, hepatic NKT cells played a minor role in DDC-induced liver fibrosis, and inhibited CCl4-induced liver fibrosis in the early stage but not in the late stage of fibrosis [19], [39]. Thus, we speculated that the suppression of schistosoma egg-driven liver fibrosis by NK1.1 positive cells was mainly dependent on NK cells.
The results presented here also demonstrated that NKG2D-Rae1 interaction might be involved in NK cell-mediated HSC death, which contributed to NK cell suppression of S. japonicum egg-induced liver fibrosis. The balance in the expression of activating and inhibitory ligands will determine whether a cell becomes a target for NK cell-mediated killing [40], [41]. RAE1 was previously identified as an NK cell activating ligand to stimulate NK cytotoxicity [42]. Here we showed that the expression of RAE1α, β, γ, and ε was detected at much higher levels on activated HSCs from S. japonicum-infected mice compared with that on quiescent HSCs from uninfected mice (Figure 6A, B). Moreover, the expression of several NK receptor inhibitory ligands, including H2-D1, H2-D4, and Clec2D was unchanged on activated HSCs (data not shown). The interaction between RAE1 and NKG2D has been reported to be an important mechanism contributing to the cytotoxicity of NK cells against activated HSCs [19], [22]. Our data also demonstrated that liver NK cells could kill activated HSCs via NKG2D/RAE1-dependent mechanism (Figure 6C, D).
In addition, we examined the role of NK cells in IL-13 expression. Our results showed that depletion of NK cells did not affect mRNA expression of IL-13 in the liver (Data not shown). And the levels of IL-13 in the culture supernatants of hepatic MNCs from anti-ASGM1-treated mice and IgG-treated mice were similar (Data not shown).
In summary, our findings suggested that after S. japonicum infection, murine hepatic NK cells were activated and suppressed liver fibrosis mainly via production of IFN-γ and via killing activated HSCs. Because of the difficulty of acquiring liver samples from humans, the function of hepatic NK cells in patients with schistosomiasis is still unknown, and clinical researches only focus on the peripheral blood mononulcear cells (PBMCs). E. Speziali et al. found that NK cells were the major source of IFN-γ when in vitro stimulating PBMCs with S. mansoni antigens. However, in individuals over 70 years old, infection positive individuals had lower percentage of IFN-γ+ NK cells than negative controls [43]. Other clinical studies showed that the cytotoxicity activity of NK cells was reduced in children infected with Schistosoma haematobium and S. mansoni compared with healthy controls [44], [45]. Using a mouse model, our study first showed the suppression role of hepatic NK cells in S. japonicum egg-induced liver fibrosis. Therefore, stimulation of NK cell activity in schistosoma-infected patients especially in the elderly and children may have beneficial effect in suppressing liver fibrosis.
|
10.1371/journal.ppat.1006084 | Microbial Community Composition Impacts Pathogen Iron Availability during Polymicrobial Infection | Iron is an essential nutrient for bacterial pathogenesis, but in the host, iron is tightly sequestered, limiting its availability for bacterial growth. Although this is an important arm of host immunity, most studies examine how bacteria respond to iron restriction in laboratory rather than host settings, where the microbiome can potentially alter pathogen strategies for acquiring iron. One of the most important transcriptional regulators controlling bacterial iron homeostasis is Fur. Here we used a combination of RNA-seq and chromatin immunoprecipitation (ChIP)-seq to characterize the iron-restricted and Fur regulons of the biofilm-forming opportunistic pathogen Aggregatibacter actinomycetemcomitans. We discovered that iron restriction and Fur regulate 4% and 3.5% of the genome, respectively. While most genes in these regulons were related to iron uptake and metabolism, we found that Fur also directly regulates the biofilm-dispersing enzyme Dispersin B, allowing A. actinomycetemcomitans to escape from iron-scarce environments. We then leveraged these datasets to assess the availability of iron to A. actinomycetemcomitans in its primary infection sites, abscesses and the oral cavity. We found that A. actinomycetemcomitans is not restricted for iron in a murine abscess mono-infection, but becomes restricted for iron upon co-infection with the oral commensal Streptococcus gordonii. Furthermore, in the transition from health to disease in human gum infection, A. actinomycetemcomitans also becomes restricted for iron. These results suggest that host iron availability is heterogeneous and dependent on the infecting bacterial community.
| One of the most well-studied phenomena in microbiology is nutritional immunity, or how the host withholds nutrients such as iron to combat infection. As part of this, researchers have characterized how many pathogens respond to iron restriction. However, these studies are often conducted in laboratory media rather than the host. As a result, they overlook how the host environment, such as its microbiome, might alter pathogen behavior regarding iron during infection. To address this gap, we used an opportunistic pathogen that causes abscess and oral cavity infections. We defined how it responds to iron restriction in vitro and then used this data to assess its iron status in vivo. Our results show that in mono-culture abscesses the pathogen is not starved for iron but in co-culture abscesses and multispecies gum disease it is starved for iron. Therefore, host environments are not uniformly restricted for iron, and the microbiome can modulate iron availability to pathogens.
| In 1944, researchers Schade and Caroline found that by adding egg whites to cultures of various microorganisms, they could stunt microbial growth. Remarkably, they could only relieve this inhibition by supplementing the cultures with iron [1]. This was the first suggestion that iron, an essential nutrient for most microbes, is withheld by the host as a means to restrict infection. Since then, an entire field has emerged on the role of iron in host immunity [2–6].
Most bacteria require at least 0.3–4 μM iron to grow [2]. Pathogens therefore meet a significant hurdle when they infect the human body, where most iron is sequestered. Inside cells, iron is bound by proteins such as hemoglobin and ferritin, and outside cells, iron is bound by proteins such as transferrin and lactoferrin [7]. As a result, free ionic iron is present in human fluids at only 10−12 μM [3], over 11 orders of magnitude below that needed by bacteria. Thus, compromises to “nutritional immunity” can greatly increase vulnerability to infection. Demonstrating this, patients with hyperferremia (iron overload) are more prone to infection, and supplementation of mammals with iron has been shown to enhance the virulence of microbial pathogens [4].
In addition to host measures, acquiring iron is made challenging because under oxidizing conditions, most iron is ferric (in the 3+ oxidation state), which is insoluble and limited in bioavailability [8]. Bacteria can circumvent these barriers by either releasing siderophore molecules that bind ferric iron with high affinity and return it to the cell or by displaying surface receptors that bind iron directly [9]. These strategies, however, must also be tightly regulated because excess iron is toxic [10]. Many bacteria regulate iron acquisition with the Ferric Uptake Regulator (Fur), a transcriptional regulator that alters gene expression in response to intracellular iron levels [11]. In general, holo-Fur (Fur bound to Fe2+) represses gene expression, whereas apo-Fur (Fur not bound to Fe2+) de-represses gene expression. Besides iron transport, Fur can also control many other aspects of bacterial physiology and behavior, including toxin production, stress resistance, and biofilm formation [11].
While most nutritional immunity studies center on the interaction between host and pathogen, polymicrobial interactions between pathogens and the host microbiota can also affect the success of pathogens in acquiring iron, either through competition [12, 13] or cooperation [14–16]. One of the most common infectious diseases, periodontitis (gum disease), is also influenced by iron [17–19]. Periodontitis is caused by the multispecies biofilm community that inhabits the gingival crevice, the pocket between the gum and tooth [20]. This site is partially anaerobic [21] and bathed in a serum exudate known as crevicular fluid. Crevicular fluid contains many potential iron sources for microbial growth including transferrin, lactoferrin, hemoglobin, and inorganic iron [22]. In healthy serum, the average total iron concentration is ~1 mg/l, whereas in periodontal crevicular fluid it increases to ~5 mg/l [22]. Although this suggests that iron is more bioavailable in disease, one of the most highly upregulated microbial community functions in human periodontitis is iron transport [23], suggesting instead that iron is limited. The bioavailability of iron in periodontitis is therefore unclear and may depend on multiple factors.
An important member of oral microbial communities is the Gram-negative bacterium Aggregatibacter actinomycetemcomitans, an opportunistic pathogen associated with aggressive periodontitis. Notably, A. actinomycetemcomitans not only infects the oral cavity but can also spread throughout the body to cause abscesses [24]. In regards to iron sources, A. actinomycetemcomitans can use ferrous (Fe2+) iron, ferric (Fe3+) iron, and hemoglobin but not transferrin or lactoferrin [25–27]. Like most pathogens, the ability of A. actinomycetemcomitans to access one of these iron sources critically determines its virulence [28, 29]. Despite this, only a few iron acquisition systems in A. actinomycetemcomitans have been characterized. These systems directly bind and transport iron [25–27] since A. actinomycetemcomitans cannot make siderophores [25], and many of them are controlled by the A. actinomycetemcomitans Fur homolog. However, outside these systems only a few targets of Fur in A. actinomycetemcomitans are known [30].
Despite the clear importance of iron to bacterial infection, we know virtually nothing about how A. actinomycetemcomitans acquires it in vivo. How does A. actinomycetemcomitans surmount iron restriction at its different infection sites? How does A. actinomycetemcomitans interact with other bacteria regarding iron? Most importantly, how does A. actinomycetemcomitans respond to iron restriction in vivo to optimize its growth and virulence? With these questions in mind, our primary goal was to comprehensively characterize the A. actinomycetemcomitans iron-restricted and Fur regulons. We then leveraged these datasets to assess the A. actinomycetemcomitans iron response in both murine and human polymicrobial infections to determine the impact of microbial community composition and disease status on iron availability.
To characterize how A. actinomycetemcomitans responds to iron restriction, we performed RNA-seq to compare its gene expression between rich media and iron-chelated media (see Tables 1–2 in S1 Dataset for study design). Since A. actinomycetemcomitans does not grow planktonically, we grew it as colony biofilms. We also used anaerobic growth conditions since soluble Fe2+ predominates and can be readily chelated with 2,2’-dipyridyl to mimic iron-restricted conditions. In total we discovered 93 genes, representing 4% of the genome, which are differentially expressed in response to iron restriction (Table 3 in S1 Dataset). Of these, over twice as many were upregulated than downregulated, and upregulated genes generally had larger fold changes than downregulated genes (S1A Fig).
Most genes downregulated in iron-restricted conditions encoded components of the anaerobic electron transport chain (Fig 1), suggesting that iron restriction hinders A. actinomycetemcomitans anaerobic respiration. These downregulated genes included the Na+-translocating NADH-quinone reductase [31], menaquinone biosynthesis [32], anaerobic respiratory reductases [33] as well as two major steps in pyruvate metabolism (Fig 1). The fact that many of these downregulated gene products contain iron suggests that restricting cellular iron usage is central to the A. actinomycetemcomitans iron starvation response. Supporting this, one of the most downregulated gene products in response to iron restriction was the iron storage protein ferritin (Fig 1).
Genes upregulated in response to iron restriction primarily encoded iron transporters and receptors. These include 5 inner membrane ABC transporters (including the characterized Afu [34] and Afe [35] systems), TonB [36], and 5 outer membrane TonB-dependent receptors (Fig 1). Substrates of these systems included inorganic iron, ferric iron siderophores, hemoglobin, and transferrin. Transporters for metals besides iron–nickel and zinc–were also differentially expressed (Fig 1). Although the significance of these non-iron transporters is unclear, co-expression of nickel with iron transporters has been observed in other bacteria [37].
Other processes upregulated in response to iron restriction included cysteine export by the CydDC transporter [38] (Fig 1). CydDC is required for the assembly of cytochrome bd (CydAB), and as CydAB is the sole aerobic respiratory oxidase in A. actinomycetemcomitans [39], this suggests that aerobic respiration in A. actinomycetemcomitans is stimulated by iron restriction. This is noteworthy considering that most of the A. actinomycetemcomitans anaerobic respiratory reductases were repressed by iron restriction. A. actinomycetemcomitans therefore seems to equate iron restriction with the presence of oxygen. Supporting this, a transporter for glycerol-3-phosphate, a carbon source that can only be catabolized by respiration [40], and 2 thioredoxins, implicated in resisting oxidative stress from oxygen [41], were also more highly expressed under iron restriction (Fig 1).
The only differentially expressed non-coding genes were 2 small RNAs (sRNA) upregulated by iron restriction (Fig 1). One of these sRNA was homologous to an iron-regulated sRNA in H. influenzae [42], and like H. influenzae, this sRNA in A. actinomycetemcomitans was also predicted to target asparagine biosynthesis (P = 0.03, TargetRNA2). The other sRNA was predicted to target nitrite reductase (P = 0.05, TargetRNA2 [43]), one of the most downregulated gene products in response to iron restriction.
We next set out to characterize the A. actinomycetemcomitans Fur regulon with the goal of defining which iron-responsive genes were controlled by this transcriptional regulator. To do this, we used RNA-seq to compare gene expression of wild type A. actinomycetemcomitans to an isogenic Δfur mutant (see Tables 1–2 in S1 Dataset for study design). Overall, we found that 386 genes, representing over 17% of the genome, were differentially expressed when Fur is absent, with slightly more genes activated than repressed (Table 3 in S1 Dataset). As expected, the Fur regulon extensively overlapped the iron-restricted regulon, with Fur-repressed genes mostly overlapping genes upregulated in iron-restricted conditions, and Fur-activated genes mostly overlapping genes downregulated in iron-restricted conditions (Table 4 in S1 Dataset). In total, the Fur regulon encompassed 70 of 93 genes in the iron-restricted regulon, indicating that as expected, Fur has a critical role in controlling A. actinomycetemcomitans iron homeostasis. Supporting this, all of the iron uptake systems that were upregulated in iron-restricted conditions were repressed by Fur, and all of the anaerobic respiratory reductases that were downregulated in iron-restricted conditions were activated by Fur (S1 Table).
Many Fur-activated genes were also related to carbon utilization (S1 Table). As these genes were not regulated by iron, we suspected that their control by Fur was due to indirect rather than direct regulatory changes. Supporting this, we identified 10 transcriptional regulators that were differentially expressed in the Δfur mutant (S2A Fig). Furthermore, promoters of several Fur-activated genes contained a DNA binding motif for the cyclic AMP (cAMP) receptor protein (CRP) (S2B Fig), a regulator that alters gene expression in response to binding cAMP [44, 45]. As the CRP regulon in A. actinomycetemcomitans has been characterized [46], we could examine its overlap with the Fur regulon. This revealed that Fur-activated genes are enriched for both CRP-repressed (P = 1 x 10−11) and CRP-activated genes (P = 3 x 10−4, one-tailed Fisher’s exact test) (Table 4 in S1 Dataset). These results show that Fur expands its control over gene expression by acting indirectly through other transcriptional regulators including CRP.
To gain a more in-depth understanding of the Fur regulon, we next used the KEGG resource [47] to reconstruct how Fur globally regulates A. actinomycetemcomitans metabolism (Fig 2, S1 Table). As Fur activity is essentially a proxy for cellular iron levels, this network could be interpreted as how A. actinomycetemcomitans metabolism might adapt to iron-restricted conditions. In general, we found that the Fur-activated regulon is metabolically diverse, encompassing several central metabolism and carbon utilization pathways. Central metabolism pathways activated by Fur included glycolysis, the pentose phosphate pathway, pyruvate metabolism, and the reductive TCA cycle, and carbon utilization pathways included those for ribose, ascorbate, citrate, gluconate, glycerol, and inositol.
In contrast, analogous pathways in the Fur-repressed regulon were much less diverse and suggest that iron-restricted A. actinomycetemcomitans, like other bacteria [48], primarily engages in fermentation, mediated by a D-lactate dehydrogenase and a zinc-dependent alcohol dehydrogenase. However, Fur both repressed and activated metabolic pathways related to amino acids, vitamins, and cofactors. For instance, chorismate (a precursor to aromatic amino acids) and serine biosynthesis were repressed by Fur, while tyrosine and serine transport were activated by Fur. Other processes regulated by Fur included autoinducer-2 (AI-2) signaling, toxin production, and biofilm formation. Specifically, the Lsr AI-2 transporter [49], leukotoxin [50], and tight adherence [51] pili that mediate surface attachment were activated by Fur, while cytolethal distending toxin [52] was repressed by Fur.
Iron restriction also caused Fur-mediated upregulation of the gene encoding Dispersin B (DspB) (Fig 1, S1 Table), an enzyme produced by A. actinomycetemcomitans to disperse from biofilms [53]. Previously we showed that dspB transcription is increased during aerobic growth via the transcriptional regulator OxyR [54]; however, the fact that we performed our experiments under strictly anaerobic conditions suggests that dspB transcription is also controlled by iron in an oxygen-independent manner. Thus, we hypothesized that iron restriction would induce transcription of dspB in a Fur-dependent manner and subsequently lead to dispersal of A. actinomycetemcomitans from biofilms. To test this hypothesis, we first searched the dspB promoter for a Fur binding motif. This revealed a sequence that overlaps the dspB transcriptional start site and lies downstream of the reported [54] OxyR binding motif (Fig 3A). We then used a dspB promoter-lacZ transcriptional fusion to measure how dspB transcription is impacted by iron restriction. First we grew A. actinomycetemcomitans as colony biofilms under anaerobic conditions, and then to restrict iron, we transferred these biofilms to media containing an iron chelator. As expected, A. actinomycetemcomitans induced transcription of dspB >5 fold upon iron restriction, and we observed this effect in two different strains of A. actinomycetemcomitans (Fig 3B), 624 (the primary strain used in this study) and VT1169. Furthermore, this effect was iron-specific as addition of FeSO4 to the chelated media abolished dspB induction (Fig 3B). We then tested the Δfur mutant under the same conditions and observed a >30 fold induction of dspB transcription. However, this induction occurred irrespective of iron levels (Fig 3B) confirming that Fur represses dspB. Genetic complementation of the Δfur mutant restored responsiveness to exogenous iron levels (Fig 3C) indicating that the response was specific to Fur.
We next used a previously described biofilm dispersal assay [54] to test if iron restriction triggers A. actinomycetemcomitans biofilm dispersal. This assay takes advantage of the fact that when A. actinomycetemcomitans is grown with shaking in a glass test tube, it forms a “ring biofilm” on the test tube (Fig 3D). The assay works by first forming a ring biofilm in a low volume of media and then adding more media. If a second biofilm then forms above the first biofilm, dispersal has occurred. Using this assay, we found that A. actinomycetemcomitans can be induced to disperse from biofilms by adding an iron chelator (Fig 3D). Notably, biofilms dispersed even under anaerobic conditions. Although the assay is qualitative, we could also observe that iron-restricted biofilms dispersed even further in the presence of oxygen (Fig 3D). This suggests that iron restriction and aerobic growth conditions can work synergistically to promote biofilm dispersal. Together, these results show that, in addition to oxygen and OxyR, dspB is regulated by iron and Fur, and this regulation mediates A. actinomycetemcomitans biofilm dispersal in response to iron restriction.
One important drawback of our experimental approach for defining the Fur regulon is the inability to distinguish between direct and indirect gene regulation. We therefore used ChIP-seq to identify Fur binding sites (see Tables 1–2 in S1 Dataset for study design). To perform ChIP-seq, we complemented the Δfur mutant with a version of Fur tagged with the VSV-G epitope, allowing for immunoprecipitation. Importantly, VSV-G tagged Fur was expressed from its native promoter on a low-copy plasmid, and it genetically complemented the Δfur mutant (Fig 3C). In taking this approach, we hoped to prevent artefactual binding events that could arise from overexpression.
In total, we identified 67 promoter regions, representing 77 genes that are directly bound by Fur in either iron-replete or iron-restricted conditions. After accounting for potential operons, this total increased to 91 genes, revealing that Fur directly regulates 3.5% of the genome (Table 3 in S1 Dataset). To our surprise, only 41 of the 91 Fur-bound promoters were differentially controlled in iron-restricted conditions or in the Δfur mutant (S1C Fig). As this has been seen in other bacteria [55], this phenomenon may be more widespread than expected. Although RNA-seq suggested Fur activates as much as it represses gene expression (Table 3 in S1 Dataset), ChIP-seq revealed that Fur primarily represses gene expression (Table 6 in S1 Dataset), with over 4 times as many genes being directly repressed by Fur than activated. As expected, promoters directly bound by Fur contained a sequence similar to reported [37, 55] Fur binding motifs (Fig 4A).
In regards to function, many of the genes directly regulated by Fur were related to iron homeostasis as well as other metabolic processes (Fig 4B). Iron transporters directly regulated by Fur included the Afu ferric iron ABC transporter [34], 3 ferric iron siderophore transporters, TonB, and 5 TonB-dependent receptors, including those for hemoglobin and transferrin [26]. Interestingly, one ferric iron siderophore transporter was not differentially expressed in either iron-restricted conditions or in the Δfur mutant (Fig 4B). Other iron-related proteins directly regulated by Fur included ferritin and Fur itself, demonstrating that as in many other bacteria with Fur homologs [11], Fur in A. actinomycetemcomitans is autoregulated.
Metabolic processes directly regulated by Fur included glycerol transport, galactoside degradation, aromatic amino acid metabolism, glucose catabolism, uracil degradation, and the biosynthesis of NAD+ and folate (Fig 4B). ChIP-seq also revealed that Fur controls both the biosynthesis and degradation of cAMP (Fig 4B), explaining the regulatory link we discovered between the Fur and CRP regulons (S2 Fig). Even though these two regulons were highly overlapping (Table 4 in S1 Dataset), CRP was not among the transcriptional regulators differentially expressed in the Δfur mutant (S2A Fig). This suggested control at the post-transcriptional level. As revealed by ChIP-seq, Fur exerts this control over CRP by regulating intracellular amounts of cAMP. Similar connections between Fur and CRP have been described in other bacteria [44, 45].
While most Fur-bound promoters were positioned ahead of coding genes, the direct Fur regulon also comprised 5 sRNA (Table 3 in S1 Dataset), including the 2 sRNA that we found are upregulated by iron restriction (Fig 4B). The promoters of 2 CRISPR-associated (cas) genes were also bound by Fur (Fig 4B). Though the promoter of the CRISPR itself was not bound by Fur, 2 other CRISPR in the genome were differentially expressed in the Δfur mutant (Table 3 in S1 Dataset). As there are a total of 3 CRISPR in our strain, the collective findings of our RNA- and ChIP-seq data suggest that Fur contributes to the regulation of all 3 CRISPR in the genome. One explanation for why A. actinomycetemcomitans evolved to regulate CRISPR with Fur is that surface-displayed iron receptors can serve as attachment sites for bacteriophages [56] (Fig 4B).
Finally, we found that Fur directly binds the dspB promoter, revealing that its upregulation in both iron-restricted conditions and the Δfur mutant (Fig 3) was due to direct de-repression by Fur. Interestingly, we found that Fur also binds to a gene in the tight adherence pili locus. As RNA-seq showed that this locus is activated by Fur, this suggests a model where iron controls the entire A. actinomycetemcomitans biofilm cycle: in the absence of iron, Fur promotes biofilm dispersal by de-repressing Dispersin B (Fig 3), whereas in the presence of iron, Fur promotes surface attachment by activating the tight adherence pili. Altogether, our ChIP-seq dataset extended our understanding of the Fur regulon, beyond that provided by RNA-seq, and gave insight into Fur’s complex role in not only iron transport but also cAMP biogenesis, viral defense, and biofilm formation.
After characterizing the iron and Fur regulons in vitro, we next set out to leverage these datasets for assessing iron availability in A. actinomycetemcomitans infections. While A. actinomycetemcomitans is most widely associated with oral cavity infections, it can also spread systemically and cause abscesses in many parts of the body [24]. Like the oral cavity, abscesses are thought to be strong targets of host iron restriction [57] suggesting that A. actinomycetemcomitans is likely restricted for iron in the abscess. To test this, we used principle component analysis (PCA) to determine whether A. actinomycetemcomitans gene expression in the abscess (using a published dataset [58]) is more similar to its gene expression in biofilms on rich (Fe+) media or on iron-chelated (Fe-) media. To our surprise, the PCA showed that the abscess lies closer to Fe+ than Fe- biofilms (Fig 5A). To quantify this relationship, we used Spearman’s correlation coefficient. This also showed that the abscess is more similar to Fe+ than Fe- biofilms (Fig 5B). We then repeated these analyses for genes in the Fur regulon, now comparing gene expression in the abscess to biofilms of the wild type and Δfur mutant. This showed that the abscess lies closer to and correlates better with the wild type than the Δfur mutant (S3 Fig). As gene expression in the Δfur mutant is essentially like that of iron-restricted wild type, the higher proximity of the abscess to the wild type suggests that gene expression within this infection better resembles that where iron is abundant than scarce. Together, these analyses indicate that A. actinomycetemcomitans is not restricted for iron in the abscess, contrasting our initial hypothesis.
To support this finding, we next decided to test the Δfur mutant in the abscess. In almost every pathogen and infection model tested, Fur is required for virulence [11]. Therefore, we reasoned that if iron is indeed not restricted in the abscess, Fur should not be required for A. actinomycetemcomitans virulence. As we anticipated, the Δfur mutant persisted just as well as the wild type in the abscess (Fig 5C), again indicating that A. actinomycetemcomitans does not face severe iron restriction in this host environment. As a final test, we also performed ChIP-seq on A. actinomycetemcomitans in the abscess, reasoning that we should be able to detect binding if iron is available. In total, we identified 18 promoters that are bound by Fur in vivo (Fig 5D). These promoters were a subset of the promoters that we identified are bound by Fur in vitro. Notably, the in vivo-bound promoters were enriched for promoters that are preferentially bound by Fur in the presence of iron (P = 0.08, one-tailed Fisher’s exact test) (Fig 5D, Table 5 in S1 Dataset), reaffirming our conclusion that A. actinomycetemcomitans is not iron-restricted in the abscess.
One possible explanation for this result is that A. actinomycetemcomitans elicits an immune response that fails to fully sequester iron from the abscess. Alternatively, iron could be low but spatially heterogeneous in the abscess (S5A Fig), forming iron-rich ‘patches’ that locally promote A. actinomycetemcomitans colonization and growth. Supporting this, A. actinomycetemcomitans proliferates in the abscess as small (<10 μm across) cell aggregates [54]. Future studies should therefore focus on examining the host response to A. actinomycetemcomitans using more direct means to quantitatively map iron levels, addressing the possibility that iron forms concentrated micron-scale patches and gradients in the abscess.
As most pathogens cause infections as part of multispecies communities [59], we next sought to determine how co-infecting bacteria influence iron availability to A. actinomycetemcomitans in vivo. In the oral cavity, some of the most prevalent bacteria are Gram-positive streptococci [60]. Previously, we showed that intricate metabolic interactions, both cooperative [39] and competitive [54], with Streptococcus gordonii enhance A. actinomycetemcomitans virulence during abscess co-infection. Therefore, we conducted RNA-seq on abscesses co-infected with A. actinomycetemcomitans and S. gordonii (see Tables 1–2 in S1 Dataset for study design). Principal component analysis showed that co-infection, while positioned close to mono-infection, shifts the abscess towards Fe- biofilms (Fig 6A). Furthermore, the correlation between co-infection and Fe- biofilms was higher than that between mono-infection and Fe- biofilms (Fig 6B). In addition, genes upregulated in response to S. gordonii were enriched for genes upregulated by iron restriction (Table 7 in S1 Dataset) including 5 involved in iron uptake (S2 Dataset). Together, these results indicate that A. actinomycetemcomitans is restricted for iron during co-infection with S. gordonii.
How is iron restricted in co-infected abscesses? One possible mechanism for this interaction is direct interspecies competition for iron, as this has been reported in other mixed-species infections [18]. However, S. gordonii and other streptococci do not have an absolute growth requirement for iron [61], suggesting that interspecies competition is likely not responsible for reducing iron availability in co-infection. Supporting this, we found little evidence that S. gordonii differentially expresses genes related to iron homeostasis in either mono- or co-infection (Table 8 in S1 Dataset, S4 Dataset). Based on these data, we hypothesize that reduced iron in co-infection results from increased A. actinomycetemcomitans intraspecies competition (S5B Fig), a result of the 5–10 fold higher A. actinomycetemcomitans cell numbers observed during co-infection [39, 54, 62]. This higher bacterial burden could also enhance the host immune response and sequestration of iron within the abscess.
Since iron restriction induces dspB expression (Fig 1), a possible implication of the iron restriction associated with co-infection is that A. actinomycetemcomitans spatially reorganizes in response to S. gordonii. Previously we showed that A. actinomycetemcomitans proliferates in the abscess as small groups of cells (aggregates) and that the size of these aggregates is controlled by dspB [54]. Therefore, we anticipate that since S. gordonii restricts iron, A. actinomycetemcomitans forms smaller aggregates in co-infection than mono-infection. Of note, this spatial reorganization could facilitate iron acquisition since theoretically more cells per aggregate, due to the higher surface area to volume ratio, would have access to iron (S5C Fig). Current studies in our lab are aimed at addressing these possibilities.
Our abscess model provided a relatively simple in vivo environment for investigating the role of iron in interactions between A. actinomycetemcomitans, the host, and a co-infecting bacterium. To investigate the role of iron in a more complex polymicrobial environment, we next analyzed expression of the A. actinomycetemcomitans iron and Fur regulons during human periodontal disease. To do this, we extracted sequencing reads mapping to A. actinomycetemcomitans from a published meta-transcriptomics dataset comparing microbial gene expression from paired healthy and diseased (periodontitis) sites in the human oral cavity [63] (see Tables 1–2 in S1 Dataset for study design). Using principal component and correlation analyses, we showed that A. actinomycetemcomitans from healthy communities lies closer to Fe+ than Fe- biofilms (Fig 7A), but correlates to the same extent with both conditions (Fig 7B). In contrast, A. actinomycetemcomitans from diseased communities was closer to (Fig 7A) and correlated better with Fe- than Fe+ biofilms (Fig 7B). We obtained similar results for the Fur regulon (S3 Fig), specifically that disease was closer than health to the Δfur mutant. Furthermore, genes upregulated in periodontitis were enriched for both genes upregulated by iron restriction and repressed by Fur (Table 9 in S1 Dataset). These included several transporters for various iron sources. Notably, 3 of the 5 genes coding for the ferrous iron transporter were upregulated in periodontitis (S2 Dataset). This suggests that ferrous iron is an important iron source for A. actinomycetemcomitans in periodontal disease. We also found that the first gene in the dspB operon is upregulated in periodontitis. As DspB is stimulated not only by low iron (Fig 3) but also oxygen [54], we propose a model where A. actinomycetemcomitans disperses and migrates to deeper, anaerobic niches of the gingival pocket [21] during disease. As to why iron is restricted at diseased compared to healthy oral sites, we suspect that this phenomenon stems from a combination of the heightened host immune response (potentially resulting in greater iron sequestration) and the higher bacterial burden (potentially resulting in greater interspecies competition for iron) (S5D Fig).
In summary, our in vitro analysis of the iron-restricted and Fur regulons of A. actinomycetemcomitans allowed us to gauge its behavior regarding iron levels in multiple infection sites. We discovered that Fur has a complex role, impacting not only A. actinomycetemcomitans iron homeostasis but also biofilm dispersal. Our observation that A. actinomycetemcomitans disperses from biofilms in iron-restricted environments suggests that this pathogen may overcome host iron restriction via actively promoting spatial re-localization. Of additional interest, we also found that the availability of iron to A. actinomycetemcomitans is heterogeneous in vivo. Indeed, while iron does not appear to be restricted in A. actinomycetemcomitans mono-species abscess infections, co-culture abscess infections and human gum disease appear to be iron-restricted infections. Collectively these results suggest that microbial pathogens use multiple methods (iron acquisition and spatial re-localization) to acquire iron during infection, and that co-infecting bacteria have a significant impact on whether a pathogen is restricted for iron.
This study was performed in accordance with recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The animal protocol was approved by the Institutional Animal Care and Use Committee of The University of Texas at Austin (protocol number 00136). Mice were anesthetized with isoflurane delivered from a precision vaporizer and euthanized by asphyxiation with CO2 followed by cervical dislocation.
A. actinomycetemcomitans 624 (a clinical isolate), A. actinomycetemcomitans VT1169 (a laboratory strain [64]) and Streptococcus gordonii Challis DL1.1 (ATCC 49818) were used in this study. Cultures were routinely grown in tryptic soy broth/agar (BD Difco) supplemented with 0.5% (w/vol) yeast extract (Fluka) (broth, TSBYE; agar, TSAYE) in a 5% CO2 atmosphere with shaking at 250 rpm for A. actinomycetemcomitans. Anaerobic cultures were grown in a vinyl chamber (Coy) supplied with the following gas mixture: 85% N2, 10% CO2, 5% H2. Colony biofilms were formed by spotting 100 μl of culture (adjusted to OD600 = 1) onto a polycarbonate, 0.2 μm pore size membrane (Whatman) placed onto the surface of a TSAYE plate.
The A. actinomycetemcomitans 624 Δfur mutant was constructed by replacing the fur gene with a gene encoding spectinomycin resistance (aad9) by natural transformation as previously described [54] (see S2 Table for primer sequences). To construct the A. actinomycetemcomitans 624 strain for ChIP-seq, the fur gene and promoter region (~500 bp upstream of the fur start codon) were PCR-amplified from A. actinomycetemcomitans 624 genomic DNA with primers fur-pro-1-F and fur-tag-1-R, the reverse primer being designed to insert the VSV-G tag directly in front of the fur stop codon. This PCR product was then ligated into the TA vector pGEM-T Easy (Promega) and TSS transformed [65] into E. coli DH5α. A plasmid with the insert was then purified and used as the template for a second PCR with primers fur-pro-2-F and fur-tag-2-R, designed to incorporate KpnI restriction sites at the ends of the insert. Like the first PCR product, the second PCR product was cloned into pGEM-T Easy (Promega) and TSS transformed [65] into E. coli DH5α. A plasmid with the insert was then purified, digested with KpnI, and the VSV-G tagged fur gene was purified by gel extraction. The target vector pYGK [66] was also digested with KpnI, dephosphorylated with calf intestinal alkaline phosphatase (CIP), and purified. The KpnI-digested VSV-G tagged fur gene was then ligated into KpnI-digested, CIP-treated pYGK and TSS transformed into DH5α. This final plasmid was purified and electroporated as described [67] into the A. actinomycetemcomitans 624 Δfur mutant to generate the ChIP-seq strain. Plasmid inserts were sequenced at the UT Austin DNA Sequencing Facility.
Colony biofilms of the A. actinomycetemcomitans 624 wild type and Δfur mutant were prepared and grown overnight under anaerobic conditions. Biofilms were then transferred to fresh locations on the same plate, incubated for 2 hours, and then transferred to new plates, either TSAYE (+Fe condition) or TSAYE + 250 μM 2,2’-dipyridyl (-Fe condition). (This concentration of 2,2’-dipyridyl reduces the growth rate and yield of planktonic A. actinomycetemcomitans by approximately 50% [25], and fully restricts the growth of A. actinomycetemcomitans iron transport mutants on streak plates [27].) Biofilms were then further incubated for 2 hours, and following this they were stored in RNAlater solution. Per replicate, 8 colony biofilms were pooled together for each treatment (+Fe wild type, -Fe wild type, +Fe Δfur, -Fe Δfur), and altogether 2 replicates were performed, each on different days (Table 1 in S1 Dataset). Biofilm RNA was extracted with RNA-Bee (Tel-Test) according to the manufacturer’s protocol. Abscess RNA was extracted as previously described [58] from 1 or 2 pooled abscesses per biological replicate. Extracted RNA was treated with RQ1 DNase (Promega) to remove DNA contamination, and DNA removal was verified by PCR of the clpX gene. RNA integrity was verified by agarose gel separation of RNA denatured with NorthernMax-Gly loading dye (Ambion). Bacterial rRNA depletion, RNA fragmentation, and RNA-seq library preparation were performed as previously described [68], except that co-infected abscess RNA-seq libraries were rRNA-depleted with the MICROBEnrich and MICROBExpress kits (Ambion) and size-selected for fragments between 130–200 bp. Libraries were sequenced on 1x100 single-end Illumina HiSeq runs at the UT Austin Genome Sequencing and Analysis Facility. RNA-seq data were deposited into the NCBI Sequence Read Archive under accessions SRP081045 and SRP093165.
Colony biofilms of A. actinomycetemcomitans 624 Δfur expressing VSV-G tagged fur were prepared and treated as described for the RNA-seq experiment, with the exception that growth plates included 10 μg/ml kanamycin to maintain the ChIP-seq plasmid. After 2 h incubation in the +Fe and -Fe conditions, biofilms were placed into 50 ml TSBYE + 1% formaldehyde per treatment, gently agitated for 20 min at room temperature, and then vigorously vortexed to dislodge the cells from the membranes. For abscess infections, the A. actinomycetemcomitans 624 Δfur strain expressing VSV-G tagged fur was first grown overnight in TSBYE + 10 μg/ml kanamycin under anaerobic conditions. Then, 3 ml of culture were washed and resuspended in 1 ml TSBYE, and 100 μl were injected into each thigh of three 10-week-old mice to form abscesses [62]. After 3 days, abscesses were harvested and each placed into 1 ml TSBYE + 5% formaldehyde. After overnight fixation, abscesses were homogenized, and the homogenates were collected by centrifugation and washed with TSBYE. Eight biofilms and 3 abscesses were used per experiment, and experiments were performed twice. At this point, the fixed biofilm cells and fixed abscess homogenates were subjected to the same ChIP procedure [37]. First, fixed samples were added with 0.5 M glycine to quench crosslinking. Samples were then washed with TBS (50 mM Tris-HCl, pH 7.5; 150 mM NaCl), each resuspended in 1 ml lysis buffer (10 mM Tris-HCl, pH 8; 100 mM NaCl; 1 mM EDTA; 0.5 mM EGTA; 0.1% deoxycholate; 0.5% N-lauroylsarcosine) + 1 mg/ml lysozyme + protease inhibitor (Sigma), and incubated at 37°C for 30 min. The samples were then chilled on ice, sonicated 2x for 5 s, or until the solution became clear, with a tip sonicator (QSonica) at 60% amplitude, and then further sonicated at 4°C for 20 min in 10 s on/10 s off cycles with a Q800R sonicator (QSonica) at 60% amplitude. Lysates were then separated from unlysed debris by centrifugation, and 25 μl of each clarified lysate was saved for the ChIP input control. A 1/10 volume of Triton X-100 (10% solution in lysis buffer) was then added to each sample, followed by 25 μl of Protein G Dynabeads (ThermoFisher) coated with anti-VSV-G monoclonal antibody (Sigma). The samples were rotated overnight at 4°C, and following this, each sample was washed 5x with 1 ml RIPA buffer (50 mM HEPES, pH 7.5; 500 mM LiCl; 1 mM EDTA; 1% Nonidet P-40; 0.7% deoxycholate), 1x with 1 ml TE (10 mM Tris-HCl, pH 8; 1 mM EDTA) + 50 mM NaCl, and resuspended in 100 μl EB (50 mM Tris-HCl, pH 7.5; 10 mM EDTA; 1% SDS). The samples were then incubated at 65°C for 30 min, separated from the Dynabeads by centrifugation, and further incubated overnight, along with the ChIP inputs, at 65°C to reverse crosslinks. Following this, the samples were brought up to 200 μl in volume with TE, incubated with 8 μl 10 mg/ml RNase A for 2 h at 37°C, and further incubated with 4 μl 20 mg/ml proteinase K for 2 h at 55°C. Finally, samples were purified with the ChIP DNA Clean & Concentrator kit (Zymo Research), and ChIP-seq libraries were prepared using the NEBNext ChIP-Seq Library Master Mix Set according to the manufacturer’s instructions. Libraries were sequenced on 1x75 single-end Illumina NextSeq runs at the UT Austin Genome Sequencing and Analysis Facility. ChIP-seq data were deposited into the NCBI Sequence Read Archive under accession SRP081045.
A. actinomycetemcomitans 624 genome sequencing and assembly were performed as previously described [62]. Annotation was performed with PGAP [69], RAST [70], and KAAS [71]. Noncoding RNA sequences [58] were extracted from the D7S-1 genome with the bedtools getfasta function [72] and mapped to the A. actinomycetemcomitans 624 genome with bowtie2 v2.2.5 in very-sensitive-local mode [73]. Raw genome sequences were deposited into the NCBI Sequence Read Archive under accession SRP064936. The genome assembly was deposited into GenBank under accession CP012959.
Raw reads were processed with cutadapt v1.9.1 (and higher) (http://cutadapt.readthedocs.org/en/stable/index.html) to (1) trim 3’ low-quality bases (cutoff: 15), (2) trim 3’ adaptors (sequence: AGATCGGAAGAGCACACGTCTGAACTCCAGTCAC), and (3) discard short reads (minimum length: 10 bases for non-human samples, or 15 bases for human samples,). Processed reads were mapped to the A. actinomycetemcomitans 624 or S. gordonii genome with bowtie2 v2.2.5 in very-sensitive-local mode [73], and only reads with mapping quality (MAPQ) scores ≥39 for non-human samples, or ≥20 for human samples, were kept for further analysis. Reads were counted per gene strand-specifically with the featureCounts function in Rsubread v1.20 (and higher) [74]. Raw read counts, excluding rRNA and tRNA, were adjusted for between-sample differences in sequencing depth with the estimateSizeFactors function in DESeq2 v1.10.1 (and higher) [75]. Raw read counts were also normalized for between-gene differences in GC content by including a normalization matrix. This normalization matrix was generated with the withinLaneNormalization function in EDASeq v2.4.1 (and higher) [76], and transformed to be on the scale of read counts, as described in the DESeq2 vignette. GC content values for the matrix were calculated with the nucBed function in bedtools v2.20 (and higher) [72]. Default DESeq2 parameters were used for estimating dispersions and performing the Wald test for differential expression analysis, and a multi-factor design, as described in the DESeq2 vignette, was used for paired analysis of healthy and diseased human from each patient [63]. Significance cutoffs were as follows. Iron-restricted, Fur, and co-infection regulons: log2 fold change (FC) > 0.5; adjusted P value < 0.05. Periodontitis: log2 FC > 0.5; non-adjusted P < 0.05. See Table 10 in S1 Dataset for a summary of the RNA-seq data analysis. Analyses were performed both locally and on the UT Austin Texas Advanced Computer Center.
Raw reads were processed as described for the RNA-seq analysis. Binding regions (peaks) were called with MOSAiCS v2.9.9 [77]. Adjustable MOSAiCS parameters were: fragLen, 200; binSize, 50; capping, 3; bgEst, automatic; FDR, 0.1. The control for colony biofilm samples was the in vitro ChIP input, and the control for abscess samples was the in vivo ChIP input. Promoters were defined for coding genes as the 200 and 50 bp up- and downstream of the start codon and for noncoding genes as the 100 and 25 bp up- and downstream of the start codon. The number of peaks overlapping each promoter was counted for each sample with the bedtools intersectBed function [72]. Peaks were only counted if they overlapped at minimum 50 bp of coding promoters or 25 bp of noncoding promoters. After tallying overlapping peaks, promoters were only considered bound by Fur if they overlapped a peak in at least 2 in vitro biofilm replicates. If one of these promoters also overlapped a peak in at least 1 of the abscess replicates, it was considered also bound by Fur in vivo. Promoters overlapping peaks in both Fe+ biofilm replicates were considered preferentially bound in the presence of iron, and promoters overlapping peaks in both Fe- biofilm replicates were considered preferentially bound in the absence of iron. See Tables 11–12 in S1 Dataset for a summary of the ChIP-seq data analysis. Analyses were performed both locally and on the UT Austin Texas Advanced Computer Center.
Promoter sequences were extracted from the A. actinomycetemcomitans 624 genome with the bedtools getfasta function [72] and submitted to MEME [78] to identify consensus binding motifs. MEME settings were: site distribution, zero or one occurrence per sequence; background model, 0-order model of sequences; minimum motif width, 15; maximum motif width, 25. A CRP box was identified among the promoters of genes activated by Fur, and a Fur box was identified among the promoters directly bound by Fur. Each binding motif was submitted to the MEME Suite program FIMO [78] to calculate the significance (FDR) of its occurrence within each promoter of its respective promoter set.
sRNA sequences were extracted from the A. actinomycetemcomitans 624 genome with the bedtools getfasta function [72] and submitted to TargetRNA2 [43] with default parameters and the A. actinomycetemcomitans strain D7S-1 genome selected. Predicted target sequences in the D7S-1 genome were mapped in fasta format to the A. actinomycetemcomitans 624 genome with bowtie2 v2.2.5 in very-sensitive-local mode [73].
The A. actinomycetemcomitans CRP regulon was determined with a microarray designed for strain HK1651 [46]. To use this data, A. actinomycetemcomitans 624 gene sequences in fasta format were mapped to the A. actinomycetemcomitans HK1651 genome with bowtie2 v2.2.5 in very-sensitive-local mode [73].
Normalized read counts were transformed with the DESeq2 [75] rlog method with blindness set to false, as described in the DESeq2 vignette, and the PCA was performed with the prcomp function in R on the 93 genes in the iron-restricted regulon or the 218 genes in the Fur regulon.
β-galactosidase assays were performed as previously described [54] using a chemiluminescent assay system (Galacto-Light Plus).
The biofilm dispersal assay was performed as previously described [54].
The murine abscess infection model was performed as previously described [39, 62], except that in the experiment testing the virulence of the Δfur mutant, anaerobic conditions were used for preparing the inoculum and plating serial dilutions of the abscess homogenates.
All RNA-seq and ChIP-seq files are available from the National Center for Biotechnology Information Sequence Read Archive (accession numbers SRP081045 and SRP093165).
|
10.1371/journal.ppat.1007198 | Immobility in the sedentary plant-parasitic nematode H. glycines is associated with remodeling of neuromuscular tissue | The sedentary plant-parasitic nematodes are considered among the most economically damaging pathogens of plants. Following infection and the establishment of a feeding site, sedentary nematodes become immobile. Loss of mobility is reversed in adult males while females never regain mobility. The structural basis for this change in mobility is unknown. We used a combination of light and transmission electron microscopy to demonstrate cell-specific muscle atrophy and sex-specific renewal of neuromuscular tissue in the sedentary nematode Heterodera glycines. We found that both females and males undergo body wall muscle atrophy and loss of attachment to the underlying cuticle during immobile developmental stages. Male H. glycines undergo somatic muscle renewal prior to molting into a mobile adult. In addition, we found developmental changes to the organization and number of motor neurons in the ventral nerve cord correlated with changes in mobility. To further examine neuronal changes associated with immobility, we used a combination of immunohistochemistry and molecular biology to characterize the GABAergic nervous system of H. glycines during mobile and immobile stages. We cloned and confirmed the function of the putative H. glycines GABA synthesis-encoding gene hg-unc-25 using heterologous rescue in C. elegans. We found a reduction in gene expression of hg-unc-25 as well as a reduction in the number of GABA-immunoreactive neurons during immobile developmental stages. Finally, we found evidence of similar muscle atrophy in the phylogenetically diverged plant-parasitic nematode Meloidogyne incognita. Together, our data demonstrate remodeling of neuromuscular structure and function during sedentary plant-parasitic nematode development.
| The sedentary plant-parasitic nematodes are an economically devastating group of pathogens. They are capable of modifying their host’s metabolism by establishing a permanent feeding site within the plant roots. Following infection, sedentary nematodes remain at their feeding site where they are only capable of slight movements associated with feeding. Upon reaching adulthood, females continue to remain immobile at their feeding site, while adult males regain mobility to search for females. Using both microscopic observation and examination of gene function, we found that immobility is associated with changes to the muscle and neurons controlling movement. Interestingly, we found that males are capable of reversing these changes upon reaching adulthood. Our findings may be used to develop targeted control strategies for these devastating parasitic nematodes while avoiding potential off-target effects to beneficial nematodes that do not show this unusual mode of development and behavior.
| The sedentary plant-parasitic nematodes are among the most damaging pathogens to agricultural crop production [1]. For example, the soybean cyst nematode, Heterodera glycines, accounts for over a billion dollars in yield loss yearly [2]. H. glycines hatches as an infective, mobile second-stage juvenile (J2) (Fig 1). Following infection, H. glycines establishes a multinucleated feeding site and develops from a vermiform J2 to lemon-shaped and sausage-shaped J4 females and males, respectively (Fig 1). While feeding, both sexes are restricted in movement to slight head motion at their feeding site [3]. During J4, males undergo extensive morphological remodeling back to a vermiform morphology. Following the final molt, the adult male is fully mobile, while the female remains lemon-shaped and immobile.
The structural basis for the loss of mobility and male-specific resumption of mobility in sedentary nematodes is not clear. Several authors have speculated that immobility is due to muscle degeneration [4–6]. For example, Bird [6] demonstrated the immobility of the sedentary nematode Meloidogyne javanica through careful dissection of nematodes from roots and subsequent behavioral assays. He later speculated that loss of mobility was due to somatic muscle atrophy [7]. An early histological examination of Meloidogyne hapla noted an absence of most somatic muscle fibers in the adult female and suggested that loss of muscle was a result of sedentary behavior [4]. However, this description was limited by the resolution of light microscopy and did not examine changes during development. Nor did this work account for possible changes in the nervous system that could impact mobility.
The structure of body wall muscles during mobile stages of sedentary plant-parasitic nematodes are similar to Caenorhabditis elegans ([8–10] and herein). The body wall muscle comprises four quadrants with striated muscle fibers oriented in an oblique angle to the longitudinal axis. In C. elegans, the majority of body wall muscle is innervated by a series of motor neurons in the ventral nerve cord (VNC) (S1 Fig) [11]. We recently demonstrated substantial variability in the number of VNC neurons among nematodes, including differences between J2 H. glycines and C. elegans [12]; however, the specific wiring pattern of somatic muscle in sedentary plant-parasitic nematodes is unknown. In C. elegans, movement is generated through muscle contractile force propagated to the outer cuticle through a thin basal lamina and epidermal layer via fibrous organelles [13]. Muscle contractions are produced via the innervation of excitatory cholinergic and inhibitory GABAergic VNC motor neurons [14,15].
Within the phylogenetic clade comprising most plant-parasitic nematodes, the sedentary life-style has arisen at least twice [16,17]. While having a similar life-cycle, the root-knot (Meloidogynidae) and cyst (Heteroderidae) nematodes are phylogenetically diverged. The sedentary nematodes are more fecund and considered more damaging than their phylogenetically closest “migratory” plant-parasitic nematode relatives [1,18]. An understanding of the neuromuscular modifications that occur during sedentary nematode development could provide novel targets for control. For example, the resumption of mobility by male cyst nematodes could be targeted without affecting non-target nematodes that do not undergo this unusual development.
Here, we conduct a detailed developmental examination of the structural basis for changes of mobility in the sedentary species H. glycines. Using light and electron microscopy, we describe the anatomical changes to the neuromusculature and surrounding epidermis of H. glycines during the transition to immobility and the resumption of mobility in males. We characterize the GABAergic system during H. glycines post-hatch development. Finally, we examine whether similar changes to body wall muscles occur in the phylogenetically diverged root-knot nematode Meloidogyne incognita. Our results demonstrate extensive remodeling of neuromuscular tissue in sedentary plant-parasitic nematodes.
To test whether the immobility of sedentary nematodes was due to degeneration of body wall muscles [4,5,7], we first analyzed the somatic muscle tissue in the mobile J2 stage of H. glycines using both the F-actin binding fluorescent probe phalloidin as well as transmission electron microscopy (TEM). Mobile J2 H. glycines somatic muscle comprise rhomboid-shaped cells longitudinally arranged in four quadrants along the length of the body (Fig 2 and S1 Fig). Each quadrant contains two rows of overlapping muscle cells and each cell contains striations obliquely oriented to the longitudinal axis. Using TEM on transverse sections following high-pressure freezing and freeze substitution, we found that individual muscle cells in J2 H. glycines comprise multiple sarcomeres separated from the cuticle by a thin basal lamina and epidermis (Fig 3A). As previously shown [8,19–21], J2 sarcomeres are arranged into obvious A- and I-bands with thick and thin filaments. The non-contractile region of body wall muscles is enriched with mitochondria. Overall, our light and electron microscopy images correspond to previous data from mobile stages of H. glycines and indicate a similar muscle structure to C. elegans.
Within six days following infection, H. glycines molts to a J3 [22]. We found that phalloidin-stained J3 H. glycines somatic muscles are disorganized and smaller than in mobile J2s (Fig 2). The body wall muscles lose their rhomboid shape and the pattern of two-row cells in each quadrant is no longer recognizable. Our TEM examination of immobile J3s also demonstrated a shrinkage of muscle and sarcomere disorganization (Fig 3B and Table 1). The sarcomeres of immobile J3s lack distinguishable A- and I-bands. Furthermore, the number of thick and thin filaments appears reduced in J3s compared with J2s. Sexual differentiation of H. glycines is first visible in late-J3 (personal observation and [23]); however, we did not observe any obvious sex-based differences in muscle structure between male and female J3s suggesting that both sexes undergo muscle atrophy. In addition to structural changes to the muscle itself, we observed that the body wall muscles are displaced internally away from the cuticle in J3 H. glycines. This displacement is correlated with a corresponding thickening of the epidermis (Fig 3B and Table 1). In C. elegans, the force of body wall muscle contraction is transduced through a thin epidermal layer via attachment to the outer cuticle [13,24]. Loss of mobility in H. glycines may be due to a combination of muscle atrophy as well as an inability to transduce contractile force to the outer cuticle. We found that the body wall head muscles and esophageal muscles of immobile stages are intact (S2A Fig). This observation is consistent with the previously reported feeding behavior of cyst nematodes and suggests that muscle atrophy is cell-specific and not due to a generalized atrophy as seen in sarcopenia of C. elegans [25,26].
J4 H. glycines females continue to feed and grow in width, while J4 males begin remodeling back into a vermiform shape (Fig 1). In phalloidin-stained J4 females (Fig 2), we detected faint stripes underlying the epidermis suggestive of muscle filaments. Interestingly, nuclear staining revealed a row of nuclei, similar in size and shape to muscle nuclei, immediately underlying the phalloidin stripes (S2B Fig). Similarly, we detected putative remnant muscles stripes in J4 females using TEM (Fig 3D). These data suggest that muscles do not completely degenerate by J4.
J4 males renew body wall muscle. While lacking the typical organization of mobile J2s or adult males, J4 male phalloidin-stained muscle is substantially larger compared with J3s and J4 females (Fig 2). Examination of J4 male muscles by TEM revealed identifiable sarcomeres, albeit less organized and smaller than in mobile J2s (Fig 3C and Table 1). Like J3 and J4 females, the muscle tissue of J4 males was located more internally than in mobile J2s or adult males, indicating that muscles had not reattached to the body wall (Table 1). When males reach adulthood, the muscles are positioned directly adjacent to the body wall (Fig 3E). The relative size of body wall muscles in adult males is larger compared to J3s or J4 females and comprises well-defined sarcomeres (Figs 2 and 3E and Table 1). Together, our light and electron microscopy data illustrate that H. glycines undergoes cell-specific muscle atrophy and sex-specific muscle regrowth during development.
In C. elegans, contraction and relaxation of most body wall muscles are regulated by motor neurons within the ventral nerve cord (VNC) [8,27]. We, therefore, examined the VNC during H. glycines development. We previously found that mobile J2 H. glycines contain 65 VNC neurons [12,28]. Here, using DAPI staining we found a gradual reduction of VNC neurons during development from the mobile J2 to sedentary J3 and J4 females (Fig 4 and Table 2). In addition, the overall pattern of the VNC in sedentary stages deviates from the linear pattern seen during mobile stage. Some nuclei of the VNC in sedentary stages are located several microns away from the ventral midline (Fig 4). Examination of the VNC by TEM also suggested a loss of fasciculation and separation from the nearest muscle during J3, while J4 males had a properly fasciculated VNC with synaptic vesicles in several processes (Fig 5). Strikingly, we found 70 neurons in the adult male VNC and a reorganization of the cord into a linear arrangement (Table 2). It appears that, similar to body wall muscles, the motor neurons in the VNC of both sexes of H. glycines degenerate during sedentary stages of development. However, males undergo neuronal remodeling that includes the addition of neurons. This addition of neurons results in more VNC neurons in adult males than in mobile J2s. Similar to C. elegans, the males of H. glycines may include sex-specific motor neurons in the VNC [29,30].
To further examine the role of motor neuron degeneration in the sedentary behavior of H. glycines, we characterized the GABAergic nervous system. In C. elegans, GABA is produced by approximately one-fourth of the C. elegans VNC neurons. Release of GABA in the VNC leads to relaxation of body wall muscles and acts in opposition to the excitatory neurotransmitter acetylcholine. GABA is synthesized by glutamate acid decarboxylase (GAD), which in C. elegans is encoded by unc-25 [14,15,31]. We cloned the putative ortholog of unc-25, from H. glycines (hg-unc-25). The predicted amino acid sequence of hg-UNC-25 is over 60% identical to C. elegans UNC-25 (S3 Fig). To determine if hg-unc-25 is functional, we attempted to rescue the C. elegans unc-25(e156) mutant with the H. glycines homolog. unc-25(e156) mutants are completely defective for production of GABA [14]. Using immunohistochemistry, we found that unc-25(e156); hg-unc-25(+) animals contained 2–4 GABA VNC, whereas unc-25(e156) mutants are completely negative for GABA-immunoreactivity (Fig 6). While our data suggest that hg-UNC-25 functions to produce GABA, wild type C. elegans contain 19 GABA-immunoreactive neurons. The lack of complete rescue may be due to mosaic expression of the rescuing construct or the absence of unidentified C. elegans-specific exonic regulatory regions in hg-unc-25.
We hypothesized that GABAergic VNC neurons were among those that degenerated during development. Therefore, we expected a reduction of hg-unc-25 expression in sedentary stages compared to the mobile J2s. Using RT-qPCR, we found that expression of hg-unc-25 is significantly reduced in J3s and J4 females compared to J2s (Fig 7A). These data suggest a reduction in GAD synthesis and possibly a concomitant reduction in GABA production. While hg-unc-25 expression levels in mobile adult males did not return to levels in mobile J2s, expression was significantly higher in the adult males compared to J4 females.
To examine GABA more directly, we utilized immunohistochemistry to characterize the GABAergic nervous system of H. glycines. We found that the pattern and morphology of GABA-immunoreactive neurons in J2 H. glycines were similar, but not identical to C. elegans [31]. We detected several GABA-immunoreactive cells in the head of H. glycines surrounding and posterior of the nerve ring that we identified as likely homologs of the RME motor neurons and the polymodal neuron AVL, respectively (Fig 7B). In addition, we identified a pair of GABA-immunoreactive neurons in the head with no known GABAergic positional homolog in C. elegans. We found 21 GABA-immunoreactive cells in the VNC of J2s (Fig 7B). Similar to C. elegans, the commissures derived from these GABA-immunoreactive neurons in H. glycines run to the dorsal side [8]. However, unlike the VNC commissures of C. elegans and Ascaris suum, which usually run along the right side [8,32], we found that the majority of GABA-immunoreactive commissures in H. glycines travel dorsally along the left side of the animal suggesting a reversal in neuronal handedness. While we found fewer GABA-immunoreactive neurons in the VNC of J4 females (x¯=12, n = 12) compared with J2s (Fig 7C), the results were inconsistent. Many individuals showed no antibody staining. Furthermore, the established antibody staining method for plant-parasitic nematodes [33] requires bisecting of the nematode for increased penetration, which, combined with the large size of the J4 female, makes identification of individual neurons and comparison with the J2 stage impossible. Similarly, and for unknown reasons, we were unsuccessful in our attempts to observe GABA-immunoreactivity in adult males. A final caveat to our immunohistochemistry studies is that we cannot differentiate true GABAergic neurons from those that potentially take up GABA. However, combined with our RT-qPCR data, our results suggest that GABAergic neurons are among those that degenerate during sedentary stages.
Among Tylenchomorpha nematodes, sedentary behavior is found in multiple genera. The root-knot nematodes (Meloidogyne spp.) also become immobile soon after infecting the plant host [6]. However, Meloidogyne spp. are phylogenetically diverged from Heterodera glycines (S4 Fig). Sister lineages to both Meloidogyne spp. and Heterodera spp. are mobile during all post-embryonic stages [16,17]. We examined mobile and immobile developmental stages of Meloidogyne incognita with light microscopy to determine if changes in neuromuscular structure are correlated with sedentary behavior in phylogenetically diverged species. Similar to H. glycines, mobile J2 M. incognita have well developed body wall muscles that atrophy following infection and the onset of sedentary stages (Fig 8). Additionally, we found fewer VNC neurons in sedentary stages of M. incognita than in the fully mobile J2 (Table 2). Finally, as a control group, we examined the ventral cord of Pratylenchus penetrans at different developmental stages. P. penetrans is classified in the same phylogentic clade as H. glycines and M. incognita [16,17], but is mobile at all stages (S4 Fig). We found that the number of ventral cord neurons in P. penetrans remained stable throughout development. Our current results with P. penetrans, combined with our previous results with the fungal-feeding Tylenchomorpha nematode Apehelenchus avenae, which shows no degeneration of VNC neurons in late development [12], suggest that VNC degeneration in H. glycines and M. incognita is specifically correlated with sedentary behavior.
We demonstrate that H. glycines undergoes a progressive atrophy of neuromuscular tissue that likely leads to immobility. Elsea (1951) suggested that the absence of body wall muscle in M. hapla adult females was “a result of the sedentary mode of life”; however, this conclusion was based on light microscopy of mobile J2s and adult females. While we cannot entirely rule out that the loss of muscle mass in H. glycines is caused by a lack of movement, our data suggests this is highly unlikely. We demonstrate that muscle atrophy occurs in conjunction with loss of mobility. Furthermore, cyst nematodes, such as H. glycines, are able to survive as fully developed, but inactive J2s, for years in a pre-hatch diapause state [34]. Upon activation by a hatching stimulant, these animals readily resume movement suggesting that inactivity alone does not result in muscle atrophy. We suggest that the combination of muscle atrophy and detachment of muscle from the cuticle combined with changes to neuron structure and gene expression cause immobility in feeding cyst nematodes.
The GABAergic system of C. elegans is essential for proper movement [14]. Our rescue of the C. elegans unc-25 mutant with hg-unc-25 demonstrates its functional conservation and corresponds with recent data demonstrating the conservation of regulatory regions of unc-25 among diverse nematode species [35]. Despite the apparent conservation of unc-25 between C. elegans and H. glycines, we found differences in the number and morphology of GABA-immunoreactive cells between these species. Interestingly, we also found that the handedness of most H. glycines GABA-immunoreactive neurons in the VNC differs from A. suum and C. elegans. Our results suggest that nematodes may be an appropriate animal phylum for exploring the evolution of neuronal handedness.
The mechanisms regulating neuromuscular remodeling in H. glycines are unknown. Previous RT-PCR data of several genes encoding muscle-related proteins showed reduced expression in adult females compared with mobile J2s [5]. A reexamination of the expression of these genes during remodeling may provide insight into the molecular basis of neuromuscular atrophy in H. glycines. Our data suggest that neuromuscular changes in H. glycines are cell specific. The absence of atrophy in head and esophageal muscles in H. glycines allows for continued feeding and head movements during the sedentary stages. In C. elegans, disruption of the BAG2 chaperone homolog UNC-23 leads to the opposite phenotype wherein the nematode undergoes a progressive degeneration of head muscles [19]. It is possible that cell autonomous changes in expression of H. glycines muscle-related genes result in the different developmental fates of head and body wall muscles. Alternatively, the muscle cell-specific atrophy seen in H. glycines could be due to selective neuronal degeneration. The head and neck muscles of C. elegans are differently innervated than the rest of the body wall muscles [11]. Whereas most C. elegans body wall muscles are innervated by the VNC, the head muscles receive innervation directly from the circumpharyngeal nerve ring. It will be interesting to determine via TEM whether a similar wiring pattern occurs in H. glycines and if muscle atrophy is due to denervation as seen in some mammals [36].
The ability to establish a specialized feeding site in their host is a significant adaptation among plant-parasitic nematodes and is often correlated with sedentary behavior. The atrophy of neuromuscular tissue is likely due to a relaxation of selection pressure for the maintenance of mobility. In addition, the degeneration of neuromuscular tissue could allow energy derived from the catabolism of muscle protein to be shunted toward reproductive development. The presence of muscle atrophy in both H. glycines and M. incognita strongly suggests the independent evolution of this behavior. However, it is important to note that within the Tylenchida, which includes both plant and insect parasitic nematodes, sedentary behavior is found in several phylogenetically separate genera [16]. It is, therefore, possible that sedentary behavior evolved first in a basal lineage within the Tylenchida and was repeatedly lost during evolution. Careful comparative analysis of these genera will be required to fully resolve these competing hypotheses.
H. glycines was originally isolated from a soybean field in Illinois, USA and cultured in the greenhouse on soybean cultivar ‘Macon’. To collect synchronized developmental stages of H. glycines, seeds of the soybean variety ‘Macon’, were germinated in a moist paper towel for three days. Soybean seedlings were then placed in pluronic gel F-127 with hatched J2 H. glycines for 24 hours [37]. Alternatively, soybean seedlings were planted into pots and inoculated with freshly hatched H. glycines second-stage juveniles (J2s). After 24 hours, roots were washed with water to remove nematodes that had not infected. H. glycines-infested roots were then planted into a pasteurized sandy loam soil and kept under a 12-hour light cycle until extraction. Infested roots were macerated with a hand blender to obtain H. glycines at specific time points. The roots were processed 6–7 days after inoculation to collect J3s and 9–10 days after inoculation to collect J4 females and males [38]. Developmental stages and sex were determined by overall body size and gonad morphology [23]. Meloidogyne incgonita was a gift from Dr. Jason Bond and maintained on ‘Rutgers’ tomato in the greenhouse. Seeds of tomato variety ‘Rutgers’ were planted into a sandy loam soil for two weeks and then inoculated with freshly hatched M. incognita J2s. After four weeks, post-infective stages of M. incognita were extracted from the tomato plants and the developmental stages were identified based on morphology [39]. P. penetrans was a gift from Dr. Terry Niblack and was cultured on corn root explants on Murashige and Skoog (MS) media [40]. Specific developmental stages of P. penetrans were isolated by synchronizing populations from eggs. P. penetrans eggs were extracted as previously described [41] Freshly hatched J2s were collected and placed on corn root explants on MS media [40]. Developmental stages were determined by the presence of molting, gonad morphology and overall body size. C. elegans strains N2 and CB156 unc-25(e156) III were cultured on NGM agar with E. coli OP50 as previously described [42].
Nematodes were fixed in 4% paraformaldehyde (Electron Microscopy Science) overnight at 4°C and washed three times with water. Nematodes were placed on a 4% agar pad with phalloidin (10 unit/ml; Thermo Fisher Scientific) and DAPI (0.2–0.5 μg/ml; Thermo Fisher Scientific) or Hoechst (0.2 mM; Thermo Fisher Scientific) to stain nuclei. An Andor system micropoint laser attached to a Zeiss Axioimager with 63x objective was used to create openings in the cuticle at regular intervals along the length of the nematode to increase the penetration of dyes. Following opening of the cuticle, nematodes were incubated in the stain overnight. Successful penetration of phalloidin in sedentary nematodes was confirmed by observing fluorescence in esophageal muscle. At least ten animals were examined for each developmental time point. All light microscopy images were captured with Zen software on a Zeiss M2 AxioImager with DIC and fluorescence optics. Images were examined in FIJI and multiple images were combined using the ImageJ stitching plugin.
Nematodes were fixed in Carnoy’s fixative (60% ethanol, 30% acetic acid, 10% chloroform) overnight in a 1.5 ml centrifuge tube and allowed to settle in the bottom of the tube [43,44]. The supernatant was removed and transferred to 75% ethanol before staining with 0.2–0.5 μg/ml of DAPI overnight in the dark at room temperature. The VNC nuclei were identified based on their size and morphology [12,45]. At least eight animals were examined for each species at each developmental time point.
Synchronized developmental stages of H. glycines were collected from soybean roots and stored overnight at 4°C. High pressure freezing and freeze substitution were modified from previous methods used for C. elegans [46,47]. Metal specimen carriers were coated with 1-hexadecene and a layer of E. coli strain OP50. Nematodes were loaded into carriers with 20% bovine serum albumin and frozen in an HPM 010 high pressure freezer. Freeze substitution was performed in 2% OsO4 (Electron Microscopy Sciences) and 0.1% uranyl acetate (Polysciences) in 2% H2O and 98% acetone in an FS-8500 freeze substitution system. Samples were kept at -90°C for 110 hours before being warmed to -20°C over five hours. Samples were then kept at -20°C for 16 hours before being warmed to 0°C over five hours. Samples were washed four times in pre-chilled 100% acetone at 0°C. The last wash was one hour. Samples were then transferred to room temperature and washed two times in 100% acetone. Samples were infiltrated with 1:1 Polybed812 (Polysciences) resin:acetone for 24 hours, 2:1 resin:acetone for 36 hours, 100% resin for 24 hours, and then changed to fresh resin for three days. All infiltration steps were conducted on an orbital shaker at room temperature. Samples were then submerged into embedding molds with resin and hardener and baked at 60°C for two days. 70 nm sections were collected using a PowerTome PC ultramicrotome with a diamond knife and collected onto formvar-coated copper slot grids. Sections were imaged with a Phillips CM200 TEM. Quantification of sarcomere size, epidermal thickness, and body area were calculated using FIJI measurement tools. Pseudocolor overlays were created with Adobe Photoshop using the Wormatlas tissue color code scheme (http://www.wormatlas.org/colorcode.htm). Multiple sections were examined from at least two animals at each time point.
C. elegans and H. glycines were fixed in 4% paraformaldehyde and 2.5% glutaraldehyde (Electron Microscopy Science) for 15 minutes at 4°C. Fixed nematodes were washed three times with PBST (8 mM Na2HPO4, 150 mM NaCl, 2 mM KH2PO4, 3 mM KCl, 0.05% Triton X-100, pH 7.4) [31]. To increase the permeability of fixed C. elegans, nematodes were washed three times after fixation with PBST and shaken in 5% β-mercaptoethanol/PBST (Sigma-Aldrich) at 37°C overnight [31]. C. elegans were then washed three times with 1 mM CaCl2/1% Triton X-100/0.1 M Tris-HCl (pH 7.5). 200 U/ml collagenase type IV was added to C. elegans and shaken vigorously for 15–30 minutes at 37°C, and washed three times with PBST. To increase the permeability of fixed H. glycines, specimens were cut into small segments [33]. H. glycines sections were incubated in 100 mM Tris/1mM CaCl2 and proteinase K (2 mg/ml) for 20–30 minutes at room temperature to facilitate antigen retrieval. Nematode segments were incubated in prechilled methanol on ice for one minute followed by one minute on prechilled acetone on ice. Segments were washed three times with PBST. After permeabilization, both nematode species were incubated in 1% BSA (Sigma-Aldrich) dissolved in PBST for at least one hour followed by incubation in a 1:100 anti-GABA (rabbit; Sigma-Aldrich) primary antibody at 4°C overnight. Nematodes were washed three times with PBST and incubated in 0.1%BSA/PBST with a 1:100 secondary anti-rabbit IgG-FITC (goat; Sigma-Aldrich) antibody at a 1:100 ratio overnight before examination. Antibody staining was observed on over 40 J2 H. glycines and 10 sedentary H. glycines.
Fresh H. glycines J2s were homogenized in liquid nitrogen to extract RNA using the Qiagen RNeasy Blood Kit (Qiagen) following the manufacturer’s instruction. Genomic DNA contamination was removed from the RNA using the Turbo DNA-free kit (Thermo Fisher Scientific). cDNA was synthesized using the Thermoscript RT-PCR system (Invitrogen). hg-unc-25 was amplified from H. glycines cDNA using primers 5’-ATGAAATTAAAGGAGCATAAAGAATC-3’ and 5’-TCACAAATCTTCTCCCAGTGTG-3’. The plasmid pPD157.60 (gift from Andrew Fire via Addgene) was used as the backbone for the construct. Forward: 5’-CATTTTTTCTACCGGTACCAATACG-3’ and reverse: 5’- ACATTTTTTTTTCTCTTTCCGTCTC-3’ primers were used to amplify the backbone including a 1.9 Kb unc-25 promoter region from pPD157.60 (Addgene). The backbone and hg-unc-25 cDNA were ligated using Gibson Assembly (ThermoFisher Scientific) following the manufacturer's instruction. Three 51-bp artificial introns were added into the hg-unc-25 cDNA to improve expression (Andrew Fire Vector Kit) using Gibson Assembly. The rescue construct unc-25p::hg-unc-25 (40 ng/ul) was co-injected with the coelomocyte marker plasmid coel::RFP (80 ng/ul) to rescue the C. elegans unc-25(e156) mutant using standard microinjection techniques [48]. Five independent stable lines were examined for rescue of GABA immunoreactivity. At least 30 nematodes from both the rescue lines and controls were examined.
Synchronized developmental stages of H. glycines were homogenized in liquid nitrogen. RNA was extracted and genomic DNA contamination was removed from the RNA. RT-qPCR was performed using the Power SYBR Green RNA-to-Ct 1 step kit on an ABI PRISM 7000 Sequence Detection System. Expression of hg-unc-25 was examined using the primers qhgunc25 forward primer 5’-TCCAAAGGGATGGAAGGTTATC-3’ and reverse primer 5’-GCCTTCAGTGCGTTTGATTT-3’. far-1, a surface associated retinol and fatty-acid protein encoding gene was compared for the relative expression of hg-unc-25 [49,50]. Each treatment group comprised three independent replicates and the experiment was repeated twice. Graphpad Prism 6 was used for statistical analysis. For the relative expression of hg-unc-25, the ΔCt between hg-unc-25 and hg-far-1 at each stage was calculated, and a one-way ANOVA was performed to calculate the mean separation [51]. Normalized data are shown.
|
10.1371/journal.pcbi.1003696 | Modeling Bi-modality Improves Characterization of Cell Cycle on Gene Expression in Single Cells | Advances in high-throughput, single cell gene expression are allowing interrogation of cell heterogeneity. However, there is concern that the cell cycle phase of a cell might bias characterizations of gene expression at the single-cell level. We assess the effect of cell cycle phase on gene expression in single cells by measuring 333 genes in 930 cells across three phases and three cell lines. We determine each cell's phase non-invasively without chemical arrest and use it as a covariate in tests of differential expression. We observe bi-modal gene expression, a previously-described phenomenon, wherein the expression of otherwise abundant genes is either strongly positive, or undetectable within individual cells. This bi-modality is likely both biologically and technically driven. Irrespective of its source, we show that it should be modeled to draw accurate inferences from single cell expression experiments. To this end, we propose a semi-continuous modeling framework based on the generalized linear model, and use it to characterize genes with consistent cell cycle effects across three cell lines. Our new computational framework improves the detection of previously characterized cell-cycle genes compared to approaches that do not account for the bi-modality of single-cell data. We use our semi-continuous modelling framework to estimate single cell gene co-expression networks. These networks suggest that in addition to having phase-dependent shifts in expression (when averaged over many cells), some, but not all, canonical cell cycle genes tend to be co-expressed in groups in single cells. We estimate the amount of single cell expression variability attributable to the cell cycle. We find that the cell cycle explains only 5%–17% of expression variability, suggesting that the cell cycle will not tend to be a large nuisance factor in analysis of the single cell transcriptome.
| Recent technological advances have enabled the measurement of gene expression in individual cells, revealing that there is substantial variability in expression, even within a homogeneous cell population. In this paper, we develop new analytical methods that account for the intrinsic, stochastic nature of single cell expression in order to characterize the effect of cell cycle on gene expression at the single-cell level. Applying these methods to populations of asynchronously cycling cells, we are able to identify large numbers of genes with cell cycle-associated expression patterns. By measuring and adjusting for cellular-level factors, we are able to derive estimates of co-expressing gene networks that more closely reflect cellular-level processes as opposed to sample-level processes. We find that cell cycle phase only accounts for a modest amount of the overall variability of gene expression within an individual cell. The analytical methods demonstrated in this paper are universally applicable to single cell expression data and represent a promising tool to the scientific community.
| With the advent of single cell expression profiling [1]–[4], the assessment of cell population heterogeneity and identification of cell subpopulations from mRNA expression is achievable [5]–[7]. However, at the single cell level, there is concern that cell cycle might interfere with the characterization of gene expression variability [8]. As many biological samples are prepared from asynchronous cell populations, where each cell is in an unknown phase of the cell cycle, it is imperative to understand the impact of cell cycle in order to account for its effect on observed expression patterns and downstream data analysis. Here, we have measured mRNA expression and cell cycle from 930 single cells derived from three cell lines in order to explore this hypothesis.
A distinctive feature of single-cell gene expression data is the bimodality of expression values. Genes can be on (and a positive expression measure is recorded) or off (and the recorded expression is zero or negligible) [9], [10]. This dichotomous characteristic of the data prevents use of the typical tools of designed experiments such as linear modeling and analysis of variance (ANOVA). We develop a novel computational framework to overcome this problem. First, a probabilistic mixture model-based framework allows the separation of positive expression values from background noise using gene-specific thresholds. After signal separation by thresholding, we model separately the frequency of expression (the fraction of cells expressing a gene) and the continuous, positive expression values. Our semi-continuous framework combines evidence from the two salient parameters of single cell expression in a statistically appropriate manner, an approach dubbed the Hurdle model [11], [12]. Extending our previous proposal of a two-sample semi-continuous test akin to the two-sample t-test, our new framework allows for testing arbitrary contrasts and allows the use of variance components/mixed models, thus bringing to bear the full power of the general linear model.
The Hurdle model allows us to identify many genes with an archetypal cell cycle expression pattern despite a frequently bimodal distribution of expression. It also suggests that stochastic variation in single cell gene expression is relatively large compared to the effect of cell cycle. We find that even in the most tightly regulated gene, cell cycle explains only 27% of the variability, while in the median gene in our data set, cell cycle explains 5%–18% of the variability, depending on the assumptions we make regarding latent technical variability. The semi-continuous model also provides a framework for estimating co-expression networks– in which edges connect genes whose partial correlations remain after removing the effect of all other genes – while adjusting for population-level nuisance factors that could bias network inference. Applying this framework to our data, we show that only a subset of canonical cell cycle genes are highly co-expressed in single cells.
In order to assess differential expression associated with actively cycling cells, expression of 333 genes was interrogated in 930 cells, across three cell lines: H9 (HTB-176), MDA-MB-231 (HTB-26), and PC3 (CRL-1435) (Figure 1A). Single cell expression was measured from flow-sorted cells and compared between cell cycle phases and cell lines via nCounter single cell profiling, a multiplexed hybridization-based detection technology that utilizes fluorescent barcodes to count individual target nucleic acid molecules [13]. This platform has been recently adapted to enable expression profiling from single cells via hybridization after a multiplexed target enrichment (MTE) in which mRNA is first converted to cDNA and then amplified [14].
Each cell was categorized as being in G0/G1, S or G2/M phase by measuring DNA content via flow-cytometry based on retention of Hoechst dye (Figure 1B and S1) [15]. Probes were selected for cell cycle associated genes (n = 119). These genes provided coverage of the entire cell cycle (Data Set S1) based on peak expression and periodicity information obtained from Cyclebase, an integrated database of bulk cell cycle expression profiling experiments that scores and ranks genes based on strength of evidence for a cell cycle associated expression pattern [16]. Probes were also included for non-cell-cycle associated genes with primary roles in the inflammatory response, and housekeeping controls without a Cyclebase ranking (n = 214). We denote probes with a Cyclebase rank (i.e. genes with the strongest evidence for cell cycle associated periodic expression) as the ranked set.
253 genes were expressed and passed quality control (see Methods). Genes showed a bimodal expression pattern in log-transformed mRNA levels (Figure 2), consistent with a burst-model of “on/off” transcription at the single cell level [17] and consistent with the kinetics of PCR amplification with low starting template concentrations, described by us and other authors [9], [10].
Expression levels for each gene were most different between cell lines (Figure 2). Many genes, including those in the ranked set showed cell line-specific expression patterns. For example, expression of TOP2A in G0/G1 varied from 70% of cells in MB-231 and PC3 to nearly universal in H9. This cell line effect was a nuisance factor we needed to adjust for in differential expression tests on cell cycle.
Nonetheless, many genes from the ranked set, such as KIF23, TOP2A, HJURP, NUSAP1, and TPX2 exhibited expression patterns consistent with cell cycle regulation (Figure 2). Figure 2 also reveals that changes in both the positive expression mean (i.e. the mean over the cells expressing that gene; PEM), and changes in the frequency of cells expressing a gene, occur throughout the cell cycle. The frequency and PEM in these genes also vary widely between cell lines, so it was important to adjust for cell line effects for accurate assessment of differential expression.
In order to test for significant differences in expression between cell cycle phases that were consistent across cell lines, we developed an ANOVA-like model (Hurdle model, see Methods) that permits adjustment for additive effects due to cell line. The Hurdle model improves the power to detect changes in single-cell expression by testing both the frequency of expression (corresponding to the relative distribution of cells between the two modes), and the PEM. Combining evidence from the discrete and continuous components of the data provides better sensitivity to changes in expression compared to test statistics based on frequencies of expression (discrete) or on the PEM (continuous) alone; or a union test (see Materials and Methods) while remaining competitive in specificity (Figures S3, S4)
Within the three cell lines tested here, significant differential expression (Bonferroni-adjusted for 253 tests at P<0.05) was observed for 78 genes in the ranked set and 28 genes in the unranked set (Figure 3A). Genes showing the strongest cell cycle associated expression patterns in bulk measurements were more likely to be identified as significant in the single-cell populations (Figure 3A–B).
For each gene, peak time was determined based on the phase (G0/G1, S or G2/M) with maximum average expression across all cell lines. Despite large cell-line-specific expression variability, peak times were broadly consistent with Cyclebase annotations (Figure 3C), and especially so within the subset of genes with strongest evidence of cycle regulation in our data (e.g. Bonferroni significant at P<0.05).
The majority of genes in the unranked set (115/143 or 80%) did not exhibit significant cell cycle effects, in concordance with their primary roles in functions unrelated to the cell cycle. Of the 28 unranked genes that exhibited a significant cell cycle phase association, we noted genes involved in cytoskeletal organization (PLAT), proliferation (PDGFA), and signaling pathways (IFNA1, IFNB1) that have been previously demonstrated to modulate progression through the cell cycle [18].
It has been argued that a substantial portion of the stochastic variability observed in single cell gene expression experiments may be caused by global changes in transcription due to cell cycling [19]. We explore this idea by examining the proportional change in the Hurdle model fit associated with inclusion and omission of cell cycle as an explanatory variable. Because the Hurdle model accounts for both the dichotomous (on/off) and continuous nature of single cell data, the change in deviance (generalized linear model log-likelihood) between nested models can be used to calculate the amount of variability explained by cell cycle. The total deviance can be partitioned into components corresponding to cell cycle effects, nuisance effects described below, and residual effects. The ratio of cell cycle deviance to the sum of cell cycle plus residual deviance can then be interpreted as the analog to the coefficient of determination in linear least squares.
We consider expression changes due to main effects and interactions of cell cycle by cell line and account for amplification efficiency and average cell line effect (see Materials and Methods). Only modest amounts of the single cell expression variability can be explained by cell cycle (Figure 4). Within the ranked gene set, cell cycle phase explains 8% of the deviance in the median gene and 27% of the deviance in the top gene (TOP2A). In unranked genes, phase explains only 5% of the deviance in the median gene.
To derive these estimates, it is important to be able to account for the nuisance factors by using the Hurdle model. If cell-to-cell variation in amplification efficiency is not removed, we underestimate the explanatory power of cell cycle on in the median ranked gene by 26% since the unmodeled deviance would include this large additional component. Similarly, other unmeasured factors may inflate the residual deviance and attenuate the apparent role of cell cycle. These factors could include errors in inferring the cell cycle phase via FACS or imperfect modeling of changes in amplification or detection efficiency between samples. To guard against this attenuation, we set an upper bound on cell-cycle-dependent variation as follows: We suppose that transcription of the gene with the most deviance attributable to cell cycle (TOP2A, 27%) would be entirely regulated in a phase-dependent manner, and we characterize other genes' cell-cycle-dependent deviance relative to this maximum. For example, a gene with 13.5% cell-cycle-dependent deviance has half as strong a cell cycle effect as TOP2A, leading to the conclusion that at most 50% of this gene's deviance could be attributable to cell cycle. Even under these generous upper bounds, cell cycle phase explains only 18% (eg, .05/.27) and 29% (eg, .08/.27) of the deviance in the median gene in the unranked and ranked sets, respectively, suggesting that even when allowing for cell line-specific cell cycle effects, cycle is generally a small factor, compared to residual variability, in gene expression variability in the human transcriptome.
Single-cell gene expression data sets have the resolution to reveal not only differential expression in response to biological variables like cell cycle phase, but also to provide insight into co-expression between genes at the cellular level (e.g. the influence of one gene on another's expression or the sharing of upstream regulatory elements). In bulk-gene expression data (e.g. microarrays), apparent co-expression arises from tissue-level factors inducing shared marginal changes in genes. For example, different radiation doses in samples will induce correlation amongst all the genes affected by radiation, regardless of whether these genes interact or even participate in the same biological processes. In contrast, single cell data allow isolation of co-expression arising from cellular-level factors, giving access to more fundamental biological relationships. If two genes are correlated across cells drawn from the same environment, then the two genes are likely to share an intimate biological relationship: they may be regulated by the same transcription factor, or one gene may directly regulate the other. The distinction between cellular and marginal co-expression follows from a probabilistic identity on conditional covariances (see Materials and Methods).
When cell cycle is not adjusted for (Figure 5 D–F), known cell cycle genes with strong evidence of marginal regulation comprise the majority of the network. These genes generally peak in phase G2/M, suggesting that the co-expression is mostly driven by the coincident peak in average expression. The networks adjusted for cell cycle at least partially remove marginal effects (Figure 5 A–C). In some cell cycle genes, substantial evidence for co-expression remains, but now additional co-expression is detected in genes without a previously described cell cycle role. In the unadjusted estimates, marginal shifts in expression in canonical cell cycle genes overwhelm subtler co-expression in unranked genes. Even though cell cycle variability is modest compared to residual variability, cell cycle is a substantial source of biological variability in the ranked genes and is in a sense confounded with the co-expression patterns.
In an attempt to quantify the performance of the Hurdle model and the effect of cell-cycle adjustments, we examined network properties when varying the number of edges. We call an edge peaktime concordant if it connects nodes that have the same peaktime annotated in cycle base (eg, G0/G1-G0/G1 or S-S). Over a range of network densities (30–240 edges) the unadjusted Hurdle or Raw networks contain between 45%–80% peaktime concordant edges, while the adjusted Hurdle contains only 32%–38% peaktime concordant edges.
Cell cycle adjustment in networks estimated on the raw data is not very effective compared to the unadjusted, raw networks (Figure S6). This is unsurprising, as this would occur when the model for the mean of the response is mis-specified, as is true when ignoring the bi-modality that the data exhibit (eg, Figures 2 and S2). If the Hurdle model is correct and cell cycle is additive, then the identity link cannot recover this additivity. On the other hand, the Hurdle model can still recover an additive mean model under a linear link by taking the discrete coefficient estimates to be null. Overall, the adjusted and unadjusted Hurdle networks in Figure 5 are rather different, sharing 39% of nodes (Jaccard similarity) and 51% of edges (Hamming Distance/#edges).
Combining both discrete and continuous networks (with the top 30 edges from discrete and continuous networks) allows a richer set of genes to be characterized. When discrete expression is used alone, networks primarily consist of G2/M peaking genes and unranked genes (Figure 5A). When positive, continuous expression is also used, S and G0/G1 peaking genes enter the networks (Figure 5B–C).
The adjusted, semi-continuous network depicted in Figure 5C consists of two primary sub-networks, one consisting entirely of ranked genes, and another largely consisting of weakly ranked and unranked genes. While we cannot rule out that measurement error of the inferred cycle is not partially responsible for the persistence of a subset of ranked genes, previously described mutual regulation in RNA-interference experiments [20] of some of these genes suggests that this subset is co-expressed at the single cell level as opposed to being co-expressed on average at the population level. The sub-network of ranked genes contains the central node of NUF2, a highly-conserved protein required for stable kinetochore localization of centromere-associated protein E (CENP-E) [21]. NUF2 is connected to other actors in mitotic organization such as ANLN, KIF23, and CENPF, as well as the check-point genes CCNA2 and BUB1, reflecting the central role of these genes in mitosis.
The sub-network of primarily unranked genes contains two key nodes: TUBB and CCR3. The predominance of genes associated with cell growth, like TUBB, and transmembrane proteins, like CCR3, in the unranked cluster is likely related to the actively dividing nature of the profiled cells, i.e. dividing cells must generate new scaffolding and membrane-related materials to support growth. This relatively large sub-network of unranked and weakly ranked genes is largely missed by the unadjusted analysis that is biased by the population level cell-cycle effect.
Stochastic, bimodal expression is a hallmark of single cell data [22]–[24]. Within a population of cells, detectable expression for any given gene typically resides in one of two modes, corresponding to an “on” or “off” state. Both technical and biological factors likely contribute to this bimodality. Quantities of some species of cDNA may be minute after reverse-transcription, and in this case random variation in the number of template-primer-enzyme complexes that form during each annealing phase may dominate the kinetics of the PCR [25]. But regardless of its origin, modeling bimodality improves the power of differential expression tests.
Here, we show how the Hurdle model can be adapted to complex study designs, extending our previous results describing its use for two-sample comparisons. We demonstrate the model's ability to identify many genes with a periodic expression pattern from asynchronously cultured cells utilizing a combination of FACS sorting and these new analytical techniques, including genes with little previous evidence of cell cycle associated periodic expression like MEF2D [26] and FAM189B. The Hurdle model is able to identify phase-dependent patterns of expression despite the fact that G2 and M phases are indistinguishable by DNA content. The similar rank ordering of differentially expressed genes in our single cell experiment as compared to bulk experiments and concordance in the phase of peak expression demonstrates the power of the Hurdle model. While we have applied the Hurdle model to our specific problem, the approach is general and can be applied to test any effect of interest in a single-cell gene expression dataset. We offer this modeling framework as an R package for other interested users at github.com/RGLab/SingleCellAssay.
Although we recommend the Hurdle model in general for testing for differential expression, it should be noted that its desirability is contingent on the frequency of the gene under consideration. For example, if a gene is highly expressed (eg, >90% expression), then the information to be derived from the 10% of cells that do not express a gene may not be worth the cost of an extra degree of freedom in the chi-square null distribution of the test statistic. However, even when this is the case, the Hurdle model might be preferred for methodological simplicity, since it is powered—although perhaps not always optimally—regardless of expression frequency, and does not require extensive pre-test simulations of power to yield acceptable performance. The data set considered here offers a relatively stringent test of the relative sensitivity of the Hurdle model, owing to the high expression frequency of the genes in this experiment (interquartile range ranked genes: .7–.9; unranked genes: .56–.88).
Single cell data also allows unparalleled resolution of genes' co-expression patterns. While bulk expression data can reveal correlation induced by varying biological conditions, single-cell data has the possibility to reveal co-expression driven by shared regulatory elements within the cell. However, when inferring gene expression networks, it is important to adjust for population level covariates that could confound the network estimation, especially for genes that are marginally affected by such a population level covariate (like known cell cycle genes in our experiment.) By measuring a limited set of cell cycle associated genes, we are able to identify a network of co-expressed genes with known roles in cell cycle regulation even after adjusting for cell cycle phase. It should be noted that the unadjusted network estimate would be appropriate in some circumstances, for example when a summary of the co-expression occurring on average in the population of cells is desired, as opposed to inference of co-expression occurring conditionally within defined subsets.
Work remains to derive network estimators that optimally combine information from discrete and continuous portions. Our current approach is likely theoretically naïve, since it is essentially a union test of the discrete and continuous portions, rather than a summation of signal from the two domains. We also have left unresolved the asymptotic consistency of our proposed network procedure under dimensional scaling.
It is crucial to understand the relationship between cell cycle and the stochastic nature of single cell expression as it determines the magnitude of the cell cycle's distorting effect on single cell analyses. In contrast to earlier estimates of Zopf et al. [19] we find little evidence of periodic regulation of expression among non-cell cycle associated genes. Our results are consistent with genome-wide mRNA profiling efforts utilizing bulk expression methodologies in mammalian cells where genes with cycle-dependent periodic expression patterns are limited and well-characterized [16], [27], [28]. Disparity between our findings and those of Zopf et al. may arise from differences between yeast and mammalian cells. Moreover, Zopf et al. primarily focus on a single, synthetic promoter while we sample hundreds of transcripts presumably driven by many different promoters. Whether the substantial remaining variability is inherent to the human single cell, or due to thus far latent, unmeasured biological variables remains to be explored.
Three human cell lines H9 (HTB-176), MDA-MB-231 (HTB-26) and PC3 (CRL-1435) were commercially obtained and cultured as recommended by the supplier (ATCC). Cultured cells were re-suspended in culture media containing Hoescht 33342 (Sigma) and incubated at 37°C for 60 minutes prior to sorting.
Cultured cells were flow-sorted to isolate individual cells from each of the cell lines according to phase (G0/G1, M/G2 and S). Cells were isolated and sorted using the FACSJazz (Becton Dickinson) at 500 events per second using a 100 micron nozzle. Single cells were defined by gating on forward and side scatter area/width. Phase was inferred from Hoescht 3342 DNA-fluorescent dye, then cells were individually deposited and lysed in wells of a 96-well PCR plate containing 3 uL of Cells-to-Ct lysis buffer (Life Technologies). The proportion of cells in G0/G1 phases varied from 54% of PC-3 cells to 73% of H9 cells (Supplementary Figure S1).
A set of 333 probes was designed. It contained cell cycle associated genes and provided coverage of the entire cell cycle based on peak expression and periodicity information derived from an integrated database of cell cycle expression profiling experiments [16]. Non-cell cycle associated genes had primary roles in the inflammatory response and included housekeeping controls without a Cyclebase ranking. Genes with a Cyclebase ranking <1000 were placed in the ranked set (n = 119) and all other probes were considered part of the unranked set (n = 214).
After lysis, RNA was converted to cDNA with SuperScript VILO (Life Technologies). Primers for 333 genes were pooled and cDNA was enriched in a multiplexed amplification (MTE) reaction according to the nCounter Single Cell Expression protocol (NanoString). The MTE samples were hybridized overnight at 65°C with an nCounter CodeSet containing probes for all enriched targets (cell cycle related, unrelated genes and controls) and internal controls as recommended by the manufacturer.
In single cell gene expression, we have previously found that accounting for both changes in the frequency of expression and shifts in the PEM produces more sensitive measures of differential expression compared to using either the frequency or the positive values alone, or compared to t-tests on the zero-inflated values [9], [33]. We sought to extend this framework to any model that permits a likelihood ratio test on parameters, e.g., generalized linear or generalized linear mixed models, in order to account for additive cell line effects. Let denote the expression threshold in the kth cell (so thus suppressing the gene index). Then we model(1)where , are cell line effects, , are cell cycle effects and , are interaction effects between cell line and cell cycle, and is an independent, normally distributed error. The indices and give the cell line and cell cycle of the k th cell. The cell line effects , and cell cycle effects , are vectors in , although with the linear constraint that the sum of them is zero, eg, , while , is a matrix in with the constraints that for and for .
The term accounts for cell-to-cell technical variability resulting from variation in reverse transcription and PCR amplification efficiency (see previous section). Jointly modeling the PCR efficiency along with the biological effects of interest is important as one factor can affect the other. Our modeling framework can be extended to regression-type models when the right hand side is replaced with a general term for each component, and even to generalized linear mixed models.
In general, let be a vector of parameters for the distribution of and let be a vector of parameters for . Then when the distribution of is divided in this fashion, inference about proceeds conditional on . The log likelihood is then additive in the and parameters. Classical hypothesis tests with chi-square asymptotic null distribution, such as Wald or likelihood ratio tests on specific components of and are null can be conducted separately. Then the test statistics are added together, combining and summarizing the evidence from the two processes, with the degrees of freedom in the null distribution doubled for the purpose of assigning significance. This approach is dubbed the “Hurdle” model and has been used in economics for several decades [34], [35].
We extend the conditional, neighborhood-based algorithm of Meinshausen-Bulmann [36] to estimate co-expression networks using the Hurdle model. The standard Meinshausen-Bulmann algorithm uses L1-penalized regressions to estimate partial correlations between vertices (genes) by treating each vertex as a dependent variable in a regression that includes all other vertices as independent variables. If the vertices are jointly Gaussian, non-zero coefficients correspond to statistical dependences between vertices, conditional on all other factors and so reflect a Gauss-Markov Random Field. Here, since the distribution of expression in single cells is not multivariate Gaussian, edges in our network correspond to conditional correlations (after possible application of the logit link). Although we do not attempt to show consistency of our proposed approach here, we note that Meinshausen-Bulmann-like methods have been shown to be consistent in estimating non-Gaussian graphical models under fairly general conditions [37], [38].
Then for the k th cell, following equation (1), we divide expression into discrete and continuous components, so fit regressions of the form(3)where is the expression of the gth gene in the kth cell, and is the expression vector of all except the gth gene in the kth cell, and is a vector of cellular covariates (eg pre-amplification effect, cell line, cell cycle, and their interaction). We estimate and separately, with distinct L1 penalties and for and using the R package glmnet [39]. Unpenalized vector parameters and adjust for pre-amplification effect ; cell line and cell cycle.
|
10.1371/journal.pbio.0050257 | Nanosensor Detection of an Immunoregulatory Tryptophan Influx/Kynurenine Efflux Cycle | Mammalian cells rely on cellular uptake of the essential amino acid tryptophan. Tryptophan sequestration by up-regulation of the key enzyme for tryptophan degradation, indoleamine 2,3-dioxygenase (IDO), e.g., in cancer and inflammation, is thought to suppress the immune response via T cell starvation. Additionally, the excreted tryptophan catabolites (kynurenines) induce apoptosis of lymphocytes. Whereas tryptophan transport systems have been identified, the molecular nature of kynurenine export remains unknown. To measure cytosolic tryptophan steady-state levels and flux in real time, we developed genetically encoded fluorescence resonance energy transfer nanosensors (FLIPW). The transport properties detected by FLIPW in KB cells, a human oral cancer cell line, and COS-7 cells implicate LAT1, a transporter that is present in proliferative tissues like cancer, in tryptophan uptake. Importantly, we found that this transport system mediates tryptophan/kynurenine exchange. The tryptophan influx/kynurenine efflux cycle couples tryptophan starvation to elevation of kynurenine serum levels, providing a two-pronged induction of apoptosis in neighboring cells. The strict coupling protects cells that overproduce IDO from kynurenine accumulation. Consequently, this mechanism may contribute to immunosuppression involved in autoimmunity and tumor immune escape.
| Although regulated suppression of the immune system prevents autoimmunity and is important during pregnancy to protect the fetus or after organ transplant to block graft rejection, it can be harmful if co-opted by tumors to escape detection. T cells of the immune system normally recognize and destroy abnormal cells, including cancerous and grafted tissues. This process requires the amino acid tryptophan. Foreign grafts and cancer cells can dampen the immune response by starving T cells of tryptophan through a mechanism involving uptake and conversion to kynurenines using an enzyme called indoleamine-2,3-dioxygenase (IDO). Independent of tryptophan starvation, kynurenines induce T cell death when they are excreted through an unknown mechanism. We constructed fluorescent nanosensors that report tryptophan levels in two immortalized cell lines, COS-7 cells and human oral cancer KB cells. Using single-cell, real-time imaging analysis, we demonstrate that an L-amino acid transporter protein (LAT1) exchanges tryptophan for its kynurenine degradation products. LAT1 together with IDO forms a metabolic minicycle that couples tryptophan starvation with kynurenine-induced cell death, providing a two-pronged inactivation of neighboring cells. The strict coupling protects the cells that overproduce IDO from accumulating kynurenine. This cycle may contribute to suppression of the immune response in autoimmune diseases and cancer.
| l-tryptophan is an essential amino acid necessary for protein synthesis in mammalian cells. Moreover, tryptophan is the precursor for the neurotransmitter serotonin, for the hormone melatonin, and contributes to the synthesis of the coenzymes NADH and NADPH. Degradation products of tryptophan have immunoregulatory functions. Mammalian cells cannot synthesize l-tryptophan, and thus depend on transport machineries for its uptake and protein turnover for its production. Because cell growth is strictly dependent on tryptophan serum levels, different cell types compete for this amino acid, for instance in the regulation of the immune response. Identified transporter proteins potentially involved in the uptake of tryptophan in human cells include the Na+-independent systems b0AT1 [1]; b0,+AT [2]; TAT1 [3]; y+LAT1 and y+LAT2 [4,5]; and LAT1, LAT2, LAT3, and LAT4 [6–9]. Of these, b0,+AT, LAT1, LAT2, y+LAT1, and y+LAT2 are amino acid exchangers; they countertransport cytosolic amino acids for extracellular ones. LAT1 is present in proliferative tissues and, as such, is up-regulated in many human primary tumor cells [10].
Tryptophan can be degraded through the kynurenine pathway, e.g., for the biosynthesis of vitamin B3 or niacin, which is needed for the synthesis of coenzyme NADH or NADPH. The rate-limiting step in this pathway is the opening of the indole ring by indoleamine 2,3-dioxygenase (IDO). Since the discovery that inhibition of IDO induced fetal allograft rejection in mice, the immunosuppressive function of tryptophan catabolism has been well established [11]. One proposed mechanism for the observed immunosuppression is the local depletion of tryptophan, which inhibits adaptive T cell responses by forcing them into growth arrest and inducing apoptosis [12]. As such, the immune escape of many cancer cell types correlates with up-regulated IDO expression and can, in some cases, be overcome by IDO inhibition [13,14]. In addition, products of the kynurenine pathway are immunosuppressive and may provide leads for the treatment of autoimmune disorders such as multiple sclerosis [15]. Kynurenines are natural amino acids found in mammals; however, the transport machinery for their export across the cell membrane is not known.
Traditionally, cellular uptake of molecules has been determined using radiolabeled substrates, and cellular levels have been measured in extracts via liquid chromatography or gas chromatography/mass spectrometry. Both methods are neither time resolved nor specific, and they lack high temporal or cellular/subcellular resolution. Tryptophan is aromatic, binding nonspecifically to many molecules. Given the importance of l-tryptophan for human health, an analytical tool for noninvasive, time-resolved determination of intracellular l-tryptophan levels was deemed desirable.
Fluorescent indicator proteins (FLIPs) are new tools for real-time monitoring of metabolite levels and have been used successfully for monitoring several small molecules in subcompartments in living cells. Typically, the nanosensors consist of a ligand-sensing domain, allosterically coupled to a pair of green fluorescent protein variants capable of fluorescence resonance energy transfer (FRET). FRET requires donor and acceptor fluorophores with overlapping emission and excitation spectra, respectively. After excitation of the donor, energy is transmitted to the acceptor in a nonradiative manner and emitted by the acceptor. The efficiency of this process depends on the distance between and relative orientation of the dipoles of the fluorophores. Ligand binding–induced conformational changes in the sensors result in altered FRET efficiencies, which correlate with the levels of the respective metabolites. Periplasmic binding proteins (PBPs) have been successfully exploited for the construction of FLIPs for imaging of key metabolites such as glucose [16], maltose [17], ribose [18], and glutamate [19]. However, no tryptophan-binding PBPs have been described to date; thus an alternative ligand-sensing scaffold was explored for construction of a tryptophan nanosensor.
In γ-proteobacteria like Escherichia coli, transcription of the tryptophan biosynthetic operon is regulated by attenuation [20] and by the inhibitory binding of the tryptophan-activated repressor protein, TrpR, to the trp operator [21]. Binding of l-tryptophan to the repressor results in conformational changes that enhance the repressor's affinity for the operator sequence [22]. We have exploited the ligand-induced conformational changes of TrpR for the construction of novel genetically encoded sensors for monitoring of in vivo l-tryptophan levels. First, we demonstrate the applicability of the metabolite FRET nanosensor technology to novel ligand-sensing domains and instigate new methods for the construction of nanosensors for metabolites that are only present inside the cell (and therefore are unlikely to be recognized by evolved PBPs). Second, we use a novel strategy for the optimization of the FRET signal based on the particular topology and conformation of TrpR. Third, we express the optimized FRET tryptophan sensor in KB and in COS-7 cells and find that these cells take up tryptophan by a LAT amino acid exchanger. Finally, we demonstrate that this transporter mediates tryptophan/kynurenine exchange, which, in cooperation with IDO, provides a new metabolic cycle that may contribute to the immune escape of various tumor cells as well as a favorable role of kynurenines in reducing autoimmunity, while at the same time protecting the cells that overproduce IDO from kynurenine accumulation.
The E. coli tryptophan operon repressor TrpR is an all-helical polypeptide of 108 amino acids organized into 6 α helices. This polypeptide forms a dimer that selectively binds two molecules of l-tryptophan with micromolar affinity (Figure 1A) [23]. In the active, dimeric conformation of TrpR, five of the six helices in each polypeptide are involved in intermolecular contacts [24]. With both chains contributing to each tryptophan-binding site, two TrpR polypeptides are necessary to form the two functional intermolecular binding sites (Figure 1A). In the absence of tryptophan, a part of TrpR is unfolded [25], which likely corresponds to the helix-turn-helix motifs that form the DNA-reading heads. Crystallographic analysis shows that the helix-turn-helix motif undergoes structural rearrangements upon binding of tryptophan [22], and the motif's flexibility is essential for the recognition of operator sequences [26]. In addition, tryptophan binding results in a shift of the relative distance and orientation of the N and C termini of each repressor polypeptide with respect to one another [22], which we detected as a change in FRET when TrpR was fused to a FRET fluorophore pair. The E. coli tryptophan repressor gene was sandwiched between enhanced cyan fluorescent protein (eCFP) and Venus (a yellow fluorescent protein variant) coding sequences (Figure 1B). Production of the translated fusion product FLIPW-CTY (CTY: eCFP-TrpR-VenusYFP) in E. coli was readily detected by recording the emission spectrum of the eCFP-Venus FRET signal in cell cultures. When eCFP was excited, significant energy transfer to Venus was detected, resulting in a Venus/eCFP ratio of 4 (Table S1). Addition of l-tryptophan decreased FRET efficiency of the purified protein, visible as an increase in eCFP emission intensity and a concomitant decrease in Venus fluorescence intensity, resulting in a 10% reduction in the Venus/eCFP ratio (Figure 1C). FLIPW-CTY bound l-tryptophan with an apparent dissociation constant (Kd) of 220 ± 20 μM, which is about one order of magnitude higher compared with unmodified TrpR as measured by equilibrium dialysis [23]. Small molecules are known to quench fluorophore emission efficiently due to nonspecific interactions [27]. To exclude that the negative ratio change observed for FLIP-CTY is due to unspecific effects, the FRET response was measured in the presence of d-tryptophan. Compared with l-tryptophan, unmodified TrpR has a 20-fold reduced affinity for d-tryptophan [23]. Titration of FLIPW-CTY with d-tryptophan resulted in a decrease of the FRET ratio at about 5-fold higher concentrations than l-tryptophan (Figure S1). Because d- and l-tryptophan would be expected to have the same quenching properties, this strongly suggests that the decrease in FRET ratio of FLIPW-CTY is due to a specific interaction of the sensor with tryptophan.
The active conformation of TrpR is a dimer, and two tryptophan-binding sites are formed at the dimer interface [28]. Therefore, one assumes that the functional FLIPW-CTY sensor functions as a dimer of two CTY polypeptides, with the four fluorophores being packed tightly together, potentially affecting the binding affinity due to steric hindrance or resulting in signal loss due to averaging of the fluorophore signals. Thus fusing two TrpR molecules to one fluorophore set may give rise to sensors with a single eCFP-Venus pair per sensor, which may have improved sensing characteristics. Three sensor permutations containing two TrpR copies (a and b) in a single gene product were constructed (Figure 1B). In the permutants FLIPW-TCTY (linear arrangement of TrpRa-eCFP-TrpRb-Venus) and FLIPW-CTYT (linear arrangement of eCFP-TrpRa-Venus-TrpRb) the distance between the N and C termini of the intercalated green fluorescent protein variants (eCFP in TCTY and Venus in CTYT) corresponds well to the distance between the C terminus of the first TrpR polypeptide (a) and the N terminus of the second TrpR (b) in the dimer (∼22 Å, Figure S2). One of the fluorophores in these variants is therefore rotationally constrained by these attachment points, a fact that is expected to lead to an improvement of the signal change due to decreased conformational averaging as shown for other “insertional” FRET sensors [29,30]. The third variant, FLIPW-CTTY, is a linear fusion in the order eCFP-TrpRa-TrpRb-Venus. For the construction of FLIPW-CTTY, two copies of the repressor gene were connected by a flexible linker consisting of seven glycine residues and inserted between the fluorophores. This linker was designed to loosely connect the two TrpR proteins without changing the dimer conformation, and is based on a model constructed in Modeller8v1 [31]. Whereas the FRET ratio of FLIPW-CTTY and FLIPW-TCTY changed only slightly when titrated with l-tryptophan, FLIPW-CTYT yielded a substantially improved tryptophan sensor (Figure 1C and 1D). The apparent binding constant of FLIPW-TCTY for l-tryptophan was ∼20 μM, comparable to unmodified TrpR [23]. The ratio change observed for FLIPW-CTTY could not be fitted to a single-site-binding isotherm (see Material and Methods).
When FLIPW-CTYT was titrated with l-tryptophan, an increase in FRET ratio from 2.2 to 2.8 was observed, indicating a significant change in chromophore orientation with respect to FLIPW-CTY. The ratio change observed in vitro for FLIPW-CTYT was +35%. FLIPW-CTYT bound l-tryptophan with an apparent affinity of 100 ± 10 μM. Analogous to the wild-type TrpR, FLIPW-CTYT binds ligands in order of decreasing affinity: l-5-methyl-tryptophan > l-tryptophan > d-tryptophan > l-5-hydroxy-tryptophan (Table S2) [23]. The positive ratio change permits efficient discrimination of quenching effects; thus, FLIPW-CTYT appears better suited for in vivo measurements compared to FLIPW-CTY, which shows a negative ratio change (Table S1). Therefore FLIPW-CTYT was chosen to monitor physiological tryptophan levels in mammalian cells (dynamic range ∼ 15 μM to 1 mM).
Molecular modeling was performed to rationalize the observed FRET signal changes. The original sensor, FLIPW-CTY, is predicted to dimerize, resulting in an antiparallel arrangement of the TrpR polypeptides, thus resulting in two sets of eCFP and Venus fluorophores in close vicinity at both sides of the TrpR dimer (Figure 2A). In agreement with the close vicinity of the fluorophores, FLIPW-CTY showed the highest FRET efficiency (Table S1). The FLIPW-CTYT sensor is modeled to form a functional TrpR dimer intramolecularly, resulting in a single eCFP and a single Venus molecule per sensor (Figure 2B). FLIPW-CTYT has lower absolute energy transfer efficiency, which is consistent with the greater distance between the fluorophore dipoles. The relative FRET change is higher compared to FLIPW-CTY, probably because of the rigidification of the attachment of the Venus molecule by its fusion to both TrpR monomers. The FLIPW-CTTY and FLIPW-TCTY sensors do not show sufficient ligand-dependent ratio changes to be useful as sensors. Molecular modeling may explain the different responses of the FLIPW-CTYT and FLIPW-TCTY sensors. In FLIPW-CTYT, a fluorophore is attached to the N terminus of TrpRa, leading to a different spatial arrangement and rotational probability space compared to FLIPW-TCTY, in which Venus is attached to the C terminus of TrpRb. This geometrical difference is presumably transduced into altered dipole orientations in the FLIPW-CTYT sensor (Figure S3).
To measure tryptophan flux in the cytosol of live cells, COS-7 cell cultures seeded in a 96-well microplate were transiently transfected with pTK222 for cytosolic production of FLIPW-CTYT sensor. Microscopic analysis of transfected cells showed that FLIPW-CTYT was produced in the cytosol. When microwell-grown cells expressing FLIPW-CTYT were incubated in Tyrode's buffer containing tryptophan and analyzed in a microplate reader, an increase in FRET ratio was observed, indicating an increase in cytosolic tryptophan levels as a result of uptake (Figure 3A). The rate in FRET increase depended on the external tryptophan concentration and showed Michaelis-Menten type kinetics with an apparent affinity constant KM of 0.88 ± 0.27 μM for combined transport and metabolism (Figure 3B). The FLIPW-CTYT sensor thus appears suitable to study factors influencing tryptophan transport and metabolism and can be used in high-throughput fluorescence-based assays and drug screens.
Reported intracellular tryptophan levels in mammalian cells cultivated in batch are in the 0.27–0.60 mM range [32]; thus, they are compatible with the detection range of the FLIPW-CTYT sensor. When COS-7 cells producing cytosolic FLIPW-CTYT sensor were perfused with Tyrode's buffer, the initial FRET ratio (528 nm/485 nm [Venus/eCFP] emission intensity) was stable (Figure 4A). Upon perfusion with 100 μM l-tryptophan, the FRET ratio increased instantaneously, corresponding to rising levels of cytosolic tryptophan. The FRET ratio was stable during subsequent perfusion with buffer, indicating that the cytosolic steady-state tryptophan levels remain constant and that in COS-7 cells tryptophan uniporters like TAT1 [3] do not contribute significantly to transport. When l-histidine was provided in the medium, the eCFP emission increased and Venus emission decreased, evidencing a decrease in FRET efficiency as a result of export of tryptophan from the cytosol. This way, cells could be repeatedly loaded with tryptophan and unloaded using histidine (Figure 4A). By comparison of the starting, minimum, and maximum response levels of the sensor, assuming that the Kd of the sensor expressed in cells was the same as in vitro, the cytosolic tryptophan concentration was estimated at 340 μM. The affinity of the cells for combined uptake and metabolism of tryptophan was 2.6 ± 1.1 μM (Figure 4B). The system L inhibitor 2-aminobicyclo-(2,2,1)-heptane-2-carboxylic acid (BCH) decreased the uptake of tryptophan, whereas either replacement of sodium with choline or the transport system A inhibitor N-(methylamino)isobutyric acid did not decrease uptake rates (Figure 4C). Na+ independence and BCH sensitivity are characteristic of the transport properties of LAT1- and LAT2-like amino acid counterexchangers [6]. Similar to histidine, the larger hydrophobic and aromatic amino acids leucine, isoleucine, valine, methionine, phenylalanine, and tyrosine were able to promote tryptophan export (at 100 μM concentrations, unpublished data), an observation that corresponds best to the reported substrate specificity of LAT1 [33–35]. LAT2 is expressed in the proximal tubule of the kidney [6], so it might be expressed in COS-7 cells. In addition, LAT1 is expressed in various human tumor cell lines [35] and, consequently, is likely to be present in COS-7 cells as well.
The product of tryptophan conversion by IDO is formylkynurenine (FK), which is, in turn, converted by the enzyme kynurenine formamidase to kynurenine (K) (Figure 5A). The consecutive action of kynurenine-3-hydroxylase produces 3-hydroxy-kynurenine (HK), which is further degraded by kynureninase to hydroxy-anthranilic acid (HAA) (Figure 5A). In vitro, K, HK, and HAA did not result in a FLIPW-CTYT FRET response or interfere with tryptophan binding to the sensor (unpublished data). We tested the effect of extracellular K, HK, and HAA on the intracellular tryptophan levels in tryptophan-preloaded COS-7 cells. Both extracellular K and HK resulted in a reduction of the fluorophore emission ratios in the COS-7 cells, evidencing tryptophan export. The sensor responses for the K- and HK-induced tryptophan were similar (unpublished data), and HK-induced export is shown in Figure 5B. Extracellular HAA did not result in tryptophan export (unpublished data). Thus, the findings are compatible with the role of LAT-like transporters that are present in COS-7 cells as exchangers of tryptophan for its degradation products K and HK. Tryptophan, K, and HK thus appear to be substrates for LAT1 and one may speculate that FK is a substrate as well. Since the substrate specificities of LAT1 and LAT2 differ only in the transport of small amino acids [33,36], we hypothesize that both are capable of tryptophan-kynurenine exchange.
KB cells originate from a human oral squamous cell carcinoma and express only LAT1 for uptake of neutral amino acids as shown by reverse-transcriptase (RT)-PCR [37] and RNA interference (RNAi) [38]. When KB cells expressing FLIPW-CTYT were perfused with tryptophan, the FRET ratio increased (Figure 6A). The FRET ratio did not change upon subsequent perfusion with buffer, but it decreased upon perfusion with histidine, confirming amino acid exchange. LAT1-mediated uptake and exchange of tryptophan was similar to that observed for the LAT1-like exchanger in COS-7 cells (Figure 6A) and was almost completely inhibited by BCH (Figure 6B.). The affinity of KB cells for combined tryptophan uptake and metabolism under perfusion conditions is 5.1 ± 1.0 μM (Figure 6C), which is 2-fold lower than that observed for the LAT1-like exchanger of COS-7 cells. In addition, intracellular tryptophan was exported upon perfusion with kynurenine (Figure 6D), 3-hydroxy-kynurenine (Figure 6E), but not 3-hydroxy-anthranilic acid (Figure 6F), indicating that human LAT1 mediates the exchange of tryptophan for its kynurenine degradation products.
Mammalian cells cannot synthesize the amino acid tryptophan and rely on its import as tryptophan or as components of nutrients such as peptides across the plasma membrane for basic cell functioning. Tryptophan is necessary for protein synthesis, and it accounts for ∼1.3 % of the amino acids in human proteins. Tryptophan is also the precursor of other vital molecules like serotonin, melatonin, and NAD. Moreover, kynurenines produced from tryptophan appear to play a pivotal role in immunosuppression in inflammatory diseases and cancer [39,40].
The FLIPW nanosensors described in this study allow for noninvasive, real-time, spatio-temporal imaging of intracellular tryptophan levels and flux, offering advantages over conventional analytic methods. The E. coli transcriptional regulator TrpR was used as the recognition element for the construction of FRET sensors for tryptophan. As noted previously, the use of bacterial proteins for the construction of intracellular sensors reduces the problem of interference with endogenous metabolic and signal transduction pathways in eukaryotic cells [41]. Genetically encoded nanosensors further offer the advantage of subcellular sensor targeting through judicious choice of leader sequences as demonstrated with nuclear- and ER-targeted glucose nanosensors [42] and with cell-surface display of glutamate nanosensors [19]. Most FRET nanosensors are based on the ligand-binding–induced Venus fly trap–like conformational changes of bacterial PBPs [16–19], which consist of two well-structured lobes with the ligand-binding site located at the interface. TrpR is about three times smaller than the average PBP and is partially unfolded in the absence of tryptophan [25]. In the presence of tryptophan, the protein adopts the condensed conformation observed in crystal structures [22], and the concomitant conformational changes allow for the detection of tryptophan binding by FRET.
FRET has been a successful reporter signal for small molecule sensors [43,44]. According to the Förster theory, the efficiency of the energy transfer depends on the distance between the fluorophores and their dipole orientation [45]. These small molecule nanosensors can be engineered by modification of linker sequences between reporter and sensing domains and/or insertion of fluorophores in surface loops of the sensing domain, resulting in increased and/or reversed signal outputs of FRET nanosensors [29]. Because TrpR dimerizes to form its ligand-binding sites, we applied a novel approach for engineering of the FRET signal. Insertion of a second TrpR coding sequence in the principal FLIPW-CTY sensor changed the FRET response depending on the position of the insertion site. While insertion before eCFP and between eCFP and Venus almost eliminated the FRET response, a TrpR copy after Venus reversed the FRET response and increased the ratio change. Comparison of structural models of the FLIPW-CTY and FLIPW-CTYT sensors predicted that the fluorophores would be closer together in FLIPW-CTY. Since FRET efficiency is inversely correlated with the distance between the fluorophores as described in the Förster equation [45], the experimentally determined FRET ratio and the models are consistent.
FLIPW-CTYT was used for monitoring tryptophan uptake in cell cultures grown in 96-well microtiter plates, making the sensor suitable for high-throughput assays in which the effect of drugs or small interfering RNAs (siRNAs) is tested systematically [46]. The effective KM for combined tryptophan uptake and metabolism in COS-7 cells in microplate assays and during perfusion was similar with values in the low micromolar range. We found that a LAT1-like transporter is responsible for the observed tryptophan exchange in COS-7 cells. Its transport and exchange characteristics are very similar to that of human LAT1 as expressed in KB cells. The reported affinity of human LAT1 for tryptophan uptake is 21 μM, about 4-fold lower than determined in this study [35]. However, the values are difficult to compare, because the former uptake experiments were performed under static conditions with LAT1 expressed in oocytes and the obtained affinities relate to the sum of intracellular pools of free, incorporated, and degraded tryptophan in the large oocytes. On the other hand, the affinities obtained using FLIPW-CTYT have been determined for the pool of free tryptophan in the targeted subcellular compartment, i.e., cytosol, under perfusion conditions with histidine as a defined counterexchange substrate. Possibly, the physiological conditions of amino acid exchange in the vascular system are mimicked more accurately by perfusion.
The transporters LAT1 and LAT2 are heteromeric obligatory counterexchangers of large, neutral amino acids with a 1:1 exchange stoichiometry (SLC3 and SLC7) [6,36]. As exchangers, they do not change the net intracellular amino acid concentration, but rather modify their relative concentrations. Perfusion of FLIPW-CTYT–expressing COS-7 and KB cells with tryptophan and histidine yielded high-resolution data of the real-time dynamics of free cytosolic tryptophan resulting from system L exchange activity. Importantly, we found that LAT transporters can exchange kynurenines and tryptophan. Since the individual intracellular and extracellular substrate selectivity of the LAT transporters are similar [36], kynurenine-tryptophan exchange may be bidirectional.
Tryptophan-kynurenine exchange may be part of an endogenous immunosuppressive mechanism during autoimmunity and may support the immune escape of proliferative cell types—like cancer cells—by enhancing the depletion of the local tryptophan pool and increasing the serum kynurenine concentrations (Figure 7). Kynurenines and hydroxy-kynurenine are natural amino acids that are produced from tryptophan through IDO, whose enzymatic activity is necessary for immune escape [13]. Macrophages and dendritic cells express IDO for endogenous suppression of the immune system [40,47]. Several human cancer cell lines, including KB cells, up-regulate IDO activity in the presence of the proinflammatory cytokine interferon-γ [48]. In addition, IDO can be up-regulated by the mutation of the tumor suppressor Bin1 [14], which is lost or attenuated in several cancer types [49]. Intracellularly produced kynurenines serve as substrates for the exchange for extracellular tryptophan by LAT transporters, which are expressed in many tissues and human primary tumors [10]. Effectively, tryptophan is sequestered from the local environment and kynurenines accumulate in the serum. The kynurenines contribute to the pool of amino acids that can be taken up in exchange for intracellular tryptophan by surrounding cells expressing LAT transporters, a process which results overall in a tryptophan flux toward the IDO-producing cells. Since resting human T cells express only transporters of system L for the transport of l-tryptophan [50], the tryptophan-kynurenine exchange mechanism helps to deplete the intracellular tryptophan as well. Both the accumulation of kynurenines and depletion of tryptophan arrest T cell growth and induce apoptosis [39,40,51]. Thus, tryptophan-kynurenine exchange results in double trouble for T cells (Figure 7). At the same time, IDO-overproducing cells are protected from the apoptotic effect of kynurenines by the strict counterexchange of tryptophan and its stoichiometric degradation products. Cells, e.g., human macrophages, expressing transport systems with higher affinities for tryptophan than LAT1 will be able to continue to proliferate during ongoing tryptophan sequestration by the combined activity of LAT1 and IDO [50]. FLIPW sensors can now be used to test whether T cells take up kynurenines using the same pathway leading to a further drain of the essential tryptophan. The sensors can also be used to identify novel drugs and regulatory factors in genomic RNAi screens or screens of chemical libraries.
FRET nanosensors are unique tools for studying intracellular, small molecule steady-state levels and fluxes in vivo and in real time. Ultimately, complete metabolic routes can be monitored by using nanosensors that selectively detect single intermediates. For this means, a set of FRET nanosensors has been constructed that use the ligand-induced conformational changes of PBPs [16–19]. As the FLIPW sensors demonstrate, other protein scaffolds that undergo conformational changes upon ligand binding can provide sensing domains for nanosensors with specificities not represented in the PBP family, such as tryptophan. E. coli tryptophan repressor TrpR is not part of a protein family with different substrate specificities, which could be used for the expansion of the current set of nanosensors. However, the wealth of bacterial transcriptional regulators, which change affinity for operator sequences upon binding of effectors, may provide potential sensing domains for novel FRET metabolite nanosensors.
The FLIPW-CTYT nanosensor has proven to be a robust system with multiple advantages over conventional methods for intracellular tryptophan detection. The new sensor thus provides a complementary tool for monitoring steady state levels, uptake, and counterexchange, and will be an important tool for analyzing the factors that control tryptophan flux in living cells. As the kynurenine/tryptophan exchange demonstrates, such factors might contribute to important cellular processes such as inflammation and immune escape.
All chemicals were of analytical grade and purchased from Sigma-Aldrich (http://www.sigmaaldrich.com). E. coli strains DH5α, TOP10 F′, and BL21(DE3)gold (Stratagene; http://www.stratagene.com) were used for transformation of Gateway reactions, cloning, and protein production, respectively.
The E. coli trpR gene [52] (EcoGene EG11029, TrpR: UniProt P0A881) was amplified from genomic DNA by PCR for cloning in plasmid pGWF1 through pDONR using forward primer (5′-GGGGACAAGTTTGTACAAAAAAGCAGGCTCGGCCCAACAATCACCCTATTCAGC-3′) and reverse primer (5′-GGGGACCACTTTGTACAAGAAAGCTGGGTT ATCGCTTTTCAGCAACACCTCTTC-3′) using the Gateway protocol provided by the manufacturer (Invitrogen; http://www.invitrogen.com). Plasmid pGWF1 is based on the pRSETb expression vector (Novagen; http://www.novagen.com) and contains genes for eCFP and Venus cloned under control of the bacteriophage T7 promoter. Between the gene sequences of eCFP and Venus, a chloramphenicol resistance gene and lethal ccdB gene are flanked by attP DNA sequences for insertion of DNA sequences using Gateway cloning technology. The trpR gene was sandwiched between the eCFP and Venus coding sequences resulting in plasmid pTK164. The protein sequence encoded on pTK164 was denoted FLIPW-CTY. By PCR, trpR copies flanked with BamHI or HindIII restriction site sequences were produced. Twin cassette sensor variants were constructed by insertion of trpR copies into pTK164 using unique BamHI and HindIII restriction sites respectively before the eCFP coding sequence (resulting in pTK203) and after the Venus encoding sequence (resulting in pTK204), resulting in sensor variants encoding the repressor dimer in a single gene. A construct in which two trpR copies were connected with a Gly7 linker was denoted pTK205. The gene products of pTK203, pTK204, and pTK205 were denoted FLIPW-TCTY, FLIPW-CTYT, and FLIPW-CTTY, respectively. FLIPW constructs were harbored in E. coli BL21(DE3)gold, and sensor proteins were produced and purified as described previously [17].
Purified sensor was added to a dilution series of ligand in 20 mM MOPS, pH 7.0, in the range of 10−2 to 10−6 M and analyzed in a monochromator microplate reader (Safire, Tecan; http://www.tecan.com; eCFP excitation 433/12 nm, eCFP emission 485/12 nm, and Venus emission 528/12 nm). eCFP shows two emission peaks at 476 nm and 501 nm [53]. The eCFP emission used for the ratio calculation was determined at 485 nm. Protein was diluted to give Venus readouts of 20,000–30,000 at a manual gain of between 70–75. By using the change in FRET ratio upon binding of ligand, affinity constants (Kd) were determined by fitting the titration curves to a single-site binding isotherm:
with [L], ligand concentration; n, number of equal binding sites; R, ratio; Rapo, ratio in the absence of ligand; and Rsat, ratio at saturation with ligand. Three independent protein preparations were analyzed and each protein preparation was analyzed in triplicate.
Structural models of FLIPW sensors were constructed using the crystal structures of Trp repressor in complex with L-Trp and Venus. Proteins were manually docked in the various topologies using MAGE (http://rd.plos.org/pbio.0050257).
For cytosolic expression in COS-7 and KB cells, the gene encoding CTYT was amplified by PCR with primers encoding unique BamHI and EcoRI restriction sites at the 5′ and 3′ end, respectively, and cloned into BamHI/EcoRI digested pcDNA3.1(+) vector (Invitrogen), resulting in plasmid pTK222. COS-7 cells were grown in Dulbecco's modified Eagle's medium (high glucose; DMEM, Gibco; http://www.invitrogen.com) with 10% fetal calf serum and 50 units (U)/ml penicillin and 50 μg/ml streptomycin (Gibco). KB cells were grown in modified Eagle's medium alpha (MEM-α, Gibco) with 10% fetal calf serum and 50 U/ml penicillin and 50 μg/ml streptomycin (Gibco). Cells were cultured at 37 °C and 5% CO2. For imaging, cells were cultured in 8-well LabTekII German tissue culture glass slides (Nalge Nunc International; http://www.nalgenunc.com) and transiently transfected at 50%–70% confluence using Lipofectamine 2000 Reagent (Invitrogen) in Opti-MEM I reduced serum medium (Gibco). After transfection, cells were cultured for 16 h in Opti-MEM I followed by 3 h in growth medium prior to imaging. Transfection efficiency as determined by counting fluorescing cells was at least 30% for COS-7 and up to 5% for KB.
Adherent COS-7 cells in 96-well microplates were washed once with 100 μl Tyrode's buffer (119 mM NaCl, 2.5 mM KCl, 2 mM CaCl2, 2 mM MgCl2, 25 mM HEPES, 30 mM glucose, pH 7.3–7.4). The initial FRET ratio was measured by recording the eCFP and Venus emissions at 485 nm and 528 nm, respectively, after excitation of eCFP at 433 nm in a Safire monochromator microplate reader (Tecan). Standard deviation of the initial ratios was less than 10%. After addition of 100 μl tryptophan in Tyrode's buffer the FRET ratio was recorded with 2-min intervals for up to 2 h. Uptake rates were determined from linear parts in the initial FRET change and fitted with the nonlinear regression program Origin 6.1 (OriginLab; http://www.originlab.com).
Ratio imaging was performed on an inverted fluorescence microscope (DM IRE2, Leica; http://www.leica.com) with a Quant EM digital camera (Photometrics; http://www.photomet.com) and 20× oil immersion, 63× water immersion lenses (HC PL APO 20×/0.7 or HCX PL APO, Leica). Dual emission intensity ratios were simultaneously recorded using a DualView with an OI-5-EM filter set (eCFP 480/30; eYFP 535/40; Optical Insights; http://www.optical-insights.com) and Slidebook 4.2 (Intelligent Imaging Innovations; http://www.intelligent-imaging.com/). A Lambda DG4 (Sutter Instruments; http://www.sutter.com) provided excitation light. Images were acquired within the linear detection range of the camera and exposure times varied between 50 and 200 ms, depending on the expression level, with software binning between 2 and 3. Fluorescence intensities for eCFP and Venus were typically in the range of 6,000–8,000 and 12,000–16,000, respectively. Typical background values were around 1,000. Cells were perfused with Tyrode's buffer supplemented with ligands at flow rates of 1.0 ml/min in a vacuum chamber (Vacu-Cell VC-MPC-TW, C&L Instruments; http://www.fluorescence.com/) with a total volume of 0.1 ml. Buffers were exchanged using an eight-way automated ValveBank (AutoMate; http://www.autom8.com/). Inhibitors BCH and α-(methylamino)isobutyric acid (MeIAB) were used at 5 mM concentrations. For determination of the substrate specificity of tryptophan exchange, all 20 amino acids were tested at 100 μM concentrations. Analyses were repeated at least three times with similar results.
The Protein Data Bank (http://www.pdb.org) accession numbers for structures discussed in this text are: 1TRO (Trp repressor with operator), 1MYW (Venus), and 1WRP (Trp repressor in complex with L-Trp).
|
10.1371/journal.pgen.1006755 | Identifying systematic heterogeneity patterns in genetic association meta-analysis studies | Progress in mapping loci associated with common complex diseases or quantitative inherited traits has been expedited by large-scale meta-analyses combining information across multiple studies, assembled through collaborative networks of researchers. Participating studies will usually have been independently designed and implemented in unique settings that are potential sources of phenotype, ancestry or other variability that could introduce between-study heterogeneity into a meta-analysis. Heterogeneity tests based on individual genetic variants (e.g. Q, I2) are not suited to identifying locus-specific from more systematic multi-locus or genome-wide patterns of heterogeneity. We have developed and evaluated an aggregate heterogeneity M statistic that combines between-study heterogeneity information across multiple genetic variants, to reveal systematic patterns of heterogeneity that elude conventional single variant analysis. Application to a GWAS meta-analysis of coronary disease with 48 contributing studies uncovered substantial systematic between-study heterogeneity, which could be partly explained by age-of-disease onset, family-history of disease and ancestry. Future meta-analyses of diseases and traits with multiple known genetic associations can use this approach to identify outlier studies and thereby optimize power to detect novel genetic associations.
| Meta-analysis of genome-wide association studies (GWAS) is a valuable tool for the discovery of genes that protect or predispose individuals to common complex diseases. It can though be hampered by excessive heterogeneity among its participating studies. To date, the impact of heterogeneity is assessed locally on an individual SNP basis using Q, I2 and τ2 statistics. Here, we present a new heterogeneity statistic, M that assesses genomic (multi-SNP) patterns of heterogeneity in GWAS meta-analysis with enhanced power compared to conventional methods. When applied to a recent GWAS meta-analysis of coronary artery disease, the new statistic revealed substantial patterns of systematic heterogeneity, much of which was attributed to differences in ancestry, age-of—disease onset and family history of disease. The new method can dissect genomic heterogeneity patterns to flag underperforming studies that could comprise the power of the meta-analysis as well as identify influential studies with advantageous design features to inform future meta-analyses of multifactorial disease.
| The common disease—common variant (CD-CV) hypothesis has been confirmed by the discovery of thousands of robustly associated loci for a wide variety of complex diseases and quantitative inherited traits [1]. The genetic effects conferred by common susceptibility loci tend to be small (per-allele disease odds ratios < 1.2 or trait variance < 0.2%) [2] with the consequence that they are frequently only reliably detected in association studies based on upwards of tens of thousands of individuals. Such large sample sizes require considerable resources to complete the necessary participant recruitment, phenotyping and genotyping, resources that are unlikely to be available to individual research groups.
In response, collaborative networks of researchers have formed consortia in order to assemble large collections of genome-wide association data [3]. Participating studies that were independently commissioned are likely to include specific and varied design features, for instance the precise specification of the phenotype or ascertainment criterion, environmental risk factor profiles or genetic ancestry. These sources of variation could influence the meta-analysis and introduce genetic heterogeneity of effect sizes between participating studies, which would reduce power to detect an overall genetic association. Heterogeneity analysis is currently performed on a variant-by-variant basis, which is potentially sensitive to locus-specific effects, for example specific gene-environment interactions that affect a minority of contributing studies. Furthermore, as the true effect sizes of genetic associations tend to be small with relatively large variances at the individual study level, single variants contain modest information on systematic between-study heterogeneity. Together, these features might mask outlier studies in a meta-analysis that show systematic patterns of heterogeneity due to design features that affect the majority of the associated variants. For example, many diseases have a variety of clinical presentations that could affect the case-mix under alternative recruitment frameworks. In multi-ethnic stroke meta-analysis, the distribution of ischaemic and haemorrhagic cases might differ among populations [4]. Furthermore, sub-phenotypes of disease might have larger or smaller genetic components. For example, although the majority of coronary artery disease (CAD) associated loci showed similar effect sizes in analyses based on the subset of cases with myocardial infarction alone versus a broader CAD phenotype (coronary stenosis >50%, acute coronary syndrome and chronic stable angina), discrepant effect sizes were evident at the HDAC9 and ABO loci [3]. Moreover, sampling patients with younger or older age-of-onset of disease or with or without a family-history of disease could affect genetic risk profiles according to the multifactorial liability threshold model [5].
We have therefore developed an analytic approach to search for systematic between-study heterogeneity patterns in genetic association meta-analysis projects. Our approach builds upon the established random-effects meta-analysis method [6], to combine information from multiple genetic variants into an integrated heterogeneity statistic. We first assess the analytic power of the new method to compare its performance with a conventional method to detect heterogeneity and then confirm the size and further explore the power of the new method in computer simulation exercises. Finally, we apply the method to a recent GWAS meta-analysis of CAD [3].
To empirically assess the theoretical distributions of M, SPRE statistics for 2, 10, 25 or 50 variants were randomly sampled from a Φ(0, 1) distribution in 10,000 replicates to approximate the null hypothesis of no systematic heterogeneity. The empirical and theoretical distributions of M match very closely irrespective of the number of variants (S1 Fig and S1 Table).
The analytic power of M to detect heterogeneity was compared with Cochran’s Q statistic [7, 8], a method that is routinely used to detect heterogeneity in meta-analyses and also underpins the I2 inconsistency index [9]. Multiple testing of V variants (for Q) and S studies (for M) was allowed for by applying Bonferroni’s adjustment to ensure that the family-wise error rates (FWER) for each method were equally controlled. Fig 1 shows the comparative power for 10, 25 and 50 variants in 10, 15 and 30 studies; the effect sizes for the S-1 “non-outlier” studies were held constant (loge(odds ratio) = 0.182 i.e. odds ratio = 1.2) to model homogeneous effects. The effect sizes for the variants in the outlier study were the product of the “non-outlier” effect size (i.e. loge(odds ratio) = 0.182) and a parameter (fold-change) to model a continuous series of systematic heterogeneity patterns. All studies were equally weighted (standard error of loge(odds ratio) = 0.1). It is clear that under all scenarios examined (Fig 1), that M had greater power than Q to detect systematic heterogeneity patterns. The power of M to detect systematic heterogeneity increased as the fold-change parameter differed from 1 as well as with larger numbers of variants but was slightly attenuated as the number of studies (and multiple testing burden) increased.
We then used Monte-Carlo computer simulations to empirically assess the type 1 and 2 error rates in a more complex series of “real world” meta-analysis scenarios for differing numbers of variants and studies. Variants were modelled to confer disease risks of varying magnitudes (S2 Table); the effect size distribution across the variants was inspired by an overview of GWAS findings [10], which documented the well-established pattern of a progressively larger number of variants with smaller effect sizes. Random variation in effect sizes for the variants in different studies was included by sampling the effect sizes (i.e. β coefficients scaled as loge(odds ratio)) from a Φ(β, σ = 0.10) distribution; this induces a background random heterogeneity pattern that affects all studies upon which we attempt to detect an overlying systematic heterogeneity pattern that only affects a single outlier study. Studies were assigned equal weights in the meta-analysis by fixing the standard errors of the simulated effect sizes based on the median value of standard errors for variants in a recent GWAS meta-analysis [3] (i.e. SE(β) = 0.08). Thus each variant was modelled with a background heterogeneity inconsistency index [11] I2 = 100 × 0.102 / (0.102 + 0.082) = 60.5%. Table 1 presents empirical type 1 error rates derived from 1,000 replicates to compare with Bonferroni corrected asymptotic p-values < 0.05. The type 1 errors for the M statistics were mostly conservatively controlled in these simulation scenarios.
Simulations were then performed to further assess the power of the M statistic to detect outlier studies included in a meta-analysis on a background of random heterogeneity. Table 1 shows the results from simulations where a single outlier study was included in the meta-analysis that showed a random pattern of association (i.e. the β coefficients for the V variants in the outlier study were sampled from a Φ(β = 0, σ = 0.10) distribution i.e. fold-change = zero). The power of M to identify the “null” outlier study increased with the number of variants but there was little impact on power varying the number of studies in the meta-analysis. We then examined scenarios where an outlier study in a meta-analysis was selected to show systematically stronger effects than the other participating studies (Fig 2). Again the power of M statistic to detect the outlier study increased with the number of variants included in the meta-analysis. Varying the number of studies in the meta-analysis had relatively little impact on the power to detect systematic outliers. Similarly, the power of M statistic to diagnose an outlier study showing systematically weaker effects than other participating studies increased with the number of variants interrogated in the meta-analysis. We also studied the impact of the background level of heterogeneity on power; this showed that it is easier to identify outlier studies with the M statistic if the average level of heterogeneity is low (S2 Fig).
The CARDIoGRAMplusC4D consortium has recently reported a GWAS meta-analysis of 60,801 CAD cases and 123,504 controls assembled from 48 studies [3]. Participants had been recruited from several ancestry groups (African American, Hispanic American, East Asian, South Asian, Middle Eastern and European). The CAD cases included patients with clinical diagnoses of myocardial infarction with or without ST-elevation, other acute coronary syndromes or chronic stable angina, as well as patients who had undergone a revascularization procedure or had angiographic evidence of stenosis (>50%) in at least 1 coronary vessel. The majority of the studies recruited CAD cases retrospectively (i.e. prevalent cases), the other prospective studies included a mixture of incident and prevalent disease. The controls included population samples who were unscreened for CAD (e.g. the UK 1958 Birth Cohort and National Blood Service controls genotyped as part of the Welcome Trust Case Control Consortium [12]) in addition to samples from volunteers with no personal history of coronary disease or individuals who had undergone coronary angiography but had no radiological evidence of vessel stenosis. Various GWAS SNP arrays had been genotyped by the studies so genotype imputation to the 1000 genomes phase 1, version 3 haplotype training set was used to facilitate the meta-analysis by maximizing the available mapping information.
In an additive-effects-only association analysis, 46 discrete CAD loci surpassed the conventional genome-wide significance threshold (i.e. P < 5 × 10−8). Variants within the 46 loci were mostly well imputed with 82% of the variants having an imputation quality score > 0.9. A lead variant (i.e. the variant with the smallest p-value) for each of these loci was selected for aggregate heterogeneity analysis, 35 of these variants showed some degree of between-study effect size heterogeneity (i.e. I2 > 0) (S3 Table). The 46 lead variants were in linkage equilibrium with each other.
Inspection of the M statistics for the 48 studies suggested that some studies showed systematic differences from the average genetic effect (Fig 3 and S4 Table). Notably, studies 9, 38 and 48 showed significantly stronger effects than average (Bonferroni corrected p-values < 0.05) while studies 10, 19, 24 and 28 showed significantly weaker effects (Bonferroni corrected p-values < 0.05). An inverse-variance weighted meta-analysis of the M statistics revealed substantial variability in the average effect across studies (I2 = 85.9%) (Fig 3). In an attempt to resolve underlying design factors that contributed to this systematic between-study heterogeneity pattern, we applied a random-effects meta-regression method [13] to the M statistics. We examined three potential sources of systematic heterogeneity that might have influenced the CARDIoGRAMplusC4D meta-analysis 1) ancestry, 2) family-history and 3) age-of-onset of disease (S5 Table). The participating studies had been independently commissioned and designed with overlapping disease case ascertainment criteria; accordingly we assigned the studies into earlier-onset (≤ 55 years) and later-onset of disease groups and flagged studies that ascertained cases with a positive family-history of disease (S5 Table).
A meta-regression of the M statistics with ancestry coded into 6 groups (African and Hispanic American, South and East Asian, Middle Eastern and European) suggested that some of the variability in average effect size could be explained by ancestry (F5, 42 = 2.52, P = 0.044) (Fig 4A). The 3 East Asian studies collectively appear to show stronger than average effects when compared with all other ancestry groups (F1, 46 = 4.75, P = 0.034). There was no evidence that the average effects for the 38 European studies (F1, 46 = 1.24, P = 0.271) or the 4 South Asian studies (F1, 46 = 2.99, P = 0.090) were systematically different.
Meta-regressions of the M statistics suggested that studies that included early-onset cases of disease (F1, 46 = 20.65, P = 0.00004) or included a family-history of CAD in the ascertainment scheme (F1, 46 = 29.49, P = 2.0 × 10−6) showed systematically stronger than average effects (Fig 4B–4D). Finally, a multiple meta-regression analysis of East Asian ancestry, early-onset and family-history of disease showed that these factors jointly explained a significant proportion of the systematic between-study variation of average effect size (F3, 44 = 13.91, P = 1.6 × 10−6; adjusted R2 = 53.2%) (Table 2). Additional factors examined as potential contributors to the systematic between-study differences observed included: imputation quality, genotype call rate, Hardy Weinberg equilibrium thresholds, percentage of myocardial infarction cases and case-control ratio. Their contribution to between-study variation of average effect size was negligible.
The CARDIoGRAMplusC4D consortium studied an extended list of independently associated variants that define additional discrete loci based upon false discovery rate (FDR) criteria [3](S6 Table). These variants incremented the heritability explained over that conferred by GWAS-significant loci and might offer greater insights into heterogeneity patterns in these data. We therefore repeated the M statistic analysis with 214 variants (P < 0.00005, FDR < 5%), which confirmed the presence of systematic heterogeneity patterns in the 1000 genomes meta-analysis (S3 Fig) as well as flagging individual outlier studies (S4 Fig). Four studies, that showed insignificant outlier patterns with 46 GWAS-significant variants showed significant evidence in this analysis of FDR variants (S7 Table) and three studies that were outliers in the GWAS 46 are now insignificant. A meta-regression confirmed that East Asian ancestry, early-onset and family-history showed systematically stronger than average effects (F3, 44 = 9.47, P = 0.0001; adjusted R2 = 44.8%) with family-history as the most important predictor of systematic heterogeneity in this dataset (S8 Table).
To compare our M analysis with a conventional single-variant strategy, we re-examined the set of GWAS-significant variants in a series of meta-regressions of three joint predictors, East Asian ancestry, early-onset and family-history. After correction for multiple testing of 46 variants, one variant (rs2891168) detected evidence of stronger associations with early-onset and family-history (F3, 44 = 6.71, P = 0.0008; adjusted R2 = 44.3%) and another variant (rs6689306) showed stronger associations with East Asian ancestry (F3, 44 = 7.69, P = 0.0003; adjusted R2 = 71.5%) (S5 and S6 Figs, S9 Table).
We present here a novel statistical approach that integrates information across multiple variants to explore background patterns of systematic between-study heterogeneity in genetic association meta-analyses. Although we have focused on examples drawn from case-control analysis where genetic association statistics have been computed by logistic regression, the method is equally applicable to other normally distributed regression statistics e.g. linear regression analysis of quantitative genetic associations. We hypothesised that design features such as ascertainment criteria for disease cases or genetic ancestry might induce genetic heterogeneity in a meta-analysis. If these design features systematically reduce the average effect size in some of the studies participating in the meta-analysis, then the overall power to detect genetic signals will be reduced. This is an important consideration, since genetic effects for CD-CV are typically small in magnitude requiring very large sample sizes for reliable detection; there is strong pressure to undertake increasingly large meta-analyses. As meta-analysis consortia expand to attain larger sample sizes, the risk that they will become increasingly diverse in terms of underlying design features must surely increase.
Analytic and Monte Carlo simulations demonstrate the potential of the proposed M statistic to detect systematic patterns of between-study heterogeneity. These calculations were based on a specified uniform level of heterogeneity for each variant and showed that the conventional approach to detecting heterogeneity (e.g. Cochran’s Q statistic) is relatively underpowered to detect systematic patterns. To maximize the power of detecting systematic heterogeneity patterns, we recommend analysing as many independently (i.e. in linkage equilibrium) and strongly associated variants as possible. In the future it would be interesting to extend the M approach including variants in linkage disequilibrium (LD) as this development might further enhance its power. It is anticipated that lead variants will show varying levels of heterogeneity, indeed several are likely to show little or no statistical evidence of heterogeneity (i.e. I2 < 25%). Such variants do though include some information relevant to detecting systematic weaker or stronger effects than average so we recommend that all firmly associated lead variants are included in the calculation of M statistics. Our simulations also assumed equal weightings for each contributing study, we anticipate that outlier studies with larger sample sizes (and thus weightings) will be prominent and outliers with small weightings are likely to be obscure. We also found that the background level of heterogeneity influences the power to detect outlier studies, panels of strongly associated variants that individually show moderate levels of heterogeneity (25% < I2 < 50%) are well suited to this approach.
We tested our new methods on data assembled for the CARDIoGRAMplusC4D GWAS meta-analysis of CAD risk [3]. Although there was marked heterogeneity of effect sizes across the participating studies (Fig 3), all studies showed positive associations with coronary disease risk (Fig 4) and thus made useful contributions to the overall discovery GWAS objective. Meta-regression of the M statistics revealed patterns of systematic heterogeneity that were linked to specific design features, East Asian ancestry, age-of-onset of disease and family-history. The latter two features are predicted by the multifactorial threshold model [5] to induce genetic enrichment [14]. Of note, 50 years ago the early-onset of coronary disease was recognised as a potent risk factor increasing sibling recurrence risks six-fold [15]. Although the magnitudes of the enrichment of average genetic effect size were quite modest (14% for East Asian ancestry, 15% for family-history, 11% for early-onset), we estimate that this could reduce the required sample size of cases and controls to detect an associated locus by up to 50%. Population genetic diversity, differences in the underlying rates of CAD and the relative contribution of individual risk factors, as well as lower use of preventive therapies in East Asia versus Europe (and other regions) may contribute to the enriched genetic signal observed in East Asian studies [16, 17]. A follow-up meta-regression analysis of individual variants confirmed the role of ancestry, age—of-onset and family-history as significant predictors of systematic heterogeneity. Meta-regression of multiple potential explanatory factors inevitably carries a multiple statistical testing burden, and our present results should be interpreted as an exploration of the substantial systematic heterogeneity patterning evident in Fig 4. The M statistic approach is advantaged over conventional single-variant methods in that information across multiple variants is aggregated reducing the dimensionality of the multiple comparison problem. Finally, we were unable to detect any systematic heterogeneity patterning attributable to the proportion of CAD cases suffering a myocardial infarction confirming the findings of the CARDIoGRAMplusC4D consortium [3].
There are several potential sources of heterogeneity that might affect genetic association meta-analysis studies. Controls for a common disease might be drawn from unscreened population samples in some studies or screened for the disease and filtered in other studies, this is predicted to dilute genetic signals in studies using population controls [18]. Environmental risk factor profiles might vary from study to study so disease cases sampled from a relatively low risk population would tend to be genetically enriched. Varying levels of LD can also induce heterogeneity [19], a situation that is particularly important for meta-analyses involving multiple ancestry groups where the lead variant is a tagging rather than the causal variant. For example, African ancestry populations typically show more haplotype diversity and lower levels of LD across the genome than European or in turn East Asian populations [20]. Thus in a multi-ethnic meta-analysis, signals detected by tagging SNPs could show systematic weaker (in low LD populations) or stronger (in high LD populations) effects that could be detected by the M statistic approach.
Given the momentum of the GWAS approach to identify more and more loci with improved genotype imputation training sets [21], it is inevitable that increasingly large and potentially diverse meta-analysis projects will be conceived. For diseases and traits with multiple known genetic signals, there is now an opportunity to assess the respective contributions of participating studies in newly commissioned meta-analyses. Outlier studies flagged with discrepant M statistics, particularly those with weaker than average effects, can be reviewed as part of the routine quality control of GWAS meta-analysis in case there are design or analytic issues that need attention to maximize power. For design issues that might be difficult to resolve, it would be useful to assess the power of performing meta-analysis in the presence and absence of the studies with discrepant M statistics. Following the final meta-analysis, meta-regression of M statistics including variants tagging previously known as well as newly discovered loci can be used to explore potential design features that might show systematic aggregate effects that are obscured in heterogeneity analyses of individual loci, and influence future study design.
Random-effects meta-analysis is a statistical procedure originally devised by epidemiologists to integrate summary information from multiple independent yet related interventional studies to estimate two parameters, Θ, the average treatment effect across the contributing studies and τ2, the extent of inter-study variability (or heterogeneity) in the treatment effects [22]. The effects evident in each study are assumed to be have been sampled from a normal distribution with two additive variance components, random within-study error σ2 and between-study variation (i.e. heterogeneity) τ2, so that ys, the measured effect in the sth study, can be modeled by: ys = Θ + εs + u where εs ~ Φ(0, σ2s), u ~ Φ(0, τ2) and Φ denotes the cumulative probability distribution function of a normal random variable.
The first step in the analysis is to estimate the magnitude of τ2, which can be undertaken by several algorithms [22]. This is followed by an inverse-variance weighted (i.e. 1/(τ^2+σs2)), least squares estimation of the average treatment effect (Θ) (which ignores the study-specific random effects) and its associated standard error (E, the “standard error of the prediction”).
Standardized predicted random effects (SPRE) can then be calculated for each of the studies as SPRE=(ys−θ)/τ^2+σs2−E2; these are precision-weighted, standard normally distributed statistics (i.e. SPRE ~ N(0, 1)) that summarize the extent and the direction that individual studies differ from the average treatment effect. If there is no evidence of heterogeneity of effects (i.e. τ2 = 0), then the SPREs are identical to standardized predicted fixed effects derived from a fixed-effects meta-analysis. A normal probability plot of the SPRE statistics provides a convenient visual way to detect outlier studies that might be unduly influencing the estimate of the average treatment effect that complements inspection of a Forest plot.
Consider now a genetic association meta-analysis project comprising S studies with summary-level results for V genetic variants. Genetic effect-sizes (and their standard errors) have been estimated in each study by regression modelling to substitute for the treatment effects described above. Assume that the variants selected for heterogeneity analysis are truly associated with the disease or quantitative trait and are in linkage equilibrium (i.e. uncorrelated) with each other. Informative variants could include 1) published variants that have previously shown strong evidence of association or 2) the lead variants at GWAS-significant loci in a post-hoc heterogeneity analysis. The genetic effects need to be synchronized so that the average Θ estimates for each variant are positive (i.e. all average effects are “in the same direction” consistent with higher disease risks or levels of a quantitative trait); this can be achieved by judicious “flipping” of the regression coefficients submitted by participating studies.
For each of V variants, estimate τ2, Θ and E using the random-effects procedure detailed above and calculate and store SPRE statistics for each of S studies in a regular array SPREsv (S1 Methods).
Subsequently, a “mean” aggregate statistic can be calculated that summarizes between-study heterogeneity across multiple genetic variants:
Ms= 1V∑v=1VSPREsv.
Under the assumption that Ms is a linear combination of V mutually independent, standard normal random variables, then Ms will be normally distributed with expectation (i.e. mean) 0 and variance 1/V (S2 Methods). Positive or negative values of Ms indicate that the study shows systematically larger or smaller genetic effects than the average effect, statistically significant deviations are found where |Ms| exceeds an appropriate threshold; we corrected for multiple testing of S studies by applying the Bonferroni procedure to control the family-wise error rate (FWER) < 0.05. We developed a Stata script (getmstatistic) based on the metareg programme [23] to calculate Ms statistics using τ2 estimates derived from the restricted maximum log-likelihood (REML) method. Additionally, an R package (Rgetmstatistic) for getmstatistic has been developed.
To support the use of this newly proposed statistic, we examined the impact of several systematic heterogeneity scenarios by means of analytic and Monte-Carlo simulation-based power analyses. We first compared our new method with Cochran’s Q statistic, a widely used and computationally simple method to screen for between-study heterogeneity [7, 8]. Q statistics approximate a chi-squared distribution in large samples [24], for each scenario non-centrality parameters were equated with calculated Q statistics (i.e. treating Q as a log likelihood ratio statistic [25]; [26]). The non-centrality parameter was then used in standard chi-squared power calculations ([26]), with an allowance for multiple testing of V variants by applying Bonferroni’s correction to control the family-wide error rate (FWER) to 5%. Denote the power to detect heterogeneity in a meta-analysis of the vth variant by ωv, then the overall power to detect at least one heterogeneous variant is
ω=1−∏1V1−ωv
To calculate the analytic power of M, it is convenient to introduce a Wald statistic (M2), the squared-standardized M statistic i.e. M2=(MSEM)2 where SEM=(1V)12, which is approximately chi-squared distributed on 1 degree of freedom. M2 can then be substitute for the non-centrality parameter in standard chi-squared power calculations [26] allowing for multiple testing of S studies by applying Bonferroni’s correction to control the family-wide error rate (FWER) to 5%. Denote the power (ω) to detect heterogeneity in a meta-analysis for the sth study by ωs, then the overall power to detect at least one heterogeneous variant is
ω=1−∏1S1−ωs
The above analytic power calculations were performed using scripts and in-built procedures in Stata 10.1.
We also carried out Monte-Carlo simulations for scenarios where a systematic heterogeneity pattern is superimposed on a background random heterogeneity pattern, this allows for the possibility that real world heterogeneity patterns have multiple sources and complexity. These simulations allowed the comparison of the distributions of empirical with asymptotic p-values, with empirical p-values calculated using the (r+1)/(n+1) estimator [27] where r represents the rank of the simulated statistic and n the total number of replicates in the simulation exercise.
To explore the impact of design features on the magnitude of M that vary between individual studies participating in a meta-analysis, we apply a random-effects meta-regression procedure (metareg) in Stata 10.1 to regress towards the average deviation of the observed effects of studies. This analysis is based upon study-specific M statistics to summarize the studies’ overall deviation from the average effect with precision weighting (i.e. 1/SEMs for the sth study—see S2 Methods) to allow for differing sample sizes in different studies.
The studies contributing to the CARDIoGRAMplusC4D study were approved by the ethics committees of the respective medical faculties, and informed consent was obtained from all participants. Summary genetic association data were anonymously meta-analysed and reported here.
Software to calculate M statistics is available at the following url: getmstatistic, https://magosil86.github.io/getmstatistic
Supplemental data includes the membership of the CARDIoGRAMplusC4D Consortium, six figures and nine tables.
|
10.1371/journal.pntd.0002154 | High Content Analysis of Primary Macrophages Hosting Proliferating Leishmania Amastigotes: Application to Anti-leishmanial Drug Discovery | Human leishmaniases are parasitic diseases causing severe morbidity and mortality. No vaccine is available and numerous factors limit the use of current therapies. There is thus an urgent need for innovative initiatives to identify new chemotypes displaying selective activity against intracellular Leishmania amastigotes that develop and proliferate inside macrophages, thereby causing the pathology of leishmaniasis.
We have developed a biologically sound High Content Analysis assay, based on the use of homogeneous populations of primary mouse macrophages hosting Leishmania amazonensis amastigotes. In contrast to classical promastigote-based screens, our assay more closely mimics the environment where intracellular amastigotes are growing within acidic parasitophorous vacuoles of their host cells. This multi-parametric assay provides quantitative data that accurately monitors the parasitic load of amastigotes-hosting macrophage cultures for the discovery of leishmanicidal compounds, but also their potential toxic effect on host macrophages. We validated our approach by using a small set of compounds of leishmanicidal drugs and recently published chemical entities. Based on their intramacrophagic leishmanicidal activity and their toxicity against host cells, compounds were classified as irrelevant or relevant for entering the next step in the drug discovery pipeline.
Our assay represents a new screening platform that overcomes several limitations in anti-leishmanial drug discovery. First, the ability to detect toxicity on primary macrophages allows for discovery of compounds able to cross the membranes of macrophage, vacuole and amastigote, thereby accelerating the hit to lead development process for compounds selectively targeting intracellular parasites. Second, our assay allows discovery of anti-leishmanials that interfere with biological functions of the macrophage required for parasite development and growth, such as organelle trafficking/acidification or production of microbicidal effectors. These data thus validate a novel phenotypic screening assay using virulent Leishmania amastigotes growing inside primary macrophage to identify new chemical entities with bona fide drug potential.
| Leishmaniases are neglected diseases caused by protozoan parasites that belong to the genus Leishmania. No vaccine exists against any form of leishmaniasis and most of the existing anti-leishmanial drugs have serious side effects. Current strategies for discovering new leishmanicidal molecules are largely using inappropriate types of host cells or employing the irrelevant insect-specific parasite stage. We presented here an innovative and biologically relevant drug screening assay based on the use of the mammalian macrophage host cells and Leishmania pathogenic amastigotes. Our visual fluorescence assay has been established with the objective to screen diverse small-molecules including chemicals and natural compounds that selectively target intra-macrophagic amastigotes without displaying toxicity for the macrophage host cells. The validation of a miniaturized assay, relying on automated handling of biological materials and fluorescent imaging probes, image acquisition, data storage and analysis is presented. Based on robust statistical methods and quality control metrics, the data analysis pipeline allows for the classification of compounds based on their effect on parasite and macrophage survival observed after three days of treatment.
| Human leishmaniases are diseases that are endemic throughout tropical and subtropical areas causing severe morbidity and mortality with an estimated worldwide incidence of 1,500,000 newly reported clinical cases per year. More than 12 million people are currently infected and 350 million people are at risk [1].
Causal agents of human leishmaniases are Euglenozoan digenetic parasites belonging to the Trypanosomatidae family and the genus Leishmania. They are cycling between i) blood-feeding phlebotomine insects where they develop as extracellular flagellated promastigotes, and ii) a range of mammals, including rodents, canids and human where they develop as obligatory intracellular amastigotes. Depending on both the Leishmania species and the mammal host genetic and immune status, four main clinical forms of leishmaniasis can be distinguished, including i) the visceral form, which causes long-term fever, weight loss, hepatosplenomegaly and pancytopenia, and is usually fatal if left untreated, ii) the cutaneous form characterized by single or multiple lesions, which generally self-cures over months, iii) the muco-cutaneous form affecting mainly nasopharyngeal mucosa characterized by extensive tissue destruction causing severe facial disfiguration and respiratory disturbances, and iv) a diffuse form in which non-ulcerating lesions are spread over large skin areas. Leishmaniases result in a strong social stigma and marginalization for infected individuals and has an important negative impact on welfare and productivity of people from developing countries [2]–[4].
The current treatment of leishmaniasis essentially relies on chemotherapy as, to date, neither preventive nor therapeutic vaccines are available. Pentavalent antimonials have been used for more than 70 years and are still the recommended first line of treatment. However, these drugs not only require a long course of parenteral treatment with repeated injections and careful monitoring in health centers, but also display variable efficacy and toxic side effects [1], [5]–[7]. Alternative therapeutic approaches based on the use of Amphotericin B (AmphoB), an antifungal polyene antibiotic, and its lipid-carrier formulations have been successfully applied when first line drugs were no longer effective [8], [9]. Nevertheless, the very high costs of these drugs limit their widespread use. In recent years, new molecules such as the anti-neoplastic agent Miltefosine, new treatments such as drug combinations, or new administration routes like topical formulations and oral administration, have been introduced in anti-leishmanial chemotherapy. However, in view of the recent spreading of Leishmania showing resistance to the limited number of existing drugs in various parts of the world, there is an urgent need to develop new, safe, fast acting, and affordable treatments.
While several new chemical scaffolds have been identified recently, they were issued from screening campaigns that were primarily based on the use of extracellular promastigotes [10], [11], i.e. the developmental stage that differentiates within the mid gut of sand fly hosts. It is noteworthy that promastigotes differ significantly from amastigotes with respect to morphology, the composition of their surface glycocalyx, and metabolism. These biological differences reflect distinct developmental programs that adapt Leishmania for extra- and intracellular survival in the phlebotomine mid gut and the macrophage parasitophorous vacuole (PV), respectively [12]–[16]. Conceivably, this stage specific biology has important consequences on how the parasite responds to chemicals [17]–[19]. When comparing the leishmanicidal activity of antimony in infected animals versus cultured parasites, the antimony susceptibility determined in vivo correlated better with the in vitro assays performed with intracellular amastigotes than with extracellular promastigotes [20]. Moreover, host features such as the permeability of plasma- and PV- membranes or the presence of molecules able to directly interfere with or metabolize chemicals are key parameters to consider for discovering compounds with selective activity against the intracellular amastigote stage. Leishmania survival within macrophages depends on the capacity of amastigotes to evade or to resist the innate host cytotoxic activities. It is possible that rescue of these activities will lead to efficient intracellular amastigote destruction. Thus, an assay based on the phenotype classification of primary macrophages hosting Leishmania amastigotes may allow to discover compounds that confer to infected macrophage full leishmanicidal activities [18], [21]–[24]. With the aim to establish a biologically relevant screening system that accounts for all of these considerations, we designed and validated a miniaturized High Content Analysis assay, relying on primary mouse macrophages hosting virulent L. amazonensis amastigotes. This approach overcomes limitations associated with promastigote-based screens [19], [25]. It allows the rapid selection of compounds that are able to interfere with Leishmania amastigote growth and survival within primary macrophages either directly, or indirectly by modifying macrophage organelle trafficking or acidification required for intracellular parasite growth. Based on robust statistical methods, quality control metrics, hit identification classification and validation, we developed a powerful data analysis pipeline that provides for each tested compound metrics on amastigote load and their toxic effect on host macrophages at the single cell level and for the entire sample population analyzed.
Female Swiss nu/nu and BALB/c mice, between 8- and 12-week of age, were obtained from Charles River. Bone marrow cell suspensions recovered from tibias and femurs of BALB/c mice were suspended in DMEM medium (Gibco, life technologies) containing 4 g/L glucose, 1 mM pyruvate and 3.97 mM L-Alanyl-L-Glutamine, 10% heat-inactivated fetal calf serum (FCS, Dominique Dutscher SAS), streptomycin (50 µg/mL) and penicillin (50 IU/mL) (Biochrom AG, IBS International) (culture medium) and with 50 ng/mL recombinant mouse CSF-1 (rmCSF-1) (ImmunoTools). Cells were distributed in bacteriologic plastic flasks (Corning Life Science, 7×105 cells/ml) and were incubated at 37°C in a 7.5% CO2 air atmosphere for 6 days.
Six days old bone marrow-derived, loosely adherent macrophages were washed with Dulbecco's phosphate buffered solution (PBS) and detached by gentle flushing (25 min at 37°C) with pre-warmed 1% EDTA in PBS without Ca2+ and Mg2+ (Biochrom AG). Recovered macrophages were suspended in either culture medium for HCS assay or cold PBS with 2% FCS and 0.05% sodium azide (PBS-FCS-Az) for FACS quality controls. Macrophages for FACS analysis were transferred to round-bottomed 96-well plates (Corning Costar) at a concentration of 3×105 cells/well. All subsequent steps were performed on ice and with ice-cold reagents. Cells were centrifuged (300 g) for 5 minutes and then incubated in PBS-FCS-Az supplemented with 10% donkey serum for 20 minutes. After centrifugation, cells were incubated for 30 minutes in PBS-FCS-Az containing a combination of fluorescent reporter-conjugated antibodies. Flow cytometry results were acquired on a Gallios flow cytometer (Beckman Coulter) and data analyzed with the Kaluza software package (Beckman Coulter).
The anti-mouse mAbs were purchased from Pharmingen for FITC-labeled 2G9 anti I-Ad/I-Ed clone, or eBioscience for the followings clones: e450-conjugated N418 anti-CD11c (p150/90), APC-conjugated M1/70 anti-CD11b/CR3 α-chain, PE-conjugated 16-10A1 anti-CD80/B7-1, APC-conjugated BM8 anti-F4/80, and PE-conjugated AFS98 anti-CD115.
L. amazonensis strain LV79 (MPRO/BR/1972/M1841) was genetically modified by chromosomal integration of the fluorescent DsRed2 molecule [26] and propagated in Swiss nu/nu mice by subcutaneous injection of 106 amastigotes into hind footpad. Six to eight weeks after amastigote inoculation lesions were excised and amastigotes purified by a modified version of the method originally described by M. Rabinovitch and colleagues [27]. Briefly, lesions were minced in PBS supplemented with streptomycin (100 µg/mL) and penicillin (100 IU/mL), and disrupted by hand in a glass homogenizer. Tissue debris were removed by 2 rounds of centrifugation at 30 g for 10 mn at 4°C. Amastigotes present in the supernatant were washed 2 times by centrifugation at 1500 g for 10 mn at 4°C before distribution in macrophage cultures. A high number of live amastigotes expressing homogenous levels of DsRed2 were recovered as determined by FACS analysis (data not shown).
Reference compounds, Leucine Methyl Ester (Leu-OMe) [27], AmphoB and cycloheximide, were solubilized in DMSO (Sigma-Aldrich). Based on literature data, we selected 60 compounds with established or potential leishmanicidal, anti-fungal or anti-microbial and cytotoxic activities to validate our experimental and data analysis pipelines. Details about their origins, working concentrations and known activity are provided in Table S1. We defined the following control conditions: C− 1% DMSO-, C+, 0.5 µM AmphoB- and C† 180 µM cycloheximide. All compounds were assayed at 10 µM or as stated in Table S1. The detailed plate maps used in this study with the position of controls are depicted in Tables S2 and S3 (see table legend for details).
Six days-old bone marrow-derived adherent macrophages were recovered as described above and deposited in culture-treated flat-optically clear bottom black 384-well plates (CellCarrier plate, PerkinElmer) at a density of 1.5×104 cells in 70 µl of medium supplemented with 12 ng/ml of rmCSF-1 per well, resulting in a 80% confluence monolayer without formation of cellular aggregates. Five hours later, purified DsRed2-expressing amastigotes were added to the macrophages at a multiplicity of 3 parasites per host cell (MOI = 3) (30 µl/well). Macrophage cultures were further incubated at 34°C, which is the permissive temperature for the surviving and multiplication of LV79 amastigotes [28], [29]. After an overnight incubation period, more than 85% of macrophages harbored intracellular parasites that were already multiplying in growing PVs [30]. At this time, compounds (Tables S2, S3) were added to macrophages (1 µl/well) resulting in a final concentration of 1% DMSO in each well. The cultures were then maintained 3 days at 34°C until processing for image acquisition.
One hour before image acquisition, the cells were incubated with vital cell-permeant dyes Hoechst 33342 (12 µM) and LysoTracker DND-26 (1 µM) (life technologies). Optimization of operating parameters included an automated dispensing of biological material (macrophages and amastigotes) and chemicals (vehicle, compounds) using the Te-MO 96-channel pipetting head of a TECAN Freedom EVOware platform located under laminar flow in a BSL2 facility. The homogeneity and the reproducibility of all pipetting procedures were assessed by quantitative image analysis during assay development (data not shown).
After 60 minutes of contact with fluorescent reporters, three channel images were acquired in a fully automated and unbiased manner using a spinning disk confocal microscope (OPERA QEHS, PerkinElmer Technologies) and a 10× air objective (NA = 0.4) using the following sequential acquisition settings: (i) 561 nm laser line excitation, filter 600/40 for DsRed2 detection, (ii) 488 nm laser line excitation, filter 540/75 for Lysotracker DND-26 detection and (iii) 405 nm laser line excitation, filter 450/50 for Hoechst 33342 detection. Fifteen images per channel, covering the entire surface of each well, were collected for reliable statistical analysis taking into account potential cell-distribution and spatial compound effect biases.
The images were transferred to the Columbus Conductor™ Database (Perkin Elmer Technologies) for storage and further analysis. The image analysis was performed by batches in Columbus using custom designed image analysis scripts developed beforehand with the Acapella Image analysis software (version 2.5 - Perkin Elmer Technologies). The script was subdivided in three object segmentation subroutines detecting successively and independently the nuclei, the PV and the Am with their respective associated features (number, size, and intensity); the living macrophage population characterisation is based on a combination of host cell nucleus size and intensity features which are key characteristics of the relative fitness of the macrophage population. Finally, all the quantitative data generated were exported in readable file format to be subsequently analyzed in the data analysis workflow described below.
To validate our pipeline, selections of image analysis outputs, including macrophage nuclei, PV and Am counts that best represented typical compound-induced phenotypes, were used. We applied a standardized data analysis workflow to automate and validate the interpretation of the large amount of data generated by image acquisition and analysis; it consists of the following three main classical hands-off steps [31]:
Current screening protocols for the discovery of anti-leishmanial compounds are compromised by both the types of parasites and host cells employed. First, the use of culture-derived promastigotes and axenic amastigotes are not reflecting the biology and environment of parasites inside the macrophage. In addition Pescher and colleagues recently demonstrated that axenic amastigotes were not able to induce acute visceral disease in hamsters compared to tissue-derived amastigotes thus minimizing the potential interest of using host-free parasites [35]. Second, the use of macrophage cell lines as host cells for Leishmania is problematic due to the mandatory use of chemicals to induce terminal macrophage differentiation, chemicals that are known to result in modulation of macrophage sensitivity to compounds, thus compromising the interpretation of screening results [18], [36], [37]. Finally, the combination of both i.e. using host cell lines infected with culture-derived parasites results in i) the presence of large quantity of extracellular proliferating promastigotes and ii) a lower rate of host cells hosting metacyclic promastigotes able to differentiate into cell-cycling amastigotes. To overcome these limitations, we set up a reliable assay based on the use of mouse primary macrophages and highly virulent lesions-derived amastigotes of L. amazonensis, which were genetically modified to stably express a DsRed2 fluorescent reporter [26]. Homogeneous populations of macrophages differentiated from bone marrow progenitors by incubation with Colony Stimulating Factor 1 (CSF-1) [29] (Figure S2) were distributed in 384-well plates, and amastigotes freshly prepared from Swiss nude mouse lesions [38] were then added to the macrophage monolayer. In contrast to promastigote-based protocols, L. amazonensis amastigotes were readily phagocytized by macrophages leading to a high and sustained infection rate. After a few hours, amastigotes were already multiplying, resulting in the development of large PVs, a phenotypic hallmark of a successful intracellular L. amazonensis infection (Figure 1A) [30], [39]. No extracellular amastigotes can be evidenced either by phase contrast or confocal fluorescence microscopy (Figures 1A, 1B). The PV property to accumulate the cell-permeant LysoTracker DND-26, a fluorescent weak base probe, is a signature of the sustained fusion of macrophage late endocytic organelles with amastigotes-containing phagosomes (Figure 1B) [39], [40]. When amastigote growth is blocked or amastigotes are killed in presence of a leishmanicidal agent like AmphoB, PVs are either strongly reduced in size or no longer detected (Figure 1C). The disappearance of PVs was correlated in a previous study to elimination of intracellular amastigotes as shown by fluorescence and differential interference contrast microscopy, and real-time quantitative PCR [29]. The presence/absence of PVs represents thus a powerful digital readout to monitor a leishmanicidal effect, a criterion we already used successfully in a non-automated visual assay for selecting 2-quinoline derivatives with activity against intracellular L. amazonensis amastigotes [29], [41]. Concomitantly, our assay allows monitoring the health status of host macrophages by visualizing their nuclear morphology with the permeant DNA probe Hoechst 33342. In presence of a toxic compound like cycloheximide, dead macrophages were easily differentiated from healthy cells and identified by the loss of nuclear integrity (Figure 1B, 1C, 1D). Based on this biologically relevant and quantifiable infection system we established an automated phenotypic screening pipeline described below.
To minimize any experimentally induced biases, we developed a linear procedure that consists of the sequential addition of mouse bone marrow-derived macrophages, purified tissue-derived amastigotes and chemicals in 384-well optical imaging clear bottom plates, without any washing steps (Figure 2A). This protocol minimizes the potential heterogeneity between wells, thereby avoiding sample perturbation and artifacts over the subsequent incubation period. The procedure is finalized after 3 days of co-culture by the addition of the fluorescent reporters LysoTracker DND-26 and Hoechst 33342 one hour before image acquisition. To validate our procedure, images were acquired in numerous control wells for the 3 different fluorescent reporters corresponding to the counts of DsRed2-tagged amastigotes, LysoTracker-positive PVs and macrophage Hoechst-stained nuclei (see next paragraph for a detailed description of the procedure). In order to avoid biases due to potential heterogeneity of the macrophage monolayer that could arise over time, and to increase the size of the population analyzed, image acquisitions were performed at low magnification using a dry 10× objective for the entire surface of the wells, resulting in acquisition and analysis of all cells for each sample. The images were thereafter segmented using Acapella scripts and the outputs normalized and expressed as percentages of media control (Figure 2B). The presence of 1% DMSO (C-) did not induce toxic effect on macrophages as demonstrated by the similar values obtained for the VI (Figure 2B bottom panel and Figure S3). On the contrary, the presence of a toxic compound for the macrophage, like cycloheximide, was easily evidenced by the dramatic decrease of the VI compared to untreated or DMSO-treated samples (Figure 2B, bottom panel). As expected, in presence of 2 leishmanicidal agents L-Leu-oMe and AmphoB, amastigote and PV counts were significantly reduced compared to negative controls (Figure 2B, top and middle panels) without inducing toxicity on host macrophages (Figure 2B, bottom panel). When performing a comparative analysis between amastigote and PV counts, we observed a good correlation in dose-response experiments using known leishmanicidal agents (Figure 2C and data not shown). Specifically, the IC50 of AmphoB was estimated at 0.11 µM and 0.17 µM for the PV and amastigote output, respectively, which, furthermore, are consistent with published values [10]. Because the PV readout can be efficiently quantified at lower magnification (10× objective), allowing for the analyses of the entire well per sample, we performed the subsequent screen using the presence/absence of PVs as the principal readout for parasite burden generating statistically highly relevant screening data. The box-and-whisker diagrams depicted in Figure 2B for the different variables strengthen and validate the homogeneity and the reproducibility of pipetting procedures, including the chemical distribution routine, used in this assay. The counts of macrophage nuclei, PVs, and amastigotes were subsequently used to establish a data analysis pipeline described in the following.
We applied our experimental approach to a small compound library combining known leishmanicidal drugs, newly published structures and antifungal or cytotoxic agents (Table S1). These compounds were initially tested in quadruplicates per plate in 2 different experiments (plates P1–P2 for Exp1 and P3–P4 for Exp2). In experiments 3 to 7, we used single data points for each compound on each plate (plates P5 to P9 corresponding to Exp3 to Exp7) at a concentration of 10 µM unless stated otherwise (Table S1). The precise design of the assay plates is described in Tables S2 and S3. In P5 to P9, we removed the “media” negative control, which was redundant to the “vehicle” control (1% DMSO) since no toxicity towards the macrophage and amastigote populations could be evidenced (Figure 2B and Figure S3). We also included in these plates cycloheximide-treated cells as positive control for macrophage toxicity (C†), which allowed us to more adequately classify compounds with strong toxic effect on macrophages and only very weak effect on amastigotes (Figures 1D and 2B) as shown in our assay validation plate.
We developed thereafter a stringent data analysis pipeline based on strong proven statistical methods for quality control, normalization and ranking of the results to validate and analyse the screening outcomes as described in the methods section. The three original readouts chosen from the image analysis (count of macrophages, PVs and amastigotes) are de facto independent and were used to define five variables: Total Macrophages (TM), Healthy Macrophages (HM), Viability Index (VI = HM/TM), total number of Amastigotes (Am), and the ratio between PVs and healthy macrophages (PV/HM). These variables allow for precise phenotype discrimination and efficient hit identification. Critical main hands-off steps of the data analysis pipeline are described below [31].
Among the dominant class of 18 compounds that presented a strong anti-leishmanial activity associated with high toxicity to host macrophages a majority of compounds were Paullone derivatives (Figure 3, image C3 for an illustration). Paullones have been described previously as inhibitors of cyclin-dependent kinases and glycogen synthase kinase-3, and also as inhibitors of L. donovani axenic amastigotes [42]. High toxicity described against host cells for these molecules indicated that screening on host-free parasite populations can lead to false positive hits, a conclusion also reached by De Muylder and colleagues [18]. Some Paullone derivatives were reported as either inefficient within THP1 macrophages and/or too toxic for the host cell [42]. A second class of compounds (chalcone derivatives; compounds c10, c13-16) was identified that did not induce any noticeable phenotype in our assay (Figure 3), although they have been previously identified as active against L. donovani axenic amastigotes in the micromolar range [43]. Finally, Acivicin (c49), Aphidicolin (c51) and Phenyltoxamine (c52) did not exhibit any activity on intramacrophagic amastigotes (Figure 3), even though these molecules were described previously as potent growth inhibitors for L. major promastigotes by Sharlow and coworkers [10]. Lack of leishmanicidal activity of these compounds against intramacrophagic amastigotes was certainly due to their inability to cross the host macrophage membranes surrounding the parasites. Control experiments performed on L. amazonensis promastigotes indeed confirmed the activity of these compounds towards host-free parasites, with IC50 values of 0.01 µM, 0.48 µM, and > = 5 µM for Acivicin, Aphidicolin, and Phenyltoxamine, respectively (Figure 4).
Anti-leishmanial compounds were distributed into two phenotypic categories. The first comprised 4 compounds with low leishmanicidal activity at 10 µM on intracellular amastigotes (c29, c38, c40 and c44) (Figure 3). This category includes notably the known drugs Pentamidine Isethionate (c38), a molecule active on viscerotropic and dermotropic Leishmania species [44] different from L. amazonensis, Miconazole (c40) and Clotrimazole (c44), two antifungals with known anti-leishmanial activity [45], which showed variable efficacy depending on therapy conditions and Leishmania species [46]–[48]. Moreover, while Pentamidine Isethionate in particular has been recently validated as a strong growth inhibitor of L. major [10] for both promastigote and axenic amastigote–like stages with an EC50 value similar to the value we obtained with L. amazonensis promastigotes (0.65 µM, Figure 4), it only exhibited low level of intra-macrophage activity (above 10 µM EC50) [47]. Such discrepancies between hit compounds showing leishmanicidal activities against either promastigotes or intramacrophagic amastigotes have already been observed [18], [49], reinforcing the value of our approach.
The second category includes compounds that displayed a strong intra-macrophagic anti-leishmanial phenotype such as our reference compounds, Leu-oMe (c43 and c60) and AmphoB (c55) and four compounds (c28, c32, c34 and c53) (Figure 3) identified by Guiguemde and co-workers for their activity against L. major promastigotes [50].
Our results further reinforce the need and the adequacy of cellular assays such as the one presented here for rapid and successful identification of active molecules on Leishmania cell-cycling amastigotes hosted by primary macrophages [49].
In the present study, we developed a scalable, throughput capable high content approach to select chemicals able to eliminate L. amazonensis amastigotes that are actively multiplying within the acidic giant parasitophorous vacuoles of primary macrophages.
This novel assay is simple and relies on only few experimental steps during which highly pure populations of amastigotes, compounds and fluorescent reporters are sequentially added to adherent macrophages without any washing or addition of fixative reagent. It generated robust and reproducible data based on the manipulation of large homogeneous populations of infected macrophages and allowed a dual measure of leishmanicidal activity against intramacrophagic parasites and the host macrophage health status.
By allowing real-time monitoring and kinetic studies on living adherent primary macrophages, this approach offers many advantages over assays that have been described recently in the literature with respect to assay reproducibility and infection homogeneity [11], [18], [49], [51]. Additionally, it enables the discovery of leishmanicidal compounds acting through the activation of microbicidal mechanisms of the host macrophage. Finally, the compound incubation period of our HCA assay was successfully prolonged up to six days (Figure S6) thus allowing for the discovery of slow-acting leishmanicidal compounds with kinetics similar to antimonials. Analyses of screening campaigns performed with compounds from kinase inhibitor libraries as part of the LeishDrug project sponsored by the EU's Seventh Framework Programme for Research are ongoing [52].
|
10.1371/journal.pgen.1003440 | Sex-Differential Selection and the Evolution of X Inactivation Strategies | X inactivation—the transcriptional silencing of one X chromosome copy per female somatic cell—is universal among therian mammals, yet the choice of which X to silence exhibits considerable variation among species. X inactivation strategies can range from strict paternally inherited X inactivation (PXI), which renders females haploid for all maternally inherited alleles, to unbiased random X inactivation (RXI), which equalizes expression of maternally and paternally inherited alleles in each female tissue. However, the underlying evolutionary processes that might account for this observed diversity of X inactivation strategies remain unclear. We present a theoretical population genetic analysis of X inactivation evolution and specifically consider how conditions of dominance, linkage, recombination, and sex-differential selection each influence evolutionary trajectories of X inactivation. The results indicate that a single, critical interaction between allelic dominance and sex-differential selection can select for a broad and continuous range of X inactivation strategies, including unequal rates of inactivation between maternally and paternally inherited X chromosomes. RXI is favored over complete PXI as long as alleles deleterious to female fitness are sufficiently recessive, and the criteria for RXI evolution is considerably more restrictive when fitness variation is sexually antagonistic (i.e., alleles deleterious to females are beneficial to males) relative to variation that is deleterious to both sexes. Evolutionary transitions from PXI to RXI also generally increase mean relative female fitness at the expense of decreased male fitness. These results provide a theoretical framework for predicting and interpreting the evolution of chromosome-wide expression of X-linked genes and lead to several useful predictions that could motivate future studies of allele-specific gene expression variation.
| With the exception of its most primitive members, mammal species practice X inactivation, where one copy of each X chromosome pair is silenced in each cell of the female body. The particular copy of the X that is silenced nevertheless shows considerable variability among species, and the evolutionary causes for this variability remain unclear. Here, we show that X inactivation strategies are likely to evolve in response to the sex-differential fitness properties of X-linked genetic variation. Genetic variation with similar effects on male and female fitness will generally favor the evolution of random X inactivation, potentially including preferential inactivation of the maternally inherited X chromosome. Variation with opposing fitness effects in each sex (“sexually antagonistic” variation, which includes mutations that both decrease female fitness and enhance male fitness) selects for preferential or complete inactivation of the paternally inherited X. Paternally biased X inactivation patterns appear to be common in nature, which suggests that sexually antagonistic genetic variation might be an important factor underlying the evolution of X inactivation. The theory provides a conceptual framework for understanding the evolution of X inactivation strategies and generates several novel predictions that may soon be tested with modern genome sequencing technologies.
| Mammalian females transcriptionally silence one of their two X chromosomes within each somatic cell – a process called X inactivation [1], [2]. The basic phenomenon of X inactivation occurs in all therian (non egg-laying) mammals studied to date, yet the specific X chromosome silenced exhibits considerable diversity among species. At one extreme, typical of marsupials, the paternally inherited X is universally silenced and the maternally inherited X is ubiquitously expressed ([3], [4]; hereafter referred to as paternal X inactivation or PXI). In contrast, placental mammals practice random X inactivation (RXI): each somatic cell may express either the maternally or the paternally inherited X (the other X is silenced), and female bodies are composed of a mosaic of cells that individually express one of the two X chromosome copies [5]. While RXI is generally thought to be unbiased – with each cell having an equal probability of expressing either of the two X chromosomes – recent data reveal quantitatively biased inactivation patterns in at least some placental mammal species, i.e.: differential silencing of maternally and paternally derived X chromosomes [6], [7]. Several marsupial studies similarly find evidence for partial expression of the paternally derived X, suggesting additional species-specificity of X inactivation rules (reviewed in [2], [8]).
The selective processes that might account for this observed diversity remain unclear. A leading hypothesis for the evolution of RXI is that it might be favored if segregating deleterious mutations have recessive or partially recessive fitness effects ([9]–[11]; which, on average, they do [12]–[15]). The logic underlying this hypothesis is straightforward. Females that uniformly silence a particular copy of the X (e.g., the paternally inherited copy under PXI) will be effectively haploid, and suffer the full fitness costs of mutations carried on their expressed X chromosome. RXI generates an expression pattern that is more similar to diploidy, and can potentially mask the fitness costs of carrying deleterious alleles.
While the masking hypothesis for the evolutionary origins of RXI is plausible (e.g., [1], [16]–[20]), its feasibility should be investigated in a formal population genetic model. Models of a similar evolutionary scenario, the evolution of haploid versus diploid life cycles (e.g., [21]–[25]), indicate that selection for masking of deleterious mutations favors the evolution of diploidy, if mutations are sufficiently recessive relative to the population's recombination rate ([26]–[30]; however, selection to mask somatic mutations eliminates constraints imposed by tight linkage [31]). However, these models do not incorporate the unique properties of sex-differential selection and inheritance that govern X chromosome evolution [32], so it remains unclear whether their conclusions apply to the case of RXI.
Sex differences in selection – where the fitness effects of single mutations differ in magnitude or direction between males and females – likely influence large fractions of animal genomes [33]–[36], which can have two potential consequences for the evolutionary diversification of X inactivation strategies. Stronger selection against deleterious alleles in males compared to females should decrease the average proportion of deleterious alleles carried on each paternally derived X (e.g., [10], [33]), and thereby favor expressing the paternally inherited X. “Sexually antagonistic alleles” – alleles that increase fitness when present in one sex, but decrease fitness in the other sex [37], [38] – should have the opposite effect on X inactivation. Alleles benefiting males and costly to females experience higher probabilities of paternal transmission (e.g., [39], [40]), which could generate selection to preferentially inactivate, or even ubiquitously silence, paternally inherited X chromosomes. Several models have examined how sexually antagonistic selection might favor the evolution genomic imprinting, which similarly involves the unequal expression of maternally and paternally inherited gene copies [40]–[44]. However, the effect of sexually antagonistic fitness variation on X inactivation evolution has yet to be addressed.
It is currently unclear how the population genetic parameters of dominance, sex-differential selection, and linkage and recombination might jointly influence the evolution of X inactivation strategies. We therefore developed a mathematical model of X inactivation evolution, and used this model to identify biological conditions that favor the evolution of different X inactivation states. We first consider the dichotomous case of RXI versus PXI, describe the conditions facilitating evolutionary transitions to RXI, and characterize the consequences of such transitions for mean fitness of males and females. Overall, selection on mammalian X inactivation strategies is primarily mediated by the interaction between dominance and sex-differential selection, and the situation differs considerably from scenarios that favor the evolution of diploidy. We also examine whether sex-differential selection might favor the evolution of biased X inactivation strategies (i.e., quantitatively unequal expression of maternally versus paternally derived X chromosomes) and predict the magnitude of biases likely to evolve. Our results suggest that conditions for evolving biased inactivation patterns are extremely permissive.
We focus on the simplest and most analytically tractable model that simultaneously incorporates genetic linkage, variation for fitness, and variation in the form of X inactivation. Our model follows the evolution of two bi-allelic loci. Locus A (the “fitness locus”) is X-linked and carries A1 and A2 alleles, which directly influence male and female fitness. Locus B (the “modifier locus”) carries B1 and B2 alleles, which influence the X inactivation rule in females within the population. The genotype at locus B can influence female fitness through its effect on the relative expression of A1 versus A2 alleles in heterozygotes. Variation at the B locus has no other phenotypic effect in males or females and is therefore neutral in males. We consider two scenarios of linkage for the B locus. Under X-linkage, A and B are physically linked, and recombine at a rate r, per female meiosis (the X does not recombine in males, which have only one X chromosome copy). When B is on an autosome, alleles at A and B loci segregate independently during meiosis.
To model the evolution of X inactivation, we begin with the B locus initially fixed for allele B1, and the A locus at polymorphic equilibrium given B1 fixed. We then characterize selection on and evolution of a novel B2 allele that changes a female carrier's X inactivation system. The population is assumed to be sufficiently large that genetic drift can be ignored, and each generation is discrete. The life cycle during each generation follows the order of birth, selection, recombination, mutation, random mating and syngamy. Our approach bears many similarities to models for the evolution of ploidy cycles (e.g., [26], [45] chapter 8 of [46]), and for the evolution of sexually dimorphic genomic imprinting [40], [44], with which we draw contrasts. Generalized two-locus recursions (see Methods) include sixteen different female genotypes (when allowing for parent-of-origin effects) and at least four male genotypes (four for the X-linked modifier model; eight for the autosomal modifier model). To reduce the enormous range of possible fitness parameterizations and initial population conditions, we focus our attention on a subset of idealized and biologically relevant population genetic scenarios in the following analyses.
We consider two basic forms of fitness variation at the A locus (Table 1): (1) deleterious alleles maintained at a balance between recurrent mutation and purifying selection; and (2) sexually antagonistic alleles stably maintained as balanced polymorphisms. In both scenarios, the female-deleterious allele is denoted by A1, i.e. female fitness is highest in A2 homozygotes, and fitness of heterozygous females is assumed to be intermediate to the two homozygous genotypes [formally, w(A1A1) = 1−sf≤w(A1A2), w(A2A1)≤w(A2A2) = 1, where w(AiAj) is the fitness of a female with genotype AiAj, and sf describes the fitness cost to females of being homozygous or haploid for the A1 allele: 1>sf>0]. When A1 is also deleterious to males, it will be maintained at mutation-selection balance (sm is the fitness cost to males of carrying an A1 allele; 1>sm>0; Table 1). We also consider genetic polymorphism maintained by sexual antagonism, where A2 is the deleterious allele for males (here, tm is the fitness cost to males of carrying an A2 allele; 1>tm>0; see Table 1).
In an ancestral population fixed for B1, female somatic cell lineages are assumed to silence the paternally inherited X with probability ξ11 and silence the maternal X with probability 1−ξ11 (0≤ξ11≤1). Under unbiased RXI (where ξ11 = ½), heterozygous females have fitness w(A1A2) = w(A2A1) = 1−sfh, where h represents the degree of masking of the A1 allele (0<h<1). Thus, h is a scaling factor that is analogous to the dominance coefficient of standard population genetic models. Use of the terms “dominance” and “recessivity”, applied to species with RXI, has been questioned because individual cells lack bi-allelic expression [47]. However, the dominance coefficient remains useful as a population genetic parameter, and simply quantifies the relative fitness of heterozygous versus homozygous genotypes when heterozygotes practice an unbiased RXI rule. Partial masking is clearly relevant for many X-linked disorders, which tend to be less penetrant and less severe in females than males [47].
When X inactivation is biased (ξ11≠½), fitness is function of h (as defined above) and parent-of-origin effects generated by the X inactivation rule. Assuming that female fitness decreases monotonically with the proportion of cells expressing the A1 allele, we can describe it using a generalized power function, w(x) = 1−xksf, where x represents the proportion of cells expressing the A1 allele, and k is a positive constant that describes the specific shape of the fitness decline associated with A1 expression (such functions are often used in evolutionary theory because of their flexibility; e.g., [48]–[50]). For our purposes, w(x) has two essential properties. First, fitness approaches unity when A2 is ubiquitously expressed (i.e., ), and 1−sf when A1 is ubiquitously expressed (i.e., ), which makes intuitive sense. Second, k can be defined in terms of the dominance coefficient of A1. When ξ11 = ½, x = ξ11 = 1−ξ11; therefore, h = 1/2k and k = −ln(h)/ln(2). A1 is partially dominant to A2 when h>½ (k<1) and partially recessive when h<½ (k>1). Consequently, hmat = ξ11k represents the effective dominance coefficient when A1 is maternally inherited, and hpat = (1−ξ11)k represents the effective dominance coefficient when A1 is paternally inherited (Table 1; Figure 1). Nonadditivity of allelic effects (i.e., h≠½) can arise when the fitness of an X-linked genotype is not cell-autonomous (e.g., fitness depends on the overall proportion of A1 versus A2 expression in female bodies; [43], [51]). With cell-autonomous effects, we can model total fitness as the mean fitness per cell [43], such that w(A1A2) = 1−ξ11sf and w(A2A1) = 1−(1−ξ11)sf, which represents a special case of the generalized power function (i.e., k = 1 and h = ½).
Given the outlined assumptions, we obtain the following equilibria with respect to the X-linked fitness-determining locus. When A1 is deleterious to both sexes, its equilibrium frequencies at mutation-selection balance (in females and males, respectively) are:(1)where u is the mutation rate to A1, hmat = ξ11k, and hpat = (1−ξ11)k, and k = −ln(h)/ln(2). Eq. (1) was previously derived in [52] (see Text S1). When A1 is sexually antagonistic (i.e., deleterious to females but beneficial to males), and with balancing selection acting at the A locus, the equilibrium frequencies of A1 are:(2)versions of which have been derived in several previous studies [43], [52]–[54]. The balancing selection criteria for sexually antagonistic polymorphism are provided (Text S1; for additional results and discussion, see [43], [52], [54]–[56]).
PXI is thought to represent the ancestral X inactivation state, from which RXI evolved [1], [57]. We therefore sought to define the population genetic conditions in which a rare B2 allele that causes unbiased RXI (where ξ12 = ½ represents the X inactivation strategy played by individuals heterozygous at the modifier locus) will invade a population that is initially fixed for the PXI strategy (ξ11 = 1).
Evolutionary transitions that influence gene ploidy levels (e.g., transitions from haploidy to diploidy) permit deleterious mutations to accumulate within populations, and reduce long-term population fitness (i.e., the new equilibrium fitness [21], [67]; however, epistasis can sometimes render diploidy advantageous [23]), which is why recombination is required for diploidy to evolve [26]–[30]. Under tight linkage, alleles for diploidy are co-transmitted with deleterious mutations, and fitness benefits of masking can be outweighed by the increased burden of linked, deleterious mutations. Recombination decouples such associations, and permits diploidy to evolve, despite the long-term fitness cost. Such interactions between ploidy and recombination have parallels in various other aspects of genome evolution, including the evolution of genomic imprinting [44], [68] and the establishment of gene duplicates [69], [70].
In contrast, the evolution of RXI does not require recombination (as shown above), and this insensitivity to linkage can similarly be considered in light of mean fitness changes that follow an evolutionary transition from PXI to RXI. For the mutation-selection balance model, equilibrium mean male and female fitnesses (respectively) under PXI and unbiased RXI are:(5a)and(5b)which each ignore terms of O(u2). For the parameter space where RXI can evolve [h<3/(6−2sm+sf); see eq. (3)], mean male fitness is always lower under RXI (as long as sm>0; otherwise fitness does not change). Mean female fitness becomes higher, as the evolution of RXI shifts some of the burden of purifying selection (removing deleterious mutations) from females to males. The fitness cost to males at the new equilibrium is offset by fitness gains to females, which provides an intuitive explanation as to why selection for RXI is insensitive to linkage. Tightly linked deleterious alleles may hitchhike with a modifier for RXI, yet such associations do not overturn the net benefits of masking.
For the case of an evolutionary transition from PXI to RXI driven by sexually antagonistic polymorphism [in the parameter space where RXI can evolve, i.e.: h<(1−tm)/(2−tm); see eq. (4)], the new equilibrium frequency of the male-beneficial/female-detrimental allele will be higher when tm<sf, and lower when tm>sf. Mean male fitness is therefore increased after the transition when sf>tm, decreased when sf<tm, and otherwise remains unchanged. The mean female fitness is increased under a much broader range of conditions, because the masking effect caused by RXI can sometimes offset a higher derived frequency of the female-deleterious allele. The condition necessary for female fitness to be increased following the evolution of RXI is:(6)where hcrit = (1−tm)/(2−tm). As shown in Figure 3, parameter conditions favoring the evolution of RXI generally lead to an increase in mean female fitness, whereas mean male fitness is increased in exactly half of the relevant parameter space. The new mean fitness is always increased in one sex at least, and in some cases, in both sexes.
For the model of sexually antagonistic genetic variation, our analysis of the change in equilibrium mean fitness follows the evolution of the ancestrally polymorphic locus, and ignores any sexually antagonistic X-linked mutations that might enter the population following the transition to RXI. However, the parameter space that permits an X-linked sexually antagonistic polymorphism is larger under RXI than PXI, within the relevant parameter space where RXI can evolve (PXI can nevertheless be more conducive to polymorphism under alternative dominance parameterizations [52]), and recessive sexually antagonistic alleles that benefit males weakly can more readily be maintained [54], [71]. Therefore, the evolution of RXI could increase the parameter space that permits X-linked sexually antagonistic polymorphism. Antagonistically selected X-linked loci that could not establish balanced polymorphisms in the ancestral population with PXI may do so in an RXI population. In the longer-term, this could increase male and decrease female fitness, contingent on the specific distribution of male and female selection and dominance coefficients among sexually antagonistic mutations.
Thus far we have shown that RXI can be favored by selection if female-detrimental alleles are sufficiently masked when heterozygous. However, selection might not necessarily favor the same rate of inactivation for maternally and paternally derived X chromosomes in females. Sex-differential selection generates allele frequency differences between males and females, which can favor differential expression of genes inherited from opposite-sex parents [40]. With respect to the mammalian X, polymorphism under sex-differential selection might favor the evolution of unequal inactivation rates between paternally and maternally derived X chromosomes; this is conceptually similar to genomic imprinting that involves partial, but unequal, expression of both of the parental gene copies [72]–[79]. Because X inactivation is a female-limited trait, we expect that selection will favor preferential inactivation of the chromosome with a greater probability of carrying female-deleterious alleles.
We tested this intuition by performing an invasion analysis ([46], chapter 12) to determine the evolutionary stability of different X inactivation strategies. To characterize the direction and magnitude of the bias favored by selection, we consider a population initially fixed for an arbitrary inactivation strategy, ξ11 (0<ξ11<1), and at equilibrium for fitness variation given this strategy [e.g., eqs. (1–2)], and identify the ξ11 values that are stable to invasion by an allele that alters the X inactivation ratio of female carriers (i.e. the “evolutionarily stable strategies” [80]).
Under a mutation-selection balance model, the stable paternal X inactivation value () is:(7)The exponent term [−ln(2)/ln(2h)] is positive when A1 is at least partially recessive (h<½), which we assume here and below, and the ratio determines the direction of bias favored by selection. Selection favors preferential inactivation of the paternal X when deleterious alleles are more frequently transmitted to progeny through males than females (>½ when <1), and favors preferential inactivation of the maternal X when deleterious mutations are more frequently transmitted through females (<½ when >1). Eq. (7) can be expressed as an explicit function of the selection and dominance coefficients by assuming that sfh≪1 (this is biologically reasonable given the observed negative association between dominance and effect size of deleterious mutations [12]). The critical ratio becomes:(8)Selection favors preferential inactivation of the maternally derived X when sm>sfh, and the degree of bias may be pronounced when deleterious mutations are poorly masked in females (e.g., when h is closer to ½ than to zero; Figure 4).
Adopting the same analytical approach for the case of sexually antagonistic fitness variation, the equilibrium X inactivation rule is:(9)Given the stated parameter constraints (h<½; 1>tm>0), (1−tm)−ln(2)/ln(2h) will always be positive and less than one, and selection always favors preferential inactivation of the paternally inherited X chromosome. The magnitude of this bias increases with tm and h, similar to the deleterious mutation scenario for fitness variation (Figure 4). When either the selection or the dominance coefficient is sufficiently large, both models of fitness variation can favor large X inactivation biases. Some moderate values of s, t and h (to the left of the x-axis in Figure 4) could lead to large enough biases – on the order of a few percent or more – to be empirically detectable.
X inactivation in mammals can take a variety of specific forms, ranging from strict PXI, to various forms of RXI [1], [2], [6], [7]. This range of observed diversity is particularly striking, as quantitative analyses of the maternally to paternally derived X inactivation ratio are available for relatively few mammalian species to date (the best data coming from a subset of mouse and human tissues). Modern high-throughput, genome-wide and allele-specific gene expression technologies (such as RNA-seq), can now be used to systematically analyze X inactivation ratios in any number of mammalian species or tissues [6], [81], and this new technology promises to reveal much more diversity in X inactivation patterns.
The range of X inactivation rules employed by different species might usefully be considered within a theoretical framework of sex-differential selection. The population genetic models analyzed here reveal broad opportunities for X inactivation evolution, though it is important to note that model predictions hinge upon the capacity of X inactivation ratios to evolve (i.e., there must be genetic variation for X inactivation rules). There is some evidence for alleles in mice that affect X inactivation choice (reviewed in [82], [83]). Moreover, the observed range of X inactivation patterns between mice, humans, and marsupials (see introduction) imply at least some degree of evolutionary lability for the trait. On the other hand, maternally biased X inactivation has not been observed, and it is unclear whether this reflects under-sampling (the phenomenon may exist, but has yet to be documented), intrinsic genetic constraints (as mentioned), or that biased inactivation of the maternally derived X is generally unfavorable. The evolvability of the X inactivation ratio should be considered an important issue that warrants future study.
For reasons of tractability, and in parallel with theory on the evolution of diploidy (e.g., [26], [45]; chapter 8 of [46]), we have focused our analysis on two-locus population genetics models to characterize how different forms of fitness variation and linkage will influence patterns of selection for different X inactivation rules. However, X inactivation affects the expression of many genes simultaneously, and unique patterns of genetic variation among X-linked loci could individually favor discordant X inactivation strategies. We can gain some insight into scenarios that involve multiple X-linked loci, under the assumption that each polymorphic locus contributes independently to selection on a rare, unlinked modifier allele (i.e., we assume loose linkage and no epistasis between fitness loci; net fitness effects of multi-gene haplotypes could instead apply under tight linkage between fitness loci). Under these conditions, criteria for invasion of a modifier allele will be , where λi is the leading eigenvalue associated with the two-locus system for the modifier locus and the ith of n X-linked fitness loci (i = {1, 2, …, n}) [84], [85]. With weak effects of individual fitness loci on the modifier (i.e., λi≈1), the selection coefficient associated with a rare modifier allele will be smod≈nE(λi−1), where E(λi−1) represents the arithmetic mean, and λi−1 represents selection contributed by the ith locus (invasion of the modifier allele requires that smod>0).
To the extent that X-linked loci segregate for deleterious alleles, and these have similar selection and dominance coefficients, the net effects of multiple X-linked loci on a modifier will be reinforcing, and the strength of selection on the X inactivation strategy will increase with the number of contributing loci (i.e., if E(λi−1)≠0, then smod scales approximately linearly with n, the number of contributing loci; this scaling is in agreement with multilocus models for the evolution of diploidy [28]). Although little available data directly bear upon the sex-specific selection and dominance parameters of mammalian X-linked mutations, data from other systems suggest that most mutations have small, at least partially recessive fitness effects (e.g., [12]–[15], [86]), and are deleterious to both sexes [35], [36]. If X-linked mutations have similarly small selection and dominance coefficients under RXI, then fitness variation due to deleterious alleles might favor the evolution of relatively unbiased RXI rules.
Genes that are polymorphic for sexually antagonistic alleles could exert disproportionately strong influence on the evolutionary trajectories of X inactivation evolution. Sexually antagonistic alleles generate selection of a higher order of magnitude than loci at mutation-selection balance (Figure S1). Consequently, selection due to sexually antagonistic polymorphism may plausibly eclipse the cumulative effects of selection from deleterious alleles. The relative contribution of deleterious versus antagonistically selected alleles to X-linked fitness variation is ultimately an empirical question. To the extent that sexually antagonistic fitness variation is common in mammalian populations (as may indeed be the case; e.g., [60], [64]), selection should favor the evolution of biased inactivation of the paternally inherited X.
Conflicts between different X-linked loci over the optimal ratio of maternal to paternal X inactivation could potentially be resolved by the evolution of genomic imprinting at individual X-linked genes. Previous models have considered two scenarios of sexually antagonistic selection driving the evolution of imprinting (i.e., the partial or complete silencing of a maternally or a paternally inherited copy of a single gene; [72]–[79]). First, when the optimal transcription level of a X-linked gene differs between males and females, imprinting can facilitate sex-specific adaptation by generating sexually dimorphic gene expression [41], [42]. Imprinting of the maternally inherited gene is favored under selection for higher transcription levels in females, whereas paternal imprinting is favored at genes selected for higher transcription in males [41]–[43]. Second, in genes polymorphic for sexually antagonistic alleles, sexually dimorphic imprinting can mitigate fitness costs of inheriting harmful alleles, which are preferentially transmitted from opposite-sex parents [40]. Existing models of this latter scenario consider polymorphism and imprinting at an autosomal gene [40], [44], yet the basic processes should apply to the X (as discussed in [40]) – particularly so because X-linkage promotes allele frequency differences and asymmetrical imprinting effects between males and females [41], [42], [87]. X inactivation and gene-by-gene imprinting may serve as complementary mechanisms for optimizing male and female fitness.
Evolution of RXI from an ancestral population with PXI should be relatively unconstrained, provided three conditions are met: (i) there is genetic variation for RXI (i.e., it is evolvable; see above); (ii) X-linked fitness variation is largely caused by segregating deleterious mutations; and (iii) the fitness costs of these mutations are at least partially masked under RXI (i.e., h<½, as seems likely). The availability of mutations to RXI could potentially constrain the convergent evolution of RXI in marsupials [18]. Sexually antagonistic X-linked fitness variability, if common within marsupials, could also promote the evolutionary maintenance of PXI. This hypothesis is plausible, given the pronounced sexual size dimorphism in marsupials relative to other mammalian species (e.g., [88], [89]). Such dimorphism is indicative of strong sexual selection, which could promote the accumulation of sexually antagonistic genetic variation and thereby limit opportunities to evolve RXI.
Once RXI has evolved, evolutionary reversals to strict PXI should face severe evolutionary constraints. In populations with RXI, the filtering of genetic variation by selection in females will strongly depend upon dominance. RXI permits the preferential accumulation of recessive, female-deleterious mutations, because such alleles experience weakened purifying selection. The retention of recessive alleles in populations with RXI should downwardly shift the mean dominance of segregating alleles (relative to the dominance coefficients of spontaneous mutations; e.g., [12], [90]), and increase the cost to females of becoming homozygous or effectively haploid, as they would under PXI. Filtering of mutations based on their dominance coefficients does not eliminate opportunities to evolve biased RXI, but it should severely constrain evolutionary transitions to complete PXI, which completely eliminates effects of masking. This situation is analogous to the coevolution of outcrossing rates and inbreeding depression, with the latter expected to become more severe in outbreeding populations because they shelter recessive alleles from natural selection [91], [92].
Species-specific properties of mutation and genetic variation might predictably affect patterns of selection for different X inactivation strategies. Sexual selection and sex-biased mutation rates are each likely to vary among species, and both processes can influence the relative transmission probabilities of female-deleterious alleles between maternally and paternally derived X chromosomes.
Mammalian mutation rates are often higher in males than females [93], [94], which tends to upwardly bias paternal transmission of deleterious alleles. Consider a population with unbiased RXI and a mutation rate of um and uf in males and females, respectively (see Text S1). With unbiased mutation (um = uf), selection can favor reduced expression of the maternally inherited X when sfh<sm, as implied by eq. (8). Male-biased mutation (um/uf>1) reduces this parameter space to (um/uf)sfh<sm, because males transmit a higher fraction of de-novo mutations to their daughters. We predict that preferential inactivation of the paternally inherited X will be more common (or more severe) in species with strongly male-biased mutation rates.
Sexual selection could similarly favor paternally biased X inactivation. Although strong purifying selection in males via sexual selection can reduce paternal transmission rates of deleterious alleles ([33], [35], [36]; though not all data support this possibility, e.g.: [95], [96]), it will also reduce the frequencies and contributions of deleterious alleles to fitness variation in females. Sexual selection may simultaneously increase the pervasiveness of sexual antagonism [97] and the contribution of sexually antagonistic alleles to female fitness variation. If sexually antagonistic fitness variation increases with the strength of sexual selection, then so should the degree of paternally biased X inactivation.
When the modifier locus, B, is linked to an autosome, the haplotype recursions can again be obtained using similar approaches as described above. Here, the frequency of each haplotype in females will be the same as described above for the specific case of free recombination: r = ½. Haplotype frequencies in males, following selection and meiosis, are modified to:(14)where m1 = m3, m2 = m4, and M = (x1+x3)m1+(x2+x4)m2.
Stability analysis (invasion opportunities for rare B2 alleles) follows the same approach as before, with initial conditions of variation at locus A remaining unchanged. Under the case of variation maintained by sexually antagonistic selection, effects of mutation are ignored (we set u to zero). The generic characteristic polynomial is:(15)with the partial derivatives each evaluated at the equilibrium: = x1 = 1−x2, = y1 = 1−y2, and x3 = x4 = y3 = y4 = 0. Values of and are based on eqs. (1) or (2) as appropriate.
|
10.1371/journal.pgen.1005359 | Sensory Neurons Arouse C. elegans Locomotion via Both Glutamate and Neuropeptide Release | C. elegans undergoes periods of behavioral quiescence during larval molts (termed lethargus) and as adults. Little is known about the circuit mechanisms that establish these quiescent states. Lethargus and adult locomotion quiescence is dramatically reduced in mutants lacking the neuropeptide receptor NPR-1. Here, we show that the aroused locomotion of npr-1 mutants results from the exaggerated activity in multiple classes of sensory neurons, including nociceptive (ASH), touch sensitive (ALM and PLM), and stretch sensing (DVA) neurons. These sensory neurons accelerate locomotion via both neuropeptide and glutamate release. The relative contribution of these sensory neurons to arousal differs between larval molts and adults. Our results suggest that a broad network of sensory neurons dictates transitions between aroused and quiescent behavioral states.
| Animals switch between periods of behavioral arousal and quiescence in response to environmental, developmental, and circadian cues. Little is known about the circuit mechanisms that produce these behavioral states. During larval molts, C. elegans exhibits a sleep-like state (termed lethargus) that is characterized by the absence of feeding and profound locomotion quiescence. We previously showed that mutants lacking the neuropeptide receptor NPR-1 exhibit increased arousal during larval molts, which is in part mediated by increased secretion of an arousal peptide (PDF-1). Here, we compare the circuits regulating arousal in larval molts and adults. We show that a broad network of sensory neurons arouses locomotion but that the impact of each neuron differs between lethargus and adults. We propose that this broad sensory network allows C. elegans to adapt its behavior across a broad range of developmental and physiological circumstances.
| Animals undergo periods of behavioral quiescence and arousal in response to changes in their environment and metabolic state. Arousal is defined as a state of heightened responsiveness to external stimuli coupled with increased motor activity whereas quiescence is associated with diminished responsiveness and motor activity [1]. Quiescence and arousal can persist for minutes to hours. Arousal is associated with fear, stress, hunger, and exposure to sexual partners [1], while quiescence is associated with sleep and satiety [2]. Relatively little is known about the specific circuit mechanisms leading to arousal or quiescence. In particular, it is unclear if similar mechanisms mediate quiescence and arousal in response to different cues, or at different times during development. To address this question, we have analyzed arousal and quiescence of C. elegans locomotion.
During each larval molt, C.elegans undergoes a prolonged period of profound behavioral quiescence, termed lethargus behavior, whereby locomotion and feeding behaviors are inactive for approximately 2 hours [3]. Lethargus has properties of a sleep-like state such as reduced sensory responsiveness and homeostatic rebound of quiescence following perturbation [4]. Several genes and molecular pathways involved in lethargus behavior have been identified [4–11]. Multiple sensory responses are diminished during lethargus, including those mediated by a nociceptive neuron (ASH) [12], and by mechanosensory neurons [11,13].
Mutants lacking NPR-1 Neuropeptide Y (NPY) receptors have been utilized as a model for generalized arousal. NPR-1 inhibits the activity of a central sensory circuit that is defined by gap junctions to the RMG interneuron [14]. In npr-1 mutants, responses mediated by the RMG circuit (e.g. pheromone and oxygen avoidance) are exaggerated, and this heightened acuity is associated with exaggerated locomotion (both during lethargus and in adults) [11,14–16]. Mutations that increase (e.g. npr-1) and decrease (e.g. tax-4 CNG and osm-9 TRPV) RMG circuit activity are associated with locomotion arousal and quiescence respectively [11,14,17,18].
We previously showed that locomotion quiescence during lethargus is dramatically reduced in npr-1 mutants and that this effect requires increased RMG sensory activity [11]. Subsequent studies showed that in microfluidic chambers npr-1 mutants have modest defects in lethargus quiescence when sensory cues are minimized but that dramatic quiescence defects are observed following brief stimulation with light or vibration [19,20]. Taken together, these papers suggest that npr-1 mutants exhibit aroused locomotion as a consequence of enhanced sensory activity.
The arousing effects of the RMG circuit are mediated in part by secretion of a neuropeptide, pigment dispersing factor (PDF-1) [11]. Activation of PDF receptors (PDFR-1) in peripheral mechanosensory neurons enhances sensitivity to vibration, thereby accelerating locomotion. Thus, sensory evoked activity in the RMG circuit arouses locomotion during lethargus through changes in PDF-1 and PDFR-1 signaling. These results raise several interesting questions. Which specific sensory neurons are responsible for arousal? Does the RMG circuit regulate arousal via multiple outputs (i.e. in addition to PDF-1)? Does the RMG circuit function similarly during lethargus and in adults? Is diminished sensory acuity during lethargus required for behavioral quiescence?
Here we show that glutamatergic transmission promotes arousal, we identify glutamatergic neurons and glutamate receptors that mediate arousal, and we show that arousal occurs by distinct mechanisms in lethargus and adult animals.
Adult npr-1 mutants exhibit accelerated locomotion (Fig 1A–1C), as shown in prior studies [21]. Faster adult locomotion suggests that locomotion circuit activity has been altered. Consistent with this idea, npr-1 mutant adults have enhanced sensitivity to the paralytic effects of a cholinesterase inhibitor (aldicarb) (Fig 1D–1F and S2A Fig) [22], indicating increased excitatory transmission at neuromuscular junctions (NMJs). To more directly assess changes in synaptic transmission, we recorded miniature excitatory post-synaptic currents (mEPSCs) in body muscles, which are evoked by acetylcholine (ACh) release at NMJs. The mEPSC rate observed in npr-1 adults was significantly higher than in wild type controls while mEPSC amplitudes were unaltered (Fig 1G–1I). Faster mEPSC rates suggest that ACh release from motor neurons was increased whereas unaltered mEPSC amplitudes imply that muscle responsiveness to secreted ACh was unaffected. By contrast, neither ACh release evoked by depolarizing motor neurons with a stimulating electrode (evoked EPSCs), nor transmission at GABAergic NMJs (assessed by miniature inhibitory post-synaptic currents, mIPSCs) was altered in npr-1 mutants (S1 Fig). This constellation of electrophysiological defects suggests that tonic ACh release (assessed by mEPSC rate) was enhanced in npr-1 mutants, whereas other forms of neurotransmitter release (evoked ACh release and tonic GABA release) were unaffected. Enhanced tonic ACh release at NMJs could account for the accelerated locomotion rate observed in npr-1 adults.
Prior studies showed that several behavioral phenotypes exhibited by npr-1 mutants are caused by enhanced sensitivity to environmental cues. In particular, sensory responses mediated by the RMG circuit are enhanced in npr-1 mutants [14,17,18] and this enhanced sensory acuity is required for accelerated locomotion rates during lethargus [11,20]. We did several experiments to determine if enhanced RMG circuit activity is also required for increased cholinergic transmission in npr-1 adults. A transgene restoring npr-1 expression in the RMG circuit (using the flp-21 promoter) rescued the accelerated locomotion (Fig 1B), enhanced aldicarb sensitivity (Fig 1D and S2A Fig), and faster mEPSC rate (Fig 1G–1I) defects of npr-1 adults. By contrast, an npr-1 transgene expressed in GABAergic neurons lacked rescuing activity (Fig 1D–1H). These results indicate that NPR-1 acts in the RMG circuit to slow adult locomotion. Similarly, mutations inactivating ion channels required for sensory transduction (TAX-4/CNG and OCR-2/TRPV) in the RMG circuit suppressed the npr-1 adult locomotion (Fig 1C), aldicarb sensitivity (Fig 1E and 1F and S2B and S2C Fig), and mEPSC rate (Fig 1J and 1K) defects. Collectively, these results suggest that the accelerated adult locomotion exhibited by npr-1 mutants is caused by heightened activity in the RMG sensory circuit and, consequently, corresponds to an aroused state.
We previously showed that the lethargus quiescence defects exhibited by npr-1 mutants are caused by increased secretion of Pigment dispersing factor (PDF-1) by cells in the RMG circuit [11]. Because PDF-1 secretion is also increased in npr-1 adults [11], we tested the idea that the hyperactive adult locomotion of npr-1 mutants is also caused by increased PDF signaling. Contrary to this idea, we found that pdf-1 and pdfr-1 (PDF Receptor-1) mutations reduced but did not eliminate the aldicarb hypersensitivity (Fig 2A and 2B and S2D and S2E Fig), the accelerated locomotion (Fig 2C), and increased mEPSC rate (Fig 2D and 2E) defects of npr-1 adults. Collectively, these results suggest that additional excitatory outputs from the RMG circuit (i.e. beyond PDF-1) must contribute to the aroused locomotion of npr-1 adults.
Many C. elegans sensory neurons are glutamatergic, including two neurons in the RMG circuit (ASH and ASK) and the body touch neurons [23]. To determine if glutamate release by sensory neurons is required for accelerated locomotion in npr-1 mutants, we analyzed mutations that inactivate the vesicular glutamate transporter (eat-4 VGLUT), which is primarily expressed in sensory neurons [23]. eat-4 VGLUT mutations blocked the increased motile fraction and locomotion speed of npr-1 mutants both during the L4-Adult (L4/A) molt (Fig 3A–3C) and in adults (Fig 3D and 3E). eat-4 mutations also blocked the hypersensitivity to aldicarb (Fig 3F and S2F Fig) and increased mEPSC rate (Fig 3G and 3H) defects of npr-1 adults. Transgenes restoring EAT-4 expression in touch neurons and ASH neurons partially reinstated both lethargus (Fig 3B and 3C) and adult locomotion (Fig 3D and 3E) defects in eat-4; npr-1 double mutants, whereas transgenes expressed in ASK lacked rescuing activity (Fig 3B and 3C). eat-4 transgenes had no effect on lethargus quiescence in wild type animals (S3 Fig). These results suggest that glutamate released by ASH and touch neurons arouses locomotion in L4/A and adult npr-1 mutants.
The preceding results suggest that ASH synaptic output arouses locomotion in npr-1 mutants. We did several additional experiments to test this idea. If altered ASH output were required for aroused locomotion, we would expect that npr-1 mutants lacking ASH neurons would have increased locomotion quiescence. To test this idea, we induced ASH cell death with a transgene that expresses the pro-apoptotic caspase CED-3. Killing ASH significantly decreased the L4/A motile fraction and locomotion rate in npr-1 mutants (Fig 4A–4C). By contrast, ASH ablation had little effect on the locomotion rate of npr-1 adults (Fig 4D).
To determine if ASH activity is increased in npr-1 mutants during lethargus, we examined sensory-evoked calcium responses in ASH, using the genetically encoded calcium indicator Cameleon. ASH mediates avoidance responses to copper and hyper-osmotic stimuli. Consistent with a recent study [12], the magnitude of copper (Fig 4E and 4F) and glycerol-evoked (S4A and S4B Fig) calcium transients in ASH was significantly decreased during lethargus in wild-type animals. Decreased ASH responsiveness to copper and glycerol during L4/A lethargus was blocked in npr-1 mutants, whereas ASH responsiveness in adults was unaltered in npr-1 mutants (Fig 4E and 4F and S4A and S4B Fig). Transgenes expressing NPR-1 in the RMG circuit (using the flp-21 promoter) or in ASH (using the sra-6 promoter) reinstated the L4/A decrease in copper and glycerol-evoked ASH calcium transients in npr-1 mutants (Fig 4G and 4H and S4C and S4D Fig). These results suggest that NPR-1 acts in ASH to inhibit sensory responses and that increased ASH activity is required for accelerated locomotion of npr-1 mutants during lethargus but not in adults.
To determine if increased ASH activity is sufficient to arouse locomotion, we analyzed locomotion after artificially depolarizing ASH neurons. For this experiment, we utilized transgenic animals that express rat TRPV1 capsaicin receptors in ASH neurons [24]. In these animals, capsaicin treatment evokes ASH-mediated avoidance behaviors [24]. A 5-hour capsaicin treatment had little effect on L4/A motile fraction and locomotion velocity [11], whereas capsaicin treatment significantly accelerated adult locomotion and increased aldicarb sensitivity (Figs 4I–5J and S2G Fig). These effects were not observed in animals lacking TRPV1 expression in ASH neurons (Fig 4I and 4J). Thus, forced ASH depolarization was sufficient to arouse adult but not lethargus locomotion. Collectively, these results suggest that diminished and heightened ASH activity is associated with locomotion quiescence and arousal respectively; however, the magnitude of ASH’s arousing effects differ between lethargus and adult animals.
Which glutamate receptors arouse locomotion in npr-1 mutants? Glutamate-activated cation channels, AMPA (GLR-1 and -2) and NMDA (NMR-1 and -2) receptors, mediate excitatory transmission at ASH-interneuron [25–27]. The npr-1 L4/A quiescence defect was abolished in glr-2; npr-1 double mutants (Fig 5A–5C), while glr-1 mutations had no effect (Fig 5D and 5E). By contrast, glr-1, glr-2, and nmr-1 mutations had little effect on npr-1 adult locomotion (Fig 5F and S5 Fig). Similarly, glr-2 mutations did not block the increased mEPSC rate in npr-1 adults (Fig 5G). These results suggest that GLR-2 AMPA receptors are specifically required for the aroused locomotion during the L4/A lethargus in npr-1 mutants.
Which synaptic targets of ASH and touch neurons mediate locomotion arousal? To address this question, we identified the neurons in which GLR-2 function is required. Aroused L4/A locomotion requires GLR-2 but not GLR-1 receptors; consequently, we reasoned that the relevant neurons are likely to express GLR-2 but not GLR-1. GLR-1 and GLR-2 are co-expressed in many neurons; however, a few GLR-2-expressing neurons lack GLR-1, including DVA (a stretch-activated neuron) and AIA (an interneuron in the head ganglia) [25–27]. The L4/A quiescence defect was partially restored in glr-2; npr-1 double mutants by transgenes expressing GLR-2 in DVA and AIA neurons, whereas transgenes expressed in the ventral cord interneurons (using the glr-1 promoter) failed to rescue (Fig 5B and 5C). Transgenic expression of GLR-2 in DVA or AIA had no effect on lethargus quiescence in wild type worms (S3 Sig). These results suggest that GLR-2 AMPA receptors expressed in AIA and DVA neurons arouse L4/A locomotion in npr-1 mutants. DVA receives direct synaptic input from the touch neuron PLM while AIA receives direct input from ASH [28]. Thus, increased transmission at ASH-AIA and PLM-DVA synapses could account for GLR-2’s effects on locomotion rate. Because we only observed partial rescue by glr-2 transgenes expressed in AIA and DVA, it is likely the GLR-2 function is required in additional (as yet unidentified) neurons.
How do AIA and DVA arouse locomotion? AIA neurons provide synaptic input to ASK and ASI, both of which express PDF-1 [11,29]. Thus, heightened AIA activity could arouse locomotion by enhancing PDF-1 secretion. To assess the level of PDF-1 secretion, we analyzed PDF-1::YFP fluorescence in the endolysosomal compartment of coelomocytes, which are specialized scavenger cells that internalize proteins secreted into the body cavity [30,31]. Inactivating GLR-2 did not alter PDF-1::YFP fluorescence in coelomocytes in both adult and L4/A animals (Fig 6). These results suggest that the arousing effects of GLR-2 are not mediated by changes in PDF secretion. DVA neurons receive direct synaptic input from the PLM touch neurons [32], and secrete NLP-12 (a neuropeptide that accelerates locomotion) [33]. Thus, increased DVA activity could contribute to locomotion arousal in npr-1 mutants. Three results support this idea. First, PLM neurons exhibit enhanced touch-evoked calcium responses in adult npr-1 mutants (S6 Fig). Thus, PLM neurons have increased sensory acuity in npr-1 mutants, similar to the effect we previously showed for ALM neurons [11]. Second, inducing DVA cell death (with a CED-3 transgene) significantly reduced npr-1 locomotion rate during L4/A lethargus (Fig 5H–5J), but not in adults (Fig 5K). Third, DVA secretion of NLP-12 is significantly increased in npr-1 mutants [33], indicating increased DVA activity. These results suggest that PLM neurons provide enhanced excitatory input to DVA in npr-1 mutants, which promotes aroused L4/A locomotion.
To investigate the circuit mechanisms for arousal, we analyzed the locomotion of npr-1 mutants in awake (adult) and quiescent (lethargus) states. Our results lead to five conclusions. First, multiple classes of sensory neurons contribute to arousal. Second, diminished sensory acuity is a circuit mechanism for promoting behavioral quiescence. Third, glutamate and neuropeptides are utilized as excitatory outputs from sensory neurons to arouse locomotion. Fourth, different mechanisms are utilized to arouse locomotion at different times during development. And fifth, we provide further evidence that arousal mechanisms are conserved across phylogeny.
Multiple classes of sensory neurons arouse locomotion during lethargus and in adults, including: mechanosensory neurons (ALM and PLM), a nociceptive neuron (ASH), a pheromone sensing neuron (ASK), and a stretch sensing neuron (DVA). Lethargus quiescence is accompanied by diminished sensory-evoked responses in ALM, PLM, and ASH (this study and [11–13]). PDF-1 secretion from ASK neurons is significantly reduced during lethargus, implying that ASK neurons also have diminished activity during lethargus [11]. npr-1 mutations prevent the dampened ALM (mechanosensory) and ASH (nociceptive) responses during lethargus and this was accompanied by decreased locomotion quiescence (this study and [11]). The arousing effects of npr-1 mutations are blocked (or diminished) by mutations that decrease sensory responsiveness (e.g. tax-4 CNG and osm-9 TRPV mutations) [11], or by ablating sensory neurons (e.g. ASH and DVA). Forced activation of ASH neurons arouses adult locomotion. Collectively, these results imply that a broad network of sensory neurons arouses locomotion, which allows C. elegans to adapt its behavior across a broad range of developmental and physiological circumstances.
NPR-1 promotes behavioral quiescence by diminishing the sensitivity of many sensory modalities. NPR-1 directly inhibits ASH responses and indirectly inhibits other sensory neurons (ALM, PLM, and DVA) via decreased glutamate and neuropeptide release. Thus, gating of sensory perception by NPR-1 provides a circuit mechanism for producing aroused and quiescent locomotion in C. elegans.
Our results do not exclude the possibility that additional mechanisms (beyond sensory gating by NPR-1) contribute to arousal and quiescence. Both quiescence (during lethargus) and arousal (following molts) persist in microfluidic chambers where many sensory cues are minimized [19]. In particular, oxygen tension is likely to be very low in these chambers, which would greatly diminish NPR-1’s effects on behavior [15,16]. Thus, the quiescence and arousal exhibited in microfluidic chambers implies that additional mechanisms beyond NPR-1 must contribute to expressing these behavioral states. It will be interesting to determine if these NPR-1 independent mechanisms also act by gating sensory activity.
Sensory neurons release glutamate and/or neuropeptides in response to external cues, which then engage downstream motor circuits in behavioral outputs. Our prior study shows that sensory-evoked PDF-1 secretion promotes locomotion arousal by enhancing touch neuron responsiveness. Neuropeptides also mediate arousal in flies (PDF) [34], fish and mammals (orexin/hypocretin) [35,36].
Here we show that sensory evoked glutamate release also plays a role in arousal. Mutations inactivating the EAT-4/VGLUT decreased locomotion arousal in lethargus and in adults. EAT-4 is almost exclusively expressed in sensory neurons [23] and transgenes restoring EAT-4 in touch neurons and ASH neurons re-instates locomotion arousal in npr-1 mutants. These results suggest that sensory neurons utilize both glutamate and neuropeptides as excitatory outputs to arouse locomotion.
Our results suggest that exaggerated glutamate release at ASH-AIA and PLM-DVA synapses arouses locomotion during lethargus in npr-1 mutants. ASH and PLM neurons have enhanced sensory evoked activity in npr-1 mutants, which is expected to produce enhanced glutamate release at ASH-AIA and PLM-DVA synapses. GLR-2 receptors are expressed in AIA and DVA. glr-2 mutations block the aroused L4/A locomotion of npr-1 mutants and arousal is re-instated by transgenes expressing GLR-2 in AIA and DVA. Finally, calcium responses in AIA [14], and neuropeptide secretion from DVA [33] are both enhanced in npr-1 mutants, indicating that these neurons have increased activity. We observed only partial rescue of aroused locomotion by transgenes restoring EAT-4 expression in ASH and touch neurons or by those expressing GLR-2 in AIA or DVA; consequently, it is likely that glutamate released by other sensory neurons also contributes to the aroused L4/A locomotion in npr-1 mutants.
Much less is known about the role of glutamate in arousal in other systems. Glutamate release has widespread effects throughout the brain in mammals, which complicates the analysis of its effects on arousal. Microinjection of glutamate or AMPA into lateral hypothalamic area increased locomotor activity and duration of waking episodes in rodents [37,38], while microdialysis of CNQX, an AMPA receptor antagonist, into the thalamus promotes sleep in cats [39]. Glutamate also induces fictive locomotion in lamprey [40]. In these cases, however, the circuit mechanisms underlying glutamate’s arousing effects are not known.
Mutants lacking NPR-1 exhibit accelerated locomotion in adults and during lethargus [11,18]. Several results suggest that locomotion arousal in adult and lethargus is established by a shared central sensory circuit. First, in both adult and lethargus, enhanced activity in the RMG sensory circuit accelerates locomotion, whereas decreased sensory transduction in the RMG circuit (i.e. by inactivating TAX-4 or OSM-9) abolishes npr-1’s hyperactive locomotion defect [11,14], suggesting that the RMG circuit activity stimulates arousal in both awake and quiescent states. Second, EAT-4 acts in ASH and touch neurons to mediate hyperactive locomotion of npr-1 adult and lethargus stage animals, suggesting that glutamate release from these sensory neurons is required for locomotion arousal in npr-1 mutants.
On the other hand, several results suggest that the mechanisms that arouse locomotion differ between adult and lethargus animals. Inactivating GLR-2 AMPA receptors blocks the hyperactive locomotion of npr-1 mutants during lethargus but not in adults. Aroused locomotion in npr-1 adults persists in glr-1, glr-2, and nmr-1 mutants, indicating that other glutamate receptors are responsible for arousing adult locomotion. Similarly, artificial activation of ASH accelerates adult but not lethargus locomotion. Collectively, our results suggest that multiple sensory circuits govern locomotion arousal throughout development but that the relative contribution of each circuit to arousal differs depending on the developmental stage.
Strain maintenance and genetic manipulation were performed as described [41]. Animals were cultivated at 20°C on agar nematode growth media (NGM) seeded with OP50 (for imaging and behavior) or HB101 E.coli (for electrophysiology). Wild type reference strain was N2 Bristol. Strains used in this study are as follows:
KP6048 npr-1(ky13) X
DA609 npr-1(ad609) X
KP6064 npr-1(ok1447) X
PR678 tax-4(p678) III
CX4544 ocr-2(ak47) IV
LSC27 pdf-1(tm1996) III
KP6340 pdfr-1(ok3425) III
MT6308 eat-4(ky5) III
KP0004 glr-1(n2461) III
VM487 nmr-1(ak4) II
KP6057 ocr-2(ak47) IV;npr-1(ok1447) X
KP6058 ocr-2(ak47) IV;npr-1(ky13) X
KP6060 tax-4(p678) III;npr-1(ky13) X
KP6061 tax-4(p678) III;npr-1(ok1447) X
KP6100 pdf-1(tm1996) III;npr-1(ky13) X
KP6410 pdfr-1(ok3425) III;npr-1(ky13) X
KP6349 eat-4(ky5) III; npr-1(ky13) X
CX4978 kyIs200[sra-6p::VR1, elt-2p::NLS-gfp] (Gift from Cori Bargmann)
KP6414 nmr-1(ak4) II; npr-1(ky13) X
KP6415 glr-1(n2461) III;npr-1(ky13) X
VM1123 dpy-19(n1347) glr-2(ak10) III
KP6740 dpy-19(n1347) glr-2(ak10) III; npr-1(ky13) X
KP7362 npr-1(ky13) X; nuIs439[nlp-12p::GFP]; nuIs519[nlp-12p::ced-3::GFP, vha-6::mCherry]
KP6693 nuIs472 [pdf-1p::pdf-1::venus, vha-6p::mCherry]
KP6743 npr-1(ky13) X; nuIs472
KP7194 dpy-19(n1347) glr-2(ak10) III; nuIs472
KP7195 dpy-19(n1347) glr-2(ak10) III; npr-1(ky13) X; nuIs472
AQ906 bzIs17[mec-4p::YC2.12]
KP6681 npr-1(ky13) X; bzIS17
CX9396 npr-1(ad609) X;kyEx1966[flp-21p::npr-1 SL2 GFP, ofm-1p::dsRed] (Gift from Cori Bargmann)
KP6051 npr-1(ad609) X;nuEx1519[unc-25p::npr-1::gfp, myo-2p::NLS-mCherry]
KP6053 npr-1(ad609) X;nuEx1520[unc-30p::npr-1::gfp, myo-2p::NLS-mCherry]
KP7149, KP7150 eat-4(ky5) III; npr-1(ky13) X; nuEx1613-1614[sra-6p::eat-4, myo-2p::NLS-mCherry]
KP7176, KP7177 eat-4(ky5) III; npr-1(ky13) X; nuEx1615-1616[sra-9p::eat-4, vha-6p::mCherry]
KP7198, KP7199 eat-4(ky5) III; npr-1(ky13) X; nuEx1640-1641[mec-4p::eat-4, vha-6p::mCherry]
KP7442 npr-1(ky13) X; nuEx1684[sra-6p::ced-3::GFP, sra-6p::mCherry, vha-6p::mCherry]
KP7633 nuEx1613[sra-6p::eat-4, myo-2p::NLS-mCherry]
KP7634 nuEx1640[mec-4p::eat-4, vha-6p::mCherry]
AQ3304 ljEx239[sra-6::YC.360]
KP7353 npr-1(ky13) X; ljEx239
KP7443 npr-1(ky13) X; ljEx239; nuEX1607[flp-21p::npr-1, myo-2p::NLS-mCherry]
KP7495 npr-1(ky13) X; ljEx239; nuEX1683[sra-6p::npr-1, vha-6p::mCherry]
KP7191 dpy-19(n1347) glr-2(ak10) III; npr-1(ky13) X; nuEx1637[nlp-12p::glr-2(gDNA),myo-2p::NLS-mCherry]
KP7192 dpy-19(n1347) glr-2(ak10) III; npr-1(ky13) X; nuEx1638[gcy-28(d)p::glr-2(gDNA),vha-6p::mCherry]
KP7354, KP7355, KP7356 dpy-19(n1347) glr-2(ak10) III; npr-1(ky13) X; nuEx1642-1644[glr-1p::glr-2(gDNA), vha-6p::mCherry]
KP7635 nuEx1637[nlp-12p::glr-2(gDNA),myo-2p::NLS-mCherry]
KP7636 nuEx1638[gcy-28(d)p::glr-2(gDNA),vha-6p::mCherry]
Transgenic strains were generated by microinjection of various plasmids with coinjection markers (myo-2p::NLS-mCherry (KP#1480) and vha-6p::mcherry (KP#1874)). Injection concentration was 40–50 ng/μl for all the expression constructs and 10 ng/μl for coinjection markers. The empty vector pBluescript was used to bring the final DNA concentration to 100 ng/μl. The flp-21 promoter (which is expressed in the RMG, ASH, ADL, ASK, URX, and ASI neurons [14]) was used to express transgenes in the RMG circuit.
Lethargus locomotion was analyzed as previously described [11]. Well-fed late L4 animals were transferred to full lawn OP50 bacterial plates. After 1 hour, locomotion of animals in lethargus (determined by absence of pharyngeal pumping) was recorded on a Zeiss Discovery Stereomicroscope using Axiovision software. Locomotion was recorded at 2 Hz for 60 seconds. Centroid velocity of each animal was analyzed at each frame using object-tracking software in Axiovision. Motile fraction of each animal was calculated by dividing the number of frames with positive velocity value with total number of frames. Speed of each animal was calculated by averaging the velocity value at each frame. Quantitative analysis was done using a custom written MATLAB program (Mathworks). Statistical significance was determined using one-way ANOVA with Tukey test for multiple comparisons and two-tailed Student’s t test for pairwise comparison.
Locomotion of adult animals was analyzed with the same setup as lethargus locomotion analysis described above, except that well-fed adult animals were monitored 1–1.5hr after the transfer to full lawn OP50 bacterial plates. For the capsaicin treatment (Fig 4I), 1 day-old animals were transferred to NGM plates containing 50 μM capsaicin (with food), treated with capsaicin for 5 hours, and recorded for their locomotion. Statistical significance was determined using one-way ANOVA with Tukey test for multiple comparisons and two-tailed Student’s t test for pairwise comparison.
Neurons were ablated in npr-1(ky13) mutant worms by transgenes co-expressing CED-3 and a fluorescent protein (GFP or mCherry) under the sra-6 (ASH ablation) or nlp-12 (DVA ablation) promoters. ASH or DVA ablations were confirmed after locomotion analysis by fluorescence microscopy.
Sensitivity to aldicarb was determined by analyzing the time course of paralysis following treatment with 1 mM aldicarb (Sigma-Aldrich) as previously described [42]. Briefly, movement of animals was assessed by prodding animals with a platinum wire every 10 minute following exposure to aldicarb. 20–30 animals were tested for each trial. For the capsaicin treatment (Fig 4J), adult animals were transferred to NGM plates containing 50 μM capsaicin (with food), treated with capsaicin for 2–3 hours, and assayed for their paralysis on 1 mM aldicarb plates containing 50 μM capsaicin.
Electrophysiology was performed on dissected adult worms as previously described [43]. Worms were superfused in an extracellular solution containing 127 mM NaCl, 5 mM KCl, 26 mM NaHCO3, 1.25 mM NaH2PO4, 20 mM glucose, 1 mM CaCl2, and 4 mM MgCl2, bubbled with 5% CO2, 95% O2 at 20°C. Whole cell recordings were carried out at –60 mV using an internal solution containing 105 mM CsCH3SO3, 10 mM CsCl, 15 mM CsF, 4mM MgCl2, 5mM EGTA, 0.25mM CaCl2, 10mM HEPES, and 4 mM Na2ATP, adjusted to pH 7.2 using CsOH. Under these conditions, we only observed endogenous acetylcholine EPSCs. To record GABAergic postsynaptic currents, the holding potential was 0 mV, at which we only observe mIPSCs. All recording conditions were as described [44]. To record evoked EPSCs, a 0.4 ms, 30 μA square pulse was applied to a motor neuron cell body with a stimulating electrode placed near the ventral nerve cord (one muscle distance from the recording pipette). Statistical significance was determined using one-way ANOVA with Tukey test for multiple comparisons and two-tailed Student’s t test for pairwise comparison.
Quantitative imaging of coelomocyte fluorescence was performed as previously described [11] using a Zeiss Axioskop equipped with an Olympus PlanAPO 100x (NA = 1.4) objective and a CoolSNAP HQ CCD camera (Photometrics). Worms were immobilized with 30 mg/ml BDM (Sigma). The anterior coelomocytes were imaged in L4/A lethargus (determined by absence of pharyngeal pumping), and 1 day-old adult animals. Image stacks were captured and maximum intensity projections were obtained using Metamorph 7.1 software (Universal Imaging). YFP fluorescence was normalized to the absolute mean fluorescence of 0.5 mm FluoSphere beads (Molecular Probes). Statistical significance was determined using Kolmogorov-Smirnov test.
Using Dermabond topical skin adhesive, individual worms were glued to 2% agarose pads in extracellular saline (145 mM NaCl, 5 mM KCl, 1 mM CaCl2, 5 mM MgCl2, 20 mM D-glucose, and 10 mM HEPES buffer [pH7.2]). To image copper and glycerol responses, single animals were placed in a perfusion chamber (RC-26GLP,Warner Instruments) under a constant flow rate (0.4 ml min-1) of buffer using a perfusion pencil (AutoMate). Outflow was regulated using a peristaltic pump (Econo Pump, Bio-Rad). 10mM CuCl2 (copper(II)chloride dihydrate, Sigma) or 500mM glycerol (Fisher) were delivered using the perfusion, pencil and switch between control and stimulus solutions was done using manually controlled valves. Solutions contained either 10mM CuCl2 in M13 buffer or 500mM glycerol in 40mM NaCl, 1 mM MgSO4, 1 mM CaCl2 and 5 mM KPO4. The stimulus was delivered for 10 seconds starting on the 10th second from the beginning of the movie. Optical recordings were performed on a Zeiss Axioskop 2 upright compound microscope equipped with a Dual View beam splitter and a Uniblitz Shutter. Images were recorded at 10 Hz using an iXon EM camera (Andor Technology) and captured using IQ1.9 software (Andor Technology). For ratiometric imaging, ROIY tracked the neuron in the yellow channel, and in the cyan channel, ROIC moved at a fixed offset from ROIY. F was computed as FY/FC following a correction for bleed through. No correction for bleaching was required. Ratio changes were detected and parametrized using scripts for MATLAB (The Mathworks). Briefly, the scripts average the F value for 5 preceding and including the marked start stimulus frame (F0) and the 5 frames centered on the marked peak frame (F1). ΔF was equal to (F1—F0) / F0 x 100. Touch-evoked calcium responses in PLM neurons were analyzed as previously described [11]. Statistical significance was determined using one-way ANOVA with Tukey test for multiple comparisons.
|
10.1371/journal.pntd.0002587 | Revisiting the Central Metabolism of the Bloodstream Forms of Trypanosoma brucei: Production of Acetate in the Mitochondrion Is Essential for Parasite Viability | The bloodstream forms of Trypanosoma brucei, the causative agent of sleeping sickness, rely solely on glycolysis for ATP production. It is generally accepted that pyruvate is the major end-product excreted from glucose metabolism by the proliferative long-slender bloodstream forms of the parasite, with virtually no production of succinate and acetate, the main end-products excreted from glycolysis by all the other trypanosomatid adaptative forms, including the procyclic insect form of T. brucei.
A comparative NMR analysis showed that the bloodstream long-slender and procyclic trypanosomes excreted equivalent amounts of acetate and succinate from glucose metabolism. Key enzymes of acetate production from glucose-derived pyruvate and threonine are expressed in the mitochondrion of the long-slender forms, which produces 1.4-times more acetate from glucose than from threonine in the presence of an equal amount of both carbon sources. By using a combination of reverse genetics and NMR analyses, we showed that mitochondrial production of acetate is essential for the long-slender forms, since blocking of acetate biosynthesis from both carbon sources induces cell death. This was confirmed in the absence of threonine by the lethal phenotype of RNAi-mediated depletion of the pyruvate dehydrogenase, which is involved in glucose-derived acetate production. In addition, we showed that de novo fatty acid biosynthesis from acetate is essential for this parasite, as demonstrated by a lethal phenotype and metabolic analyses of RNAi-mediated depletion of acetyl-CoA synthetase, catalyzing the first cytosolic step of this pathway.
Acetate produced in the mitochondrion from glucose and threonine is synthetically essential for the long-slender mammalian forms of T. brucei to feed the essential fatty acid biosynthesis through the “acetate shuttle” that was recently described in the procyclic insect form of the parasite. Consequently, key enzymatic steps of this pathway, particularly acetyl-CoA synthetase, constitute new attractive drug targets against trypanosomiasis.
| Many protists, including parasitic helminthes, trichomonads and trypanosomatids, produce acetate in their mitochondrion or mitochondrion-like organelle, which is excreted as a main metabolic end-product of their energy metabolism. We have recently demonstrated that mitochondrial production of acetate is essential for fatty acid biosynthesis and ATP production in the procyclic insect form of T. brucei. However, acetate metabolism has not been investigated in the long-slender bloodstream forms of the parasite, the proliferative forms responsible for the sleeping sickness. In contrast to the current view, we showed that the bloodstream forms produce almost as much acetate from glucose than the procyclic parasites. Acetate production from glucose and threonine is synthetically essential for growth and de novo synthesis of fatty acids of the bloodstream trypanosomes. These data highlight that the central metabolism of the bloodstream forms contains unexpected essential pathways, although minor in terms of metabolic flux, which could be targeted for the development of trypanocidal drugs.
| Trypanosoma brucei is a unicellular eukaryote, belonging to the protozoan order Kinetoplastida that causes sleeping sickness in humans and economically important livestock diseases [1]. This parasite undergoes a complex life cycle during transmission from the bloodstream of a mammalian host (bloodstream forms of the parasite - BSF) to the alimentary tract (procyclic form - PF) and salivary glands (epimastigote and metacyclic forms) of a blood feeding insect vector, the tsetse fly. In the bloodstream of the mammalian host, the pleomorphic BSF strains proliferate as “long-slender” BSF (LS-BSF) and differentiate into the non-proliferative “short-stumpy” trypanosomes (SS-BSF), which are preadapted for differentiation into PF in the insect midgut [2]. The environmental changes encountered by the parasite require significant morphological and metabolic adaptations, as exemplified by important qualitative and quantitative differences in glucose metabolism between BSF and PF [3], [4].
PF living in the tsetse fly midgut – where glucose is scarce or absent – have developed an elaborate energy metabolism based on amino acids, such as proline. However, when grown in standard glucose-rich conditions, they prefer glucose to proline as a carbon source [5], [6]. PF converts glucose into the partially oxidized and excreted end-products, acetate and succinate, with most of the glycolysis taking place in specialized peroxisomes called glycosomes [7]. In the course of glycolysis, phosphoenolpyruvate (PEP) is produced in the cytosol, where it is located at a branching point to feed the glycosomal ‘succinate branch’ and the mitochondrial ‘acetate and succinate branches’ (see Fig. 1). For the “succinate branches”, PEP must re-enter the glycosomes where it is converted into malate and succinate within that compartment. Malate, which moves from the glycosomes into the mitochondrion, can also be converted into succinate therein. Additionally, PEP can be converted in the cytosol into pyruvate to feed the ‘acetate branch’ (steps 1–4 in Fig. 1). In the mitochondrion, pyruvate is converted by the pyruvate dehydrogenase complex (PDH, EC 1.2.4.1, step 1) into acetyl-CoA and then into acetate by two different enzymes, i.e. acetate∶succinate CoA transferase (ASCT, EC 2.8.3.8, step 2) and acetyl-CoA thioesterase (ACH, EC 3.1.2.1, step 3) [8]–[10]. In PF, acetate production plays an important role for mitochondrial ATP production by the ASCT/SCoAS cycle (steps 2 and 4), while ACH is not involved in ATP production [10]. Acetate can also be produced from threonine, a major carbon source of PF present in the in vitro medium [6], [11], [12]. This amino acid is converted into acetate by threonine-3-dehydrogenase (TDH, EC 1.1.1.103, step 5), acetyl-CoA∶glycine C acetyltransferase (EC 2.3.1.29, step 6) and probably ASCT and/or ACH. We recently showed that PF uses a new metabolic pathway only observed in PF trypanosomes so far, named the “acetate shuttle”, which transfers acetyl-CoA from the mitochondrion to the cytosol to feed the essential cytosolic fatty acid biosynthesis [13]. In this shuttle, acetate produced in the mitochondrion from acetyl-CoA is exported in the cytosol and converted back into acetyl-CoA by the cytosolic acetyl-CoA synthetase (AMP-dependent enzyme, AceCS, EC 6.2.1.1, step 7).
In contrast to PF, BSF trypanosomes rely only on glucose for their energy production, with a 5- to 10-fold higher rate of glucose consumption [14]. It is generally accepted that the proliferative LS-BSF grown under aerobiosis convert glucose exclusively into pyruvate [15], [16], although excretion of trace amounts of other incompletely oxidized end-products such as glycerol, succinate and alanine have been reported [14], [17]. These minor glycolytic end-products are thought to be produced by “contaminating” non-proliferative SS-BSF trypanosomes that have developed a more elaborated central metabolism with a number of PF traits, including production of acetate and succinate from glycolysis [18], [19]. Consequently, recent reports consider that LS-BSF trypanosomes do not produce acetate from glucose. In the seventies, acetate production was reported from the threonine degradation in both PF and BSF trypanosomes [12], however, this metabolic pathway was not further investigated.
Here we investigated the role of glucose and threonine degradation in acetate production in the monomorphic 427 BSF strain, which proliferates as LS-BSF trypanosomes and has lost the ability to differentiate into non-proliferative SS-BSF [20], [21]. This BSF cell line produces and excretes acetate, as a minor end-product of glucose metabolism, with a metabolic flux in the same range as observed for PF. Glucose and threonine contribute almost equally to acetate production, which is essential for the viability of proliferative BSF trypanosomes, as demonstrated by reverse genetics approaches. Our data reveal unexpected metabolic similarities between PF and LS-BSF trypanosomes.
The bloodstream form of T. brucei 427 90-13 (TetR-HYG T7RNAPOL-NEO), a 427 221a line (MiTat 1.2) designed for the conditional expression of genes, was cultured at 37°C in IMDM (Iscove's Modified Dulbecco's Medium, Life Technologies) supplemented with 10% (v/v) heat-inactivated fetal calf serum (FCS), 0.25 mM ß-mercaptoethanol, 36 mM NaHCO3, 1 mM hypoxanthine, 0.16 mM thymidine, 1 mM sodium pyruvate, 0.05 mM bathocuprone and 2 mM L-cysteine [22]. To prepare threonine-depleted IMDM medium all the compounds constituting the medium, except threonine, were purchased from Sigma-Aldrich. The procyclic form of T. brucei EATRO1125 was cultured at 27°C in SDM79 medium containing 10% (v/v) heat-inactivated fetal calf serum and 35 µg/mL hemin [23].
Replacement of the threonine-3-dehydrogenase (TDH: Tb927.6.2790) by the puromycin (PAC) and blasticidin (BSD) resistance markers via homologous recombination was performed with DNA fragments containing a resistance marker gene flanked by the TDH UTR sequences. The TDH knock out was generated in the 427 90-13 BSF parental cell line, which constitutively expresses the T7 RNA polymerase gene and the tetracycline repressor under the control of a T7 RNA polymerase promoter for tetracycline inducible expression (TetR-HYG T7RNAPOL-NEO) [24]. Transfection and selection of drug-resistant clones were performed as previously reported using the Nucleofactor system [25]. The first and second TDH alleles were replaced by puromycin- and blasticidin-resistant genes, respectively. Transfected cells were selected in IMDM medium containing hygromycin B (5 µg/mL), neomycin (2.5 µg/mL), puromycin (0.1 µg/mL) and blasticidin (10 µg/mL). The selected cell line (TetR-HYG T7RNAPOL-NEO Δtdh::PAC/Δtdh::BSD) is called Δtdh.
Accession numbers (http://www.genedb.org/genedb/tryp/) of genes targeted by RNAi, acetyl-CoA synthetase (AMP-dependent enzyme, AceCS) and E2 subunit of the pyruvate dehydrogenase complex (PDH-E2), are Tb927.8.2520 and Tb927.10.7570, respectively. RNAi-mediated inhibition of gene expression in the 427 90-13 BSF parental cell line was performed by expression of stem-loop “sense/anti-sense” RNA molecules of the targeted sequences introduced into the pHD1336 (kindly provided by C. Clayton, ZMBH, Heidelberg, Germany, [email protected]). The AceCS-SAS and PDH-E2-SAS “sense/anti-sense” constructs were first generated in the pLew100 vector (kindly provided by E. Wirtz and G. Cross) [24] as described before [5], [13]. Then the AceCS-SAS and PDH-E2-SAS HindIII-BamHI cassettes extracted from the pLew100 plasmids were inserted in HindIII-BamHI digested pHD1336 vector, which contains the blasticidin resistance gene. The RNAi-harboring RNAiAceCS and RNAiPDH single mutant cell lines were produced by transfection of the 427 90-13 cell line with the NotI-linearized pHD-AceCS-SAS and pHD-PDH-E2-SAS plasmids, respectively, and selected in IMDM medium containing hygromycin B (5 µg/mL), neomycin (2.5 µg/mL) and blasticidin (10 µg/mL). For transfection of the Δtdh cell line with the pLew-PDH-E2-SAS construct, all of the four antibiotics used to select the Δtdh cell line, in addition to phleomycin (2.5 µg/mL), were included in the medium to select double mutant cell lines.
A recombinant fragment containing the full-length TDH gene was inserted into the NdeI and BamHI restriction sites of the pET28a expression vectors (Novagen) to express in BL21 Escherichia coli the TDH protein preceded by a N-terminal histidine tag (6 histidine codons). Cells were harvested by centrifugation, and recombinant proteins purified by nickel chelation chromatography (Novagen) from the insoluble fraction according to the manufacturer's instructions. The anti-TDH immune serum was raised in rabbits by five injections at 15-day intervals of 100 µg of TDH-His recombinant nickel-purified proteins, emulsified with complete (first injection) or incomplete Freund's adjuvant (Proteogenix S.A.). Antibodies raised against the T. brucei TDH protein expressed in E. coli recognize a single 36.5 kDa protein in western blots, corresponding to the calculated TDH molecular weight (36.96 kDa).
Total protein extracts of bloodstream or procyclic forms of T. brucei (5×106 cells) were separated by SDS PAGE (10%) and immunoblotted on Immobilon-P filters (Millipore) [26]. Immunodetection was performed as described [26], [27] using as primary antibodies, the mouse anti-sera against AceCS diluted 1∶100 [13], PDH-E2 diluted 1∶500 [28] or the heat shock protein 60 (hsp60) diluted 1∶10,000 [29], or the rabbit anti-sera against TDH diluted 1∶500, acetate∶succinate CoA-transferase (ASCT) diluted 1∶100 [9], acetyl-CoA thioesterase (ACH) diluted 1∶500 [10] or glycerol-3-phosphate dehydrogenase (GPDH, EC 1.1.1.8) diluted 1∶100 [30]. Goat anti-rabbit Ig/peroxidase (1∶10,000 dilution) or goat anti-mouse Ig/peroxidase were used as secondary antibody and revelation was performed using the SuperSignal West Pico Chemiluminescent Substrate as described by the manufacturer (Thermo Scientific). Images were acquired and analyzed with a KODAK Image Station 4,000 MM and quantitative analyses were performed with the KODAK MI application.
Cells were washed in PBS and lysed by sonication (5 sec at 4°C) in hypotonic lysis buffer (5 mM Na2HPO4, 0.3 mM KH2PO4). Determination of TDH and PDH enzymatic activities was performed using a spectrophotometric assay as described before [12], [31], [32].
To stain mitochondria of the wild-type cell lines, 200 nM MitoTracker Red CMXRos (Invitrogen) were added to the culture, followed by a 20 min incubation and washes in PBS. Then wild-type cells were fixed with 4% formaldehyde in PBS, permeabilized with 1% Triton X-100, and spread on poly-L-lysine-coated slides. The slides were then incubated for 45 min in PBS containing 5% BSA, followed by incubation in PBS with 2% BSA and the primary antiserum, 1∶50 diluted rabbit anti-TDH, mouse anti-AceCS or mouse anti-PDH-E1α. After washing with PBS, the slides were incubated with 2 µg/mL Alexa 594 anti-rabbit IgG conjugate or Alexa Fluor 594 anti-mouse IgG conjugate (Molecular Probes). Slides were then washed and mounted in the SlowFade antifade reagent (Invitrogen). Cells were visualized with a Leica DM5500B microscope, and images were captured by an ORCA-R2 camera (Hamamatsu) and Leica MM AF Imaging System software (MetaMorph).
The bloodstream forms (2.5×107 cells, ∼0.25 mg of proteins) or procyclic form (5×107 cells, ∼0.25 mg of proteins) of T. brucei were collected by centrifugation at 1,400 g for 10 min, washed once/twice with phosphate-buffered saline (PBS) and incubated for 5 h at 37°C in 2.5 mL of incubation buffer (PBS supplemented with 5 g/L NaHCO3, pH 7.4), with [U-13C]-glucose (4 mM) in the presence or the absence of threonine (4 mM). The same experiments were performed with regular 12C glucose as the only carbon source. The integrity of the cells during the incubation was checked by microscopic observation. 50 µL of maleate (20 mM) were added as internal reference to a 500 µL aliquot of the collected supernatant and proton NMR (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 method. This method provides an electronically synthesized reference signal [33]. Acquisition conditions were as follows: 90° flip angle, 5,000 Hz spectral width, 32 K memory size, and 9.3 sec total recycle time. Measurements were performed with 256 scans for a total time close to 40 min. Before each experiment, the phase of the ERETIC peak was precisely adjusted. Resonances of the obtained spectra were integrated and results were expressed relative to ERETIC peak integration. The linear production of pyruvate and acetate throughout the experiment was confirmed by 1H-NMR quantification of the end-products excreted by the wild type trypanosomes incubated for 6 h in PBS containing 4 mM [U-13C]-glucose (data not shown).
Cells in the late exponential phase (5×107 cells) were incubated for 16 h in 10 mL of modified IMDM medium without threonine, pyruvate, leucine, isoleucine, valine, containing 25 mM glucose, 100 µM acetate and 40 µCi of [1-14C]-acetate (55.3 mCi/mmol). Cells were checked microscopically for viability several times during incubation. Subsequently, lipids were extracted by chloroform∶methanol (2∶1, v/v) for 30 min at room temperature, and then washed three times with 0.9% NaCl. The washed lipid extracts were then evaporated and lipids were dissolved in 1 mL of methanol∶H2SO4 (40∶1, v/v). Trans-esterification of the fatty acids of the lipids was performed at 80°C for 60 min. After cooling the samples, 400 µL of hexane (99% pure) and 1.5 mL of H2O were added, and the mixture was homogenized vigorously for 20 sec. The samples were then centrifuged for 5 min at 1,000 g to separate the phases, and the hexane upper phases containing fatty acid methyl ester (FAMEs) were recovered without contact with the lower phases. FAMEs were loaded onto HPTLC plates developed in hexane/ethylether/acetic acid (90∶15∶2, v/v) and were separated (RF 0.90). They were identified by co-migration with known standards. Their radio-labeling was then determined with a STORM 860 (GE Healthcare). The values were normalized with the amounts of total esters in each sample and detected by densitometry analysis using a TLC scanner 3 (CAMAG, Muttenz, Switzerland) as already described [13].
Eight- to ten-week-old female BALB/c mice bred at the SAS Centre d'Elevage Depré (Saint Doulchard, France) were housed under conventional conditions, with food and water administered ad libitum, according to institutional guidelines. Twelve mice per group were immunocompromised by intraperitoneal injection of 300 mg/kg Endoxan 48 h prior to infection and then infected with a single intraperitoneal injection of 104 parasites suspended in 0.3 mL of fresh IMDM medium. Where appropriate, 1 mg/mL doxycycline and 50 g/L saccharose were added every 48 h to the drinking water starting three days prior to infection. Four experimental groups were studied: animals infected with wild-type parasites without (group 1) or with doxycycline (group 2) in the drinking water, animals infected with the c RNAiPDH.ni cell line (group 3) and animals infected with the RNAiPDH.i cell line cultured for 48 h in the presence of doxycycline to pre-induce down-regulation of PDH-E2 expression and then kept with doxycycline in the drinking water (group 4). To prevent the phenotypic reversion commonly observed in BSF mutants, the injected RNAiPDH.i cell line was selected from a fresh transfection and maintained in vitro up to 4 weeks post-transfection before injecting the animals. Efficient down-regulation of PDH-E2 expression was confirmed by western blot and the threonine-dependency of the selected cell line was confirmed in vitro. The health status of the animals was monitored on a daily basis and parasitaemias were counted daily.
Experiments, maintenance and care of mice complied with guidelines of the European Convention for the Protection of Vertebrate Animals used for Experimental and other Scientific Purposes (CETS n°123). Experiments were approved by the Department for the protection of animals and plants of the Préfecture de la Gironde (Identification number A33-063-324).
PF depend on acetate produced in the mitochondrion to feed fatty acid biosynthesis through the essential enzyme AceCS [13]. A western blot analysis showed that AceCS (74 kDa) was expressed at the same level in the BSF and PF, and an immunofluorescence analysis using the anti-AceCS immune serum showed a homogeneous diffuse pattern characteristic of a cytoplasmic localization (Fig. 2), suggesting that this pathway may also exist in BSF. AceCS is essential for BSF viability, as demonstrated by the death of the RNAiAceCS.i cell line three days post-induction of down-regulation of the AceCS gene expression (.ni and .i stands for uninduced and tetracycline-induced, respectively) (Fig. 3A). To investigate the role of AceCS, radiolabel incorporation into fatty acids from [1-14C]-acetate was measured for the parental and RNAiAceCS cell lines incubated in the IMDM medium. Label incorporation into fatty acids was reduced 2.1- and 8.1-fold one and two days after tetracycline addition, respectively, which correlates with the reduction of AceCS expression (Fig. 3B). Altogether these data demonstrate that proliferative BSF require acetate to feed the essential fatty acid biosynthetic pathway, as previously observed in PF [13].
The IMDM medium does not contain acetate, except the minor contribution of the 10% FCS supplement (∼5 µM) [34]. Consequently, LS-BSF may produce acetate from the catabolic pathways previously identified in PF, i.e. mitochondrial production of acetyl-CoA from glucose and threonine degradation through PDH and TDH, respectively [3], followed by conversion of acetyl-CoA into acetate by two mitochondrial enzymes, ASCT [9] and ACH [10]. A western blot analysis showed that ASCT (54 kDa), ACH (40.5 kDa), PDH-E2 (E2 subunit of PDH, 49.6 kDa) and TDH (39.5 kDa) are expressed in the 427 BSF strain, which has lost the ability to differentiate into SS-BSF (Fig. 2A). This was confirmed by determination of the PDH and TDH activities in LS-BSF, which were 4-fold and 6.7-fold lower than PF, respectively (Fig. 2A). Immunofluorescence analyses revealed colocalization of PDH-E1α (E1α subunit of PDH) and TDH with the mitochondrion-specific dye MitoTracker Red CMXRos (Invitrogen) (Fig. 2B). The mitochondrial localization of TDH is consistent with a 24-amino-acid N-terminal mitochondrial targeting signal predicted by MitoProt (http://ihg.gsf.de/ihg/mitoprot.html) with a high probability (0.82). Since BSF express the whole set of enzymes required for acetate production, we then used a combination of reverse genetics on PDH-E2 and TDH and metabolic profiling by NMR to investigate mitochondrial acetate production in BSF.
It is widely considered that LS-BSF excrete only pyruvate from glucose metabolism, while PF mainly produce acetate and succinate. To compare glucose metabolism in these two forms, 2.5×107 LS-BSF and 5×107 PF (equivalent to 0.25 mg of proteins) were incubated in 2.5 mL of PBS containing 4 mM glucose (Fig. 4) or [U-13C]-glucose (Fig. 5A and Table 1). 13C-enriched end-products excreted in the medium from [U-13C]-glucose metabolism were quantified by 1H-NMR (Table 1). It is to note that quantification errors are significant, in particular for molecules representing less than 5% of all excreted end-products. As expected, BSF mainly converted glucose into pyruvate (7761 nmol/h/108 cells), which accounts for 85.1% of the excreted end-products. In addition, BSF excreted significant amounts of alanine, acetate and succinate, which represent 9.2%, 4.9% and 0.8% of the excreted end-products from glucose metabolism, respectively (Table 1). Surprisingly, the rate of excretion of 13C-enriched acetate and succinate from [U-13C]-glucose was only 2-fold lower in BSF than in PF (446 versus 789 nmol of acetate/h/108 cells and 71 versus 156 nmol of succinate/h/108 cells, respectively) (Table 1, see Fig. 4). The unexpected similar rate of acetate and succinate excretion in both trypanosome forms is probably due to the ∼10-fold higher glycolytic rate in BSF (the rate of glycolytic end-product excretion was 9.4-fold higher in BSF compared to PF - Table 1). Consequently, the high glycolytic rate in BSF combined with the dominant conversion of glucose into pyruvate (85.1% of the excreted end-products) may have led to underestimation of the role of acetate and succinate production in LS-BSF, although their rate of production were in the same range in PF.
To confirm acetate production from glucose metabolism by LS-BSF, we conducted RNAi-mediated down-regulation of expression of the PDH-E2 gene. The RNAiPDH.i cell line showed no growth phenotype upon tetracycline induction (Fig. 6A), although the PDH-E2 protein was no longer detectable by western blot two days post-induction (Fig. 6A, inset). Metabolite profiling of the RNAiPDH.i cell line incubated in the presence of 4 mM of [U-13C]-glucose showed a 13.7-fold reduction of acetate production from glucose compared to the RNAiPDH.ni cells (55 versus 755 nmol/h/108 cells) (Table 1). It is to note that, for unknown reasons, the uninduced RNAiPDH.ni cell line produces ∼1.7-times more acetate from glucose than the parental cells.
Since both BSF and PF have been reported to produce acetate from threonine [12], which is present in the IMDM medium (0.9 mM), we investigated the threonine degradation pathway in BSF. Incubation of the parasites in threonine-depleted medium, which contains only ∼15 µM of the amino acid coming from FCS [35], did not affect growth of the wild-type and RNAiPDH.ni cells, while growth of the RNAiPDH.i mutant was abolished (Fig. 6D). To confirm that glucose and threonine degradations contribute to acetate production, both pathways were interrupted by down-regulating PDH-E2 expression in the TDH null background (Δtdh/RNAiPDH cell line). First, both TDH alleles were replaced by the puromycin (PAC) and blasticidin (BSD) markers in the Δtdh cell line, with no effect on growth rate (Fig. 6C). Deletion of both TDH alleles was confirmed by PCR analyses (Fig. 6B), western blot analyses and enzymatic assays (insets of Fig. 6C). Second, RNAi-mediated down-regulation of PDH-E2 was performed in the Δtdh background. Growth of the Δtdh/RNAiPDH.i cell lines was abolished three days post-induction before cell death seven days later (Fig. 6E). Addition of 4 mM acetate in the medium does not rescue growth of the Δtdh/RNAiPDH.i mutant (data not shown) suggesting that acetate and/or acetyl-coA need to be produced inside the mitochondrion to feed the essential mitochondrial fatty acid pathway, may be through the production of the precursor butyryl-CoA [36]. This result confirms that abolition of mitochondrial acetyl-CoA/acetate production from both glucose and threonine is lethal for BSF grown in standard medium.
To address this question we developed a metabolite profiling assay based on the ability of 1H-NMR spectrometry to distinguish 13C-enriched molecules from 12C ones. Cells were incubated in PBS with equal amounts (4 mM) of [U-13C]-glucose and unenriched threonine in order to perform a quantitative analysis of threonine-derived and glucose-derived acetate production by 1H-NMR. When [U-13C]-glucose was the only carbon source in the incubation medium, the excreted [13C]-acetate (annotated A13 in Fig. 5) was represented by two doublets with chemical shifts at around 2.0 ppm and 1.75 ppm, respectively (see Fig. 5A). It is to be noted that threonine metabolism cannot be analyzed independently since glucose is essential for BSF. Addition of threonine to the [U-13C]-glucose/PBS medium induced production of threonine-derived [12C]-acetate (386 nmol/h/108 cells) in addition to [13C]-glucose-derived [13C]-acetate (532 nmol/h/108 cells) (Fig. 5B and Table 2). This shows that in the presence of equal amounts of both carbon sources, glucose contributes ∼1.4-fold more than threonine to acetate production.
1H-NMR metabolite profiling of the single and double mutants confirmed the involvement of both glucose and threonine in acetate production. As expected, production of [13C]-glucose-derived [13C]-acetate was ∼50-times lower in the RNAiPDH.i than in the RNAiPDH.ni cells (16 versus 847 nmol/h/108 cells), while threonine-derived acetate production was not affected. Conversely, production of threonine-derived acetate was abolished in the Δtdh mutant, while [13C]-glucose-derived [13C]-acetate was not affected (Table 2 and Fig. 5C–D). Finally, production of acetate from both carbon sources was affected in the Δtdh/RNAiPDH.i double mutant cell line (Table 2 and Fig. 5E).
BALB/c mice immunocompromised by Endoxan treatment were injected with wild-type and RNAiPDH cells and kept with or without doxycycline, a stable tetracycline analog, in the drinking water to down-regulate expression of PDH-E2. Animal survival and the blood parasite levels were monitored. No differences were observed between the four groups of animals, in which parasite density started to rise at day three post-infection. All mice were dead at days 6–7 post-infection (data not shown). This shows that acetate production from glucose is not necessary for the viability of T. brucei in vivo, suggesting that a possible acetate source (threonine) that is present in the blood is absent in the threonine-depleted in vitro culture medium. As mentioned above, mammalian blood contains approximately 150 µM threonine [35], [37], which is 10-times higher than in the threonine-depleted IMDM medium. The RNAiPDH.i cell line died in IMDM medium containing 15, 37.5 and 75 µM threonine, while addition of 150 µM of the amino acid restored its growth in vitro (Fig. 6D), suggesting that the homeostatic threonine blood concentration (150 µM) is sufficient to provide BSF with the required acetyl-CoA/acetate molecules. Altogether, this demonstrates that BSF trypanosomes have developed two complementary and self-sufficient ways to maintain the essential production of acetate in the blood of mammalian hosts.
LS-BSF trypanosomes are well known for their glucose-dependency to satisfy ATP requirements [38]. Indeed, net production of all cellular ATP is fulfilled by the last glycolytic step catalyzed by pyruvate kinase, which produces pyruvate, the excreted glycolytic end-product. Excretion of significant amounts of other partially oxidized end-products of glycolysis, such as glycerol, succinate and alanine, has been previously reported [14], [17], [39], [40]. However, in the late seventies emerged a general dogma whereby pyruvate was considered the exclusive glycolytic end-product excreted from LS-BSF under aerobic conditions [15], [16], because the minor end-products were assigned to either non-growing conditions or contamination with non-dividing SS-BSF [18], [19]. Here, we used as an experimental model the 427 BSF strain, which has lost the ability to differentiate into SS-BSF, in order to focus our analysis of glucose metabolism on LS-BSF trypanosomes. This laboratory-adapted monomorphic strain is insensitive to the stumpy inductor factor, but, it successfully differentiates in vitro into bona fide SS-BSF, for instance when expression of the protein kinase target of rapamycin (TOR4) is inhibited [41]. This suggests that the 427 strain can be considered as a slender-like BSF that has lost the ability to respond to the stumpy inductor factor, and as such is the relevant model to study the metabolism of proliferative BSF. Our metabolic analyses showed that LS-BSF can produce almost as much succinate and acetate from glucose as PF incubated in the same conditions. This suggests that most, if not all, enzymes involved in the “succinate and acetate branches” previously characterized in PF are also expressed in LS-BSF. To produce acetate, PDH (step 1 in Fig. 1) converts pyruvate into acetyl-CoA, which is the substrate of ASCT (step 2) and ACH (step 3) for acetate production. ASCT expression is low in BSF (Fig. 2 and [19]), while ACH is relatively abundant (Fig. 2) with an ACH activity ∼2-fold higher than PF (data not shown). Three of the four PDH subunits have been investigated so far and are expressed in BSF (PDH-E1α and PDH-E2, see Fig. 2; PDH-E3, [42]), with a PDH enzymatic activity only 4-fold lower than in PF (Fig. 2). This relatively high PDH activity is in agreement with a recent comparative SILAC proteomics analysis showing that PDH-E1α, PDH-E1ß, PDH-E2 and PDH-E3 are 5.3-, 7.6-, 5.2- and 8.1-fold more abundant in PF than LS-BSF, respectively [43]. The same proteomics analysis in LS-BSF also detected most, if not all, of the enzymes involved in succinate production from phosphoenolpyruvate, although at a lower level of expression than in PF (between 3- and 20-fold). Altogether, this clearly demonstrates that LS-BSF have maintained the capacity to produce and excrete acetate and succinate from glycolysis. The relatively high rate of acetate and succinate production (only ∼2-fold higher in PF), while the enzymes involved in the corresponding metabolic pathways are 5- to 20-times more abundant in PF, may be due to the considerably higher glycolytic flux in BSF. We determined that the excretion rate of glycolytic end-products is 9.5-fold higher in BSF than in PF (9115 versus 968 nmol of excreted end-products/h/108 cells), which is consistent with the previously measured 5- to 10-fold difference in glycolytic flux [14]. A recent analysis of the glycolytic flux in the same BSF strain incubated in growing conditions (IMDM containing 20 mM glucose) showed a higher rate of pyruvate production compared to our analysis performed in PBS containing 4 mM glucose and threonine (19.2 versus 12.0 µmol/h/108 cells) [44]. The reduced glycolytic flux observed in PBS conditions certainly reflects the difference between non-growing conditions (PBS) and exponential growth (IMDM) with a doubling time in the range of 5 h [44]. Also, the 35% reduction in total end-product fluxes for the wild type cells depending on the available substrates (PBS/glucose versus PBS/glucose/threonine) shows that metabolic fluxes are dependent on the exact context of substrates present (Table 1).
It is important to note that our experimental procedures do not reflect physiological conditions, since trypanosomes were incubated at high density in PBS containing 4 mM glucose. Consequently, these minor glycolytic end-products might be excreted at a lower rate, or not at all, by LS-BSF trypanosomes in vivo. A recent quantitative analysis of the fate of glucose in exponentially growing 427 LS-BSF in vitro (the same strain analysis here) showed that pyruvate is the only excreted glycolytic end-product [44]. Glucose, pyruvate and glycerol were analysed in that study. Although they report an almost complete carbon balance between glucose uptake and pyruvate excretion, their analysis leaves room for small fluxes towards products they did not analyse such as acetate and succinate. We here report these end-products, with fluxes to acetate and succinate together representing ∼5% of the excreted glycolytic end products. The exact fraction of total carbon going to these end-products is difficult to assign, due to the errors in quantification of fluxes and because our results did not enable the calculation of a carbon balance between carbon uptake and excretion. Whatever the rate of acetate excretion from glucose metabolism in exponentially growing LS-BSF is, its production is essential for growth, as exemplified by the death of the RNAiPDH.i cell line incubated in the absence of threonine, the other acetate source. This was confirmed by inducing cell death upon blocking acetate production from both carbon sources in the Δtdh/RNAiPDH.i double mutant, while growth of the corresponding single mutants in standard IMDM medium was not affected (Fig. 6). The relevance of acetate production from glycolysis for LS-BSF is further strengthened by (i) the same high rate of growth of the wild-type parasite, even in the absence of threonine (Fig. 6), as recently observed by the development of a new minimal medium that supports growth of BSF [37] and (ii) the impossibility to rescue growth of the Δtdh/RNAiPDH.i double mutant by addition of sodium acetate in the medium, which cannot substitutes glucose-derived acetate production.
The above-mentioned reverse genetic experiments combined with 1H-NMR metabolic analyses also clearly demonstrate that glucose and threonine are the only significant carbon sources contributing to the essential production of acetate in the 427 BSF strain. To our knowledge, this is the first report showing acetate production from glucose in LS-BSF, while threonine has been described before as an acetate-source in BSF [12], [45]. Inhibition of a single acetate-production pathway in the RNAiPDH.i and Δtdh cell lines grown in standard IMDM medium does not affect growth of LS-BSF, indicating that a single active pathway is sufficient for growth in vitro. This is probably also true in vivo, since (i) glucose concentration is higher in the blood than in our experimental conditions (5 versus 4 mM) and (ii) the relatively low homeostatic concentration of threonine in mammalian blood (150 µM) is sufficient for acetate production, as demonstrated by the absence of growth effect of the RNAiPDH.i mutant in medium containing at least 150 µM threonine (Fig. 6D). When incubated with equal amounts of threonine and glucose (4 mM), the parasite produces ∼1.4-fold more acetate from glycolysis. Mammalian blood contains ∼30-fold more glucose than threonine (5 mM versus 150 µM), which strengthens the view that the contribution of glucose to acetate production is relevant in vivo. This contrasts with an equivalent recent analysis performed on the procyclic form of T. brucei, which prefers threonine for acetate and fatty acid productions, with a ∼2.5-fold higher contribution of threonine compared to glucose when incubated with 4 mM of both carbon sources [46].
In PF trypanosomes, the mitochondrial production of acetate is essential to feed de novo fatty acid biosynthesis through the “acetate shuttle” [13]. In this shuttle, acetate produced in the mitochondrion reaches the cytosol, where a part of it is converted by AceCS into acetyl-CoA to produce malonyl-CoA, the elongator for fatty acid biosynthesis. It is noteworthy that both the microsomal elongase-dependent and mitochondrial type II fatty acid synthase pathways use malonyl-CoA to elongate fatty acids [47]. As observed in PF, AceCS is essential for incorporation of [1-14C]-acetate into LS-BSF fatty acids (Fig. 3), indicating that AceCS is required for de novo fatty acid synthesis. In addition, Gilbert et al. previously showed that both glucose and threonine are used as carbon sources for fatty acid synthesis [47]. Altogether, this demonstrates that, like PF, proliferative BSF trypanosomes use the “acetate shuttle” to feed fatty acid biosynthetic pathways. The essential role of mitochondrial fatty acid synthesis has been documented in BSF, however, RNAi-mediated down regulation of elongase genes involved in the microsomal fatty acid biosynthesis does not affect growth of the parasite [36], [48], [49], [50]. Consequently, the lethal phenotype observed for the LS-BSF RNAiAceCS mutant is probably due to the dramatic reduction of cytosolic malonyl-CoA production required to feed mitochondrial fatty acid biosynthesis, which contribute to ∼10% of cellular fatty acid production [36].
As mentioned above, our detection of several “minor” glycolytic end-products excreted by BSF trypanosomes is consistent with most, if not all, early reports [14], [17], [39], [40] and the recent quantitative flux analysis in BSF 427 [44]. The high glycolytic flux combined with the almost exclusive conversion of glucose into pyruvate has probably caused the community to overlook these minor end-products and the metabolic pathways leading to their production, although their biosynthesis may be essential for the parasite, as described here for acetate. For instance, we also observed that LS-BSF excretes alanine from glucose (twice more than acetate, Table 1), as reported before [17]. Alanine is produced from pyruvate by transfer of an amino group coming from various possible amino acid sources. A recent metabolomic analysis showed that glutamate and hydrophobic keto acids accumulate in the BSF spent media, suggesting that glutamine and hydrophobic amino acids are possible substrates of alanine aminotransferase for alanine production [37]. The relevance of alanine production from glycolysis is strengthened by the reported accumulation of hydrophobic keto acids in the plasma and urine of infected rodents [51], [52] and the requirement of alanine aminotransferase activity for in vitro growth of the parasite [53]. Succinate production from glycolysis may also be of importance for biosynthetic pathways in LS-BSF, as exemplified by the requirement of fumarate for de novo synthesis of pyrimidine through the unusual fumarate-dependent dihydroorotate dehydrogenase [54]. Altogether, this highlights the need to reconsider and further investigate the metabolic pathways leading to minor glycolytic end-products, which are still considered negligible compared to the total carbon flux from glucose [44], in order to reveal new essential metabolic pathways that could be targeted to develop new trypanocidal molecules. Beyond glycolysis, other overlooked metabolic pathways of the central metabolism need to be revisited in LS-BSF, as exemplified by the recent observation that RNAi down-regulation of the tricarboxylic acid enzyme succinyl-CoA synthetase induces one of the most spectacular death phenotypes observed in BSF, with 100% cell death within less than 20 h post-induction [55], while this pathway is considered repressed in BSF.
|
10.1371/journal.pgen.1000769 | Interaction of HP1 and Brg1/Brm with the Globular Domain of Histone H3 Is Required for HP1-Mediated Repression | The heterochromatin-enriched HP1 proteins play a critical role in regulation of transcription. These proteins contain two related domains known as the chromo- and the chromoshadow-domain. The chromo-domain binds histone H3 tails methylated on lysine 9. However, in vivo and in vitro experiments have shown that the affinity of HP1 proteins to native methylated chromatin is relatively poor and that the opening of chromatin occurring during DNA replication facilitates their binding to nucleosomes. These observations prompted us to investigate whether HP1 proteins have additional histone binding activities, envisioning also affinity for regions potentially occluded by the nucleosome structure. We find that the chromoshadow-domain interacts with histone H3 in a region located partially inside the nucleosomal barrel at the entry/exit point of the nucleosome. Interestingly, this region is also contacted by the catalytic subunits of the human SWI/SNF complex. In vitro, efficient SWI/SNF remodeling requires this contact and is inhibited in the presence of HP1 proteins. The antagonism between SWI/SNF and HP1 proteins is also observed in vivo on a series of interferon-regulated genes. Finally, we show that SWI/SNF activity favors loading of HP1 proteins to chromatin both in vivo and in vitro. Altogether, our data suggest that HP1 chromoshadow-domains can benefit from the opening of nucleosomal structures to bind chromatin and that HP1 proteins use this property to detect and arrest unwanted chromatin remodeling.
| HP1 proteins are transcriptional regulators frequently associated with gene silencing, a phenomenon involving masking of promoter DNA by dense chromatin. Owing to their chromo-domain, these proteins can read and bind an epigenetic mark that on many non-expressed genes is present on histone H3 at the surface of the nucleosome (the fundamental packing unit of chromatin). However, the binding to this mark does not explain the repressing activity of HP1 proteins. Here, we show that these proteins can establish a second contact with histone H3, independently of the epigenetic mark. This second contact site is located inside the nucleosome, in a position likely to be inaccessible. Interestingly, this site is also contacted by a subunit of the SWI/SNF complex and this contact is required for the ATP-dependent chromatin remodeling catalyzed by SWI/SNF. We provide evidence suggesting that HP1 proteins use the SWI/SNF chromatin remodeling to gain access to the contact site inside the nucleosome and to prevent further remodeling by competing with SWI/SNF for binding at this position. These observations lead us to suggest that HP1 proteins function as gatekeepers on promoters, detecting and stopping unwanted exposure of internal nucleosomal sites.
| HP1 proteins are important regulators of heterochromatin-mediated silencing and chromosome structure in diverse eukaryotes (for recent reviews, see [1],[2]). In mammalian cells, the HP1 family is composed of HP1α, HP1β, and HP1γ. So far, only HP1β has been inactivated in the mouse, resulting in defective development of neuromuscular junctions and cerebral cortex [3]. Within the nucleus, the three HP1 isoforms all concentrate in foci of dense pericentromeric heterochromatin but are also present in the rest of the nucleoplasm. Consistent with this very general distribution, the mammalian HP1 proteins are detected not only in dense heterochromatic regions but also on active euchromatic genes [4].
On these active genes, HP1 proteins seem to be present both during phases of silencing and transcriptional activity. For example, on the Survivin gene and on an integrated HIV1 LTR, HP1β is detected on the repressed promoter, while HP1γ is recruited after transcriptional activation [5],[6]. HP1γ is however not always associated with active transcription as it participates in the repression of the MMTV LTR and the Sox2 promoter [7],[8]. It is noteworthy also that on the HIV1 LTR and on the β-major gene, HP1 proteins are co-localized with the RNA polymerase II (RNAPII), indicating that they are not creating an environment incompatible with recruitment of this polymerase [5],[9]. Taken together, these observations suggest that, at least in euchromatin, HP1 proteins are not “chromatin condensers” per se, but more likely regulators of enzymatic activities involved in transcription initiation or elongation.
HP1 proteins contain two very similar domains known as the chromo-domain (CD) and the chromoshadow-domain (CSD) separated by a less structured hinge region. The CSD is required for dimerization and interaction with many molecular partners that share a PXVXL motif [10]. It is also necessary for the recruitment of HP1 proteins to sites of DNA damage [11]. Concurrently, the CD recognizes and binds histone H3 tails methylated on lysine 9 (K9), an epigenetic mark frequently associated with transcriptional repression [12],[13]. In addition, the hinge region of HP1 proteins harbors DNA- and RNA-binding activities and the targeting of these proteins to chromatin likely results from the integration of multiple contacts [14],[15].
In contrast to the strong binding to peptides mimicking histone H3 tails methylated on K9, HP1 proteins bind only weakly to reconstituted methylated nucleosomal arrays [16]. Consistent with this, binding of HP1 proteins to purified native chromatin in vitro seems relatively inefficient [15],[17]. This binding can be improved by auxiliary factors that may help the recognition of chromatin [16], but it has also been suggested that HP1 can benefit from chromatin opening. Indeed, a more stable incorporation of HP1 proteins occurs in S phase when DNA replication disrupts the histone octamers [17]. Earlier reports also describe the presence in the nucleus of two populations of HP1 proteins with either high or low mobility [18] and it has been proposed that the more stable interaction creates the HP1 population of low mobility [3]. Binding of HP1 proteins may also benefit from ATP-dependent chromatin remodeling as HP1β co-localize with the ACF1-ISWI remodeling complex [19]. In addition, HP1α, but not HP1β and HP1γ, interacts with Brg1 and Brm, the mutually exclusive catalytic subunit of the human SWI/SNF (hSWI/SNF) complex, and this interaction favors repression of a reporter construct by a transfected Gal4-HP1α fusion protein (Figure S1A, S1B, S1C, S1D and [20],[21]).
To gain better understanding of HP1 chromatin binding and transcriptional regulation, we have here examined whether these proteins could establish alternative interactions with the histones. This allowed us to identify a contact between the CSD and a region of histone H3 located at the border of the globular domain. This region is also contacted by the hSWI/SNF subunits Brg1 and Brm, and we show that HP1 proteins have a negative effect on hSWI/SNF-mediated chromatin remodeling. Finally, we provide evidence indicating that hSWI/SNF activity is involved in the recruitment of HP1 proteins to chromatin.
We investigated whether HP1 proteins could bind histone H3 independently of the well-characterized association of the CD with methylated K9. To this end, we tested the binding of HP1α and HP1γ to either purified or recombinant B10-epitope-tagged histones immobilized on nitrocellulose membrane. As expected, the HP1 proteins bound strongly to purified histone H3 but not to histone H4 (Figure 1A, lanes 3 and 4). Interestingly, we also observed weaker but significant binding to full-length recombinant histone H3 produced in E. coli and therefore not methylated on K9 (Figure 1A, lane 1). This binding was not observed on the tail region alone (Figure 1A, lane 2). This is in accordance with earlier studies showing interaction of HP1 proteins with the globular domain of recombinant histone H3 [17],[22]. In GST pull down assays, we also observed weaker, but persisting histone H3 binding after mutation of the CD at position V22, abolishing interaction of HP1α with the methylated histone H3 tail (Figure 1B, lane 2). This again suggested the presence of additional contact points between HP1α and histone H3.
The structure of the CSD is very similar to that of the CD (Figure 1C), prompting us to probe for an interaction with the histones via this domain. To this end, we further mutated HP1α V22M at position I126 inside the CSD. This position is equivalent to I25 in the CD, an amino acid that, when mutated, prevents the domain from interacting with histone H3 [22]. This position was chosen because V22 has no equivalent in the CSD. The double mutant no longer interacted with H3, indicating that in both the CD and the CSD, the first β strand is involved in histone interaction (Figure 1B, lane 3). Mutation of the CSD at I126 also affected the repressing activity of HP1α. This was visualized by co-transfecting in MCF7 cells an MMTV/Gal4 reporter construct and expression plasmids for Gal4-HP1α fusions where HP1α was either WT or with an I126F mutation (Figure 1D and 1E).
We next mapped the region of histone H3 involved in the interaction with the HP1α CSD. A non-modified histone H3 truncation mutant spanning from aa 1 to 76 produced in E.coli was sufficient to interact with HP1α, while a shorter construct containing only the H3 tail region (aa 1–47) failed to do so. The interaction was disrupted by deletion of the CSD, confirming its implication in the contact with H3 (Figure 1F and 1G, 3 top panels). We noted also that all mutations affecting HP1 dimerization abolished the CSD-H3 interaction, while this interaction resisted incubation with a DNA intercalating agent (data not shown). This series of experiments also showed that HP1γ and Drosophila dHP1a had binding activities similar to that of HP1α (Figure 1G, bottom panels).
We finally identified aa 35 to 66 as the minimal region binding both HP1α and HP1γ (Figure 1H, lane 2 and 4). We termed this region the Shadock for “chromoShadow docking”. This region contains a PXXVXL motif resembling the PXVXL motif frequently found in proteins interacting with the CSD of HP1 proteins [10]. Mutation of the valine in this sequence (V46) abolished binding to HP1α but not HP1γ (Figure 1H, lane 5). Consistent with this, the Shadock could be further shortened to aa 44 to 66 without disrupting binding of HP1γ (Figure 1H, lane 3). These observations show that the two proteins have overlapping but not identical binding sites.
The Shadock region is located at the entry/exit site of the nucleosome but is partially hidden inside the nucleosomal barrel (Figure 2A). Interestingly, this region also includes the H3 αN helix previously shown to play an important role in nucleosome mobility [23] and mutations in this region were recently shown to affect chromatin remodeling by yeast SWI/SNF [24]. Besides, we found that histone H3 was co-immunoprecipitated with the hSWI/SNF catalytic subunit Brg1 in an in vitro assay (Figure 2C). We therefore investigated whether Brg1 would target this region during remodeling. In these experiments, we used a truncation mutant of the Brg1 protein centered on the ATPase domain (ΔBrg1-1, Figure 2B). This mutant, sufficiently short to be produced in E. coli, shows remodeling activity similar to full length Brg1 [25]. When expressed as a GST fusion, ΔBrg1-1 had affinity for histone H3 but not for other histones bound to nitrocellulose membrane (Figure 2D, lane 5). GST pull down assays further showed that ΔBrg1-1 bound to both purified and recombinant histone H3 (Figure 2E, lanes 2 and 3). Additional Brg1 deletion mutants showed that regions C-terminal of the helicase domain could mediate the interaction (Figure 2F). These regions were previously reported as essential for in vitro remodeling [25]. Binding properties of Brg1 to histone H3 could essentially be recapitulated with Brm, the alternative catalytic subunit of the hSWI/SNF complex (Figure S1E, S1F, S1G, S1H). H3 deletion mutants further revealed that the interaction of Brg1 with the histone was dependent on the region from aa 35 to 66 also involved in interaction with the HP1 proteins (Figure 2G, lane 2). Best binding was however achieved when this region was extended by 10 aa (aa 25 to 66, Figure 2G, lane 1).
To determine whether this interaction was important for chromatin remodeling, we used Restriction Enzyme Accessibility (REA) assays [26]. Nucleosomal arrays were assembled by the use of DNA templates consisting of two sets of five 5S nucleosome positioning sequences that flank DNA sufficient to assemble two nucleosomes, one of which overlaps a unique HhaI site. Accessibility of this site is increased in the presence of full length Brg1 and ATP, reflecting chromatin remodeling (in Figure 2H, lane 1 compare top and bottom panels). In these assays, we challenged the remodeling by Brg1 with H3 deletion mutants, reasoning that these polypeptides could interfere with the binding of Brg1 with its normal nucleosomal substrate. We observed a good correlation between the ability of the H3 mutants to inhibit the remodeling reaction and their ability to bind Brg1 (see the effect of 200nM and 400nM competing protein in Figure 2H and 2I, and compare with binding in Figure 2G). These observations show that the contact between Brg1 and H3 is important for remodeling.
The binding of both HP1 and Brg1/Brm proteins to the Shadock domain of histone H3 suggested a competition between these two sets of transcriptional regulators. To investigate this possibility, we tested the effect of HP1α on chromatin remodeling by the hSWI/SNF complex in REA assays. Addition of recombinant flag-tagged HP1α (F-HP1α) prevented the hSWI/SNF-dependent increase of HhaI accessibility without modifying the intrinsic accessibility of the restriction site (Figure 3A). Quantification showed that 50% inhibition of site accessibility was reached when nucleosomes, F-HP1α and hSWI/SNF were present at a molar ratio of approx. 1∶70∶10. In a similar assay F-HP1α also repressed remodeling by Brg1 and the truncated ΔBrg1 (Figure S2B).
We next analyzed the effect of F-HP1α on remodeling of a mononucleosome substrate. The mononucleosome was assembled on a 202 bp template containing a unique PstI site only 5% accessible in the absence of hSWI/SNF complex and ATP. As observed with the polynucleosomal template, F-HP1α inhibited remodeling of this substrate (Figure 3B). Repression was moderately less efficient as 50% inhibition was reached at a nucleosomes∶F-HP1α∶hSWI/SNF ratio of 1∶100∶10. Flag-tagged F-HP1γ also repressed hSWI/SNF remodeling with an efficiency similar to that of F-HP1α (Figure 3C). Measuring the kinetics of the repression revealed however that HP1γ was significantly slower than HP1α (Figure S2C).
We finally used the REA assay to test several HA-tagged HP1α constructs (Figure 3D). Consistent with a role for the CSD in the repression, truncation of the carboxyl terminal region abolished the repressing effect of HP1α on remodeling by hSWI/SNF. In contrast, mutants defective in either histone H3 K9me binding (HP1α V22M), DNA/RNA binding (ΔDNA), or interaction with Brm/Brg1 (HP1α δBrg1) were not affected in their ability to repress hSWI/SNF chromatin remodeling (Figure 3E). In addition, we observed HP1α-mediated repression on a nucleosomal array reconstituted with non-modified histones produced in E. coli (data not shown), confirming that binding of the CD to H3 K9me was not required for inhibition of hSWI/SNF activity.
We next set up to identify genes where the antagonism between Brg1/Brm and HP1 proteins could be visualized in vivo. Our attention was brought to interferon-regulated genes that are well-characterized hSWI/SNF targets and that, like the viral HIV1 LTR, are highly and rapidly inducible in response to outside stimuli [27]–[29]. To confirm the effect of hSWI/SNF on the transcription of these genes, we knocked down Brm in HeLa cells with two different siRNAs (Figure 4A, lanes 1–3). Brm was here preferred over Brg1 as Brm is degraded during each mitosis and is therefore very efficiently depleted with siRNAs [30]. This depletion resulted in repression of the interferon-inducible genes we tested, including IFIT1, IFIT3, OASL, and OAS1, with the exception of IFIT5 (Figure 4B). In contrast, these genes, again with the exception of IFIT5, were activated upon knock-down of either HP1α or HP1β (Figure 4A, lanes 4–7, and Figure 4C). These data show that several interferon-regulated genes rely on hSWI/SNF for their activation and on HP1α and HP1β for their repression.
Knock-down of HP1γ did not significantly affect the expression of the interferon-inducible genes in the absence of stimulation (Figure 4A, lanes 8–9 and 4C). However, we noted a moderate but reproducible decrease of the interferon-inducibility of the genes after depletion of HP1γ, again with the exception of IFIT5 (Figure 4D). These data are consistent with earlier reports showing an implication of HP1γ in efficient expression of some genes [5],[6].
Finally, we note that the activation of the interferon-inducible genes was observed only on 4 out of 5 tested genes, with HP1α and HP1β, but not with HP1γ or GAPDH siRNAs. In addition, our siRNAs were designed to minimize the interferon response [31]. We therefore ruled out a possible non-specific stimulation of the interferon pathway by the double-stranded siRNAs.
Earlier studies have shown that HP1 proteins bind poorly to chromatin under physiological salt conditions while they associate tightly with destabilized nucleosomes from cells in S-phase [15],[17]. This would be consistent with nucleosomal structures preventing HP1 proteins to access either the DNA or the Shadock region inside the nucleosome barrel. It also suggests a possible effect of chromatin remodeling on the loading of HP1 proteins to chromatin.
To investigate whether hSWI/SNF activity could influence HP1 recruitment, we carried out chromatin immuno-precipitations (ChIP) on the IFIT3 promoter before and after knock down of Brm with siRNAs. To minimize the impact of the histone H3K9 methylation repression mark on recruitment of HP1 proteins, we followed the promoter during transcriptional activation and we concentrated our study on HP1γ that is not associated with repression of basal IFIT3 transcriptional activity (Figure 4C). The experiments were carried out with a HeLa-derived cell line stably expressing moderate levels of epitope-tagged HP1γ, thus allowing us to detect the protein with both anti-HP1γ and anti-HA tag antibodies (Figure 5A). Upon stimulation with interferon α, recruitment of both HP1γ and Brm increased (Figure 5B, siGAPDH). This is consistent with an implication of HP1γ in efficient expression of the IFIT3 gene as observed in Figure 4D. The recruitment of HP1γ was essentially abolished upon depletion of Brm (Figure 5B, siBrm). This decreased recruitment was not due to the silencing of the IFIT3 promoter, as Brm depletion reduces the transcriptional activity of the gene only approx. 2-fold (Figure 4D). ChIP-reChIP experiments further showed that Brm and HP1γ were present on the same chromatin fragments, suggesting that their recruitment is interdependent. This co-recruitment could be visualized with anti-Brm followed by anti-HA ChIP-reChIP as well as the inverse combination (Figure 5C).
We next investigated the timing of the recruitment of HP1γ and Brg1 on an integrated HIV1 LTR (Figure 5D). As mentioned in the introduction, transcriptional activation of this promoter results in eviction of HP1ß and increased recruitment of HP1γ [5]. ChIP analysis showed that recruitment of RNAPII, Brg1 and HP1γ peaks 60 min. after induction with PMA. Subsequently, low levels of Brg1 recruitment were restored while levels of RNAPII and HP1γ remained high. These observations are therefore compatible with transient recruitment of Brg1 helping the loading of HP1γ onto sites internal to the nucleosome. They also suggest that recruitment of HP1γ may limit the duration of hSWI/SNF-mediated chromatin remodeling during transcriptional activation.
To investigate in vitro whether chromatin opening could facilitate binding of HP1 proteins to nucleosomes, we finally used a recombinant nucleosomal array associated with streptavidin beads (Figure 6A and 6B). Consistent with earlier studies, this array assembled at relatively low ionic strength was poorly bound by Drosophila dHP1a (Figure 6C, lane 2 and [16],[32]). The Drosophila protein was here preferred because it could be more efficiently purified than its human counterparts and showed limited direct binding to Brg1 (Figure S3). The binding of this protein was significantly increased in the presence of full length recombinant purified Brg1 and ATP (Figure 6C, lanes 3 and 4). The effect was inhibited in the presence of γS-ATP, showing that it was dependent on the remodeling activity of Brg1 (Figure 6C, lane 5). Some binding was also observed in the presence of high levels of Brg1 and non-hydrolysable γATP (Figure 6C, lane 6), possibly explained either by a contamination of our preparation of Brg1 with ATP or by the interaction of dHP1a with Brg1.
We show here that, in addition to the contact of the CD with methylated H3K9, HP1 proteins use the CSD to associate with histone H3 at the level of the entry/exit point of the nucleosome barrel. This region on histone H3 that we termed Shadock is also contacted by Brg1 and Brm, the catalytic subunits of the hSWI/SNF complex, and chromatin remodeling can favor loading of HP1 proteins onto chromatin. We provide also evidence suggesting that this loading ultimately decreases the efficiency of hSWI/SNF remodeling.
FRAP experiments have shown that the CSD domain is required for the binding of HP1 proteins to native chromatin in vivo, with actually a stronger contribution than the CD when observation are made in euchromatic regions [33]. In addition, deletion of the CSD abolishes HP1-mediated transcriptional repression in transfection assays [34]. In the same type of experiment, we show that a point mutation in the CSD disrupting the interaction with the H3 Shadock region (I126F) is sufficient to interfere with the repressing activity. Furthermore, our REA assays show that the CSD is required for inhibition of chromatin remodeling by hSWI/SNF and that this remodeling complex contacts the same region on H3 as the HP1 proteins. Taken together, these observations strongly suggest that HP1 chromatin binding and repression activities are largely mediated by the histone binding activity of the CSD.
ChIP and ChIP reChIP assays show that Brm is required for the loading of HP1γ to the IFIT3 promoter during transcriptional activation. We note that on this promoter, we have another example of the switch from HP1α/HP1ß to HP1γ during transcriptional activation, also observed on the HIV1 LTR and on the Survivin promoter [5],[6]. This switch could suggest that, in vivo, histone H3K9 methylation is the determining factor for the recruitment of HP1α and HP1ß, while recruitment of HP1γ would be more dependent on chromatin opening by hSWI/SNF or the RNAPII. The role of HP1 proteins on active promoters is still enigmatic. The transient recruitment of Brg1 to the HIV1 promoter in sync with HP1γ recruitment shown in Figure 5D suggests that HP1 proteins could be involved in limiting remodeling on activated promoters. However, HP1 proteins may also have a role further downstream in connection with mRNA maturation [35].
The presence of HP1α on the IFIT1 and IFIT3 promoters and thereby the direct regulation of these genes by the HP1 protein was established by ChIP experiments (data not shown). However, the ChIP approach did not allow us to estimate the impact of hSWI/SNF remodeling on the recruitment of HP1α, as depletion of Brm leads to repression of the IFIT1 and IFIT3 promoters. Under these conditions, it was not possible to part between enhanced recruitment due to increased histone H3K9 methylation and decreased recruitment associated with reduced hSWI/SNF activity. However, our in vitro nucleosome-array binding assay suggests that HP1 proteins other than HP1γ can benefit from chromatin remodeling to bind nucleosomes. It must here be noted that this assay does not allow us to discriminate between histone and DNA binding. Nevertheless, repression of hSWI/SNF remodeling does not appear to rely on DNA binding as an HP1α mutant no longer binding DNA still represses. In addition, HP1γ does not bind to DNA in our hands and prefers nucleosomes (data not shown). Yet, it is efficient in repressing hSWI/SNF remodeling. Finally, we note that methylation of histone H3 on K9 is not required for repression of hSWI/SNF remodeling in vitro, further suggesting that the contact of the CSD with the Shadock, and not other contacts, is important for the repressing activity of HP1 proteins.
The interaction of HP1α CSD with histone H3 required HP1 dimerization and was disrupted by the mutation of V46 in the PXVXL-like sequence in the Shadock. It is therefore possible that this interaction can be structurally compared to that of other molecular partners of CSDs such as CAF1 or TIF1 proteins [36]. Interestingly, the rH3(35–66)V46A and rH3(44–66) mutants bind HP1γ but not HP1α, while H3 constructs with a full Shadock region bind both HP1 proteins. These observations show that HP1α and HP1γ have neighboring but distinct sites of interaction on the histone. While this manuscript was in revision, it was shown that phosphorylation of histone H3 on tyrosine 41 by JAK2 compromises the binding of the chromoshadow-domain of HP1α to an H3 peptide spanning from aa 31 to 56 [37]. This modification is outside the HP1γ binding site and it therefore seems possible that HP1α and HP1γ are differentially regulated by post-translational modifications in the H3 Shadock region.
The region of histone H3 contacted by the HP1 proteins also associates with the hSWI/SNF subunits Brg1 and Brm. This is compatible with earlier studies in yeast showing that residues K56 and L61 are involved in SWI/SNF recruitment [38],[39]. In yeast, it was suggested that the H3 αN helix is targeted and remodeled by the SWI/SNF complex [40]. More recently, mutagenesis within this region was shown to affect the efficiency of yeast SWI/SNF remodeling in vitro [24]. Our competition experiments with polypeptides mimicking the H3 αN helix further suggest that interaction of Brg1 with this region is essential for the chromatin remodeling activity of the hSWI/SNF complex. Consistent with this, the sequences downstream of the helicase domain of Brg1/Brm that mediate the interaction with the H3 Shadock domain have earlier been reported as essential for efficient remodeling in vitro [25]. We therefore speculate that the competing binding of Brg1/Brm and HP1 proteins to a same region of histone H3 during unwanted remodeling can at least in part explain the inhibiting effect of HP1 on chromatin opening by hSWI/SNF.
In vitro binding assays suggest that Brg1 has an affinity for the globular domain of H3 that is higher than that of HP1α (Figure S4). In a mechanism based on competition of Brg1 and HP1 proteins for binding to overlapping sites on histone H3, this difference in affinity may explain that a 7-fold excess of HP1α over Brg1 is required to obtain 50% inhibition of hSWI/SNF remodeling in the REA assays. It must finally be noted that the activity of the hSWI/SNF complex is also inhibited by the Polycomb Group (PcG) class II complex in vitro [26],[41]. This complex, involved in gene silencing, includes a CD protein that, like HP1 proteins, binds methylated histone H3 tails with a preference for methylation on K27 rather than K9 [42],[43]. However, Polycomb proteins contain no CSDs and it seems that PRC1 relies essentially on a structuring effect on the nucleosomal template, repressing remodeling by creating more condensed chromatin [44].
hSWI/SNF and HP1 proteins have many common target promoters including E2F1, human thymidine kinase, c-Myc, Sox2, Cyclin E, and the MMTV and HIV1 LTRs [5], [7], [45]–[51]. Likewise, we show here that several interferon-inducible genes that require hSWI/SNF for their activation are under the negative control of HP1α and HP1β. We note however that not all hSWI/SNF target genes we tested were affected by knock-down of HP1 proteins, including for example DraL and SPARC. These genes were both expressed at relatively high levels in the cells we used and may therefore not have any repressive structure on their promoter (data not shown). All considered, we suggest that on promoters where SWI/SNF functions as a repressor, the opening of the chromatin may, as previously suggested, be the event initiating HP1 stable recruitment [8],[20]. Where SWI/SNF functions as an activator, HP1 proteins recruited by methylated histone H3 tails may instead function as sensors of unwanted SWI/SNF activity, binding nucleosome domains uncovered by the remodeling, and thereby block the reaction (see proposed model in Figure 7). Finally, after the activation, additional HP1 proteins may be involved in controlling excessive remodeling activity. In that sense, HP1 proteins could very generally function as gatekeepers using the exposure of domains internal to the nucleosome to detect and restrict chromatin opening.
MCF7 cells were transfected by calcium phosphate precipitation as previously described [52]. When indicated, 10−6 M dexamethasone (dex) was added to the medium. 40 hours post-transfection, luciferase assays were performed using the Promega luciferase kit, following the instructions of the manufacturer.
Flag-epitope tagged human HP1α and Brg1 were expressed in SF9 cells using a bac-to-bac expression system (Gibco) and purified on M2 anti-flag beads (Sigma). hSWI/SNF was purified from a Flag-tagged Ini1 HeLa cell line as described previously [53]. HA-tagged HP1α constructs were expressed in E.coli and purified using their additional 6xHis affinity tag. HA-HP1αΔDNA contains a deletion from aa 87 to 111. HA-HP1αΔBrg1 was previously described [20]. The polynucleosome template was assembled by gradient salt dialysis with HeLa core histones and a 5S arrays fragment [53],[54]. The mononucleosome template was assembled on a 202 bp TPT containing DNA fragment by salt dialysis [55] and incubated at 37°C for 3 hours before use. REA assays were performed as previously described [26].
GST-HP1 fusions and GST-ΔBrg1 were constructed in pGEX3X and pET41 plasmids respectively. HA-tagged HP1 and B10-tagged histone H3 were constructed in pET-28 and pET15b plasmids respectively that also providing a 6xHis affinity tag used for purification. Pull-down experiments were performed in ELB buffer (50mM HEPES pH 7, 250mM NaCl, 1mM EDTA, 0.1% NP40, 1xComplete protease inhibitor cocktail from Roche). Bound proteins were eluted in 100mM Tris pH 8, 20mM glutathione, resolved by SDS-PAGE and detected by western blotting using the B10 monoclonal mouse anti-estrogen receptor α antibodies (Euromedex ERB10-As) or anti-histone H3 rabbit polyclonal (Abcam ab1791). Overlay assays and nucleosome array binding assays were previously described [14],[16].
siRNAs were synthesized by Dharmacon (ON-TARGET plus) : GAPDH (D-001830-01), Brm dh5 (J-017253-05), Brm dh8 (J-017253-08), HP1α (5′CACAAAUUGUGAUAGCAUU3′), HP1β (5′AGCUCAUGUUCCUGAUGAA3′) and HP1γ (5′AUCUGACAGUGAAUCUGAU 3′). siRNA were transfected into HeLa cells at 50 nM final concentration using DhamaFECT#1. Cells were harvested 3 days after transfection. RNAs were extracted using the Nucleobond RNA extract kit (Macherey-Nagel). mRNAs levels were quantified by real-time PCR after reverse-transcription performed at 50°C with SuperScriptIII (Invitrogene). Proteins were extracted as described previously [30] and detected by western blotting using anti-Brm (ab15597 Abcam) and anti-HP1α (2G9), anti-HP1β (1A9), and anti-HP1γ (1G6) from Euromedex.
ChIP and ChIP-reChIP experiments were performed as previously described [56] using Jurkat J-Lat TAT-IRES-GFP clone A1 (NIH AIDS Research & Reference Reagent Program), or HeLa derived clones expressing HA-tagged HP1γ [5] and the following antibodies: anti-RNAPII (ab5095 Abcam), anti-Brg1 (2E12 Euromedex), anti-HP1γ (42S2 Millipore), anti-HA (12CA5), and anti-Brm (ab15597 Abcam). The eluted DNAs were detected by real-time PCR.
Real-time PCR were performed with the SYBR Green kit Brilliant II (Agilent) reagents in a Mx3000 qPCR machine (Stratagene). The primers were the following:
IFIT1-F: ACACCTGAAAGGCCAGAATGAGGA,
IFIT1-R: TGCCAGTCTGCCCATGTGGTAATA,
IFIT3-F: AGCAAGAACATGCTGACCAAGCAG,
IFIT3-R: ACTTCAGTTGTGTCCACCCTTCCT,
OASL-F: ATGTTGGACGAAGGCTTCACCACT,
OASL-R: ATCTGTACCCTTCTGCCACGTTGA,
OAS1-F: GTTCTCCACCTGCTTCACAGAACT,
OAS1-R: CGAAATCCCTGGGCTGTGTTGAAA,
IFIT5-F: ATGGCCGCTTTCAGGAATTTCACC,
IFIT5-R: AGCACTTGTCAGTTTGGTGCGAAG,
HPRT-F: TATGGACAGGACTGAACGTCTTGC,
HPRT-R: TGAGCACACAGAGGGCTACAAT.
IFIT3_TS_F: AAAGCACAGACCTAACAGCACCCT,
IFIT3_TS_R: CATGATGGCTGTTTCCCTGCAGTT.
HIV1 TS primers were previously described [5].
|
10.1371/journal.pgen.1000374 | A Drosophila Model for EGFR-Ras and PI3K-Dependent Human Glioma | Gliomas, the most common malignant tumors of the nervous system, frequently harbor mutations that activate the epidermal growth factor receptor (EGFR) and phosphatidylinositol-3 kinase (PI3K) signaling pathways. To investigate the genetic basis of this disease, we developed a glioma model in Drosophila. We found that constitutive coactivation of EGFR-Ras and PI3K pathways in Drosophila glia and glial precursors gives rise to neoplastic, invasive glial cells that create transplantable tumor-like growths, mimicking human glioma. Our model represents a robust organotypic and cell-type-specific Drosophila cancer model in which malignant cells are created by mutations in signature genes and pathways thought to be driving forces in a homologous human cancer. Genetic analyses demonstrated that EGFR and PI3K initiate malignant neoplastic transformation via a combinatorial genetic network composed primarily of other pathways commonly mutated or activated in human glioma, including the Tor, Myc, G1 Cyclins-Cdks, and Rb-E2F pathways. This network acts synergistically to coordinately stimulate cell cycle entry and progression, protein translation, and inappropriate cellular growth and migration. In particular, we found that the fly orthologs of CyclinE, Cdc25, and Myc are key rate-limiting genes required for glial neoplasia. Moreover, orthologs of Sin1, Rictor, and Cdk4 are genes required only for abnormal neoplastic glial proliferation but not for glial development. These and other genes within this network may represent important therapeutic targets in human glioma.
| Malignant gliomas, tumors composed of glial cells and their precursors, are the most common and deadly human brain tumors. These tumors infiltrate the brain and proliferate rapidly, properties that render them largely incurable even with current therapies. Mutations in genes within the EGFR-Ras and PI3K signaling pathways are common in malignant gliomas, although how these genes specifically control glial pathogenesis is unclear. To investigate the genetic basis of this disease, we developed a glioma model in the fruit fly, Drosophila melanogaster. We found that constitutive coactivation of the EGFR-Ras and PI3K pathways in Drosophila glia gives rise to highly proliferative and invasive neoplastic cells that create transplantable tumor-like growths, mimicking human glioma. This represents a robust cell-type-specific Drosophila cancer model in which malignant cells are created by mutations in genetic pathways thought to be driving forces in a homologous human cancer. Genetic analyses demonstrated that EGFR-Ras and PI3K induce fly glial neoplasia through activation of a combinatorial genetic network composed, in part, of other genetic pathways also commonly mutated in human glioma. This network acts synergistically to coordinately stimulate cellular proliferation, protein translation, and inappropriate migration. Rate-limiting genes within this network may represent important therapeutic targets in human glioma.
| Malignant gliomas, neoplasms of glial cells and their precursors, are the most common tumors of the central nervous system (CNS). These tumors typically proliferate rapidly, diffusely infiltrate the brain, and resist standard chemotherapies, properties that render them largely incurable. One key to developing more effective therapies against these tumors is to understand the genetic and molecular logic underlying gliomagenesis. The most frequent genetic lesions in gliomas include mutation or amplification of the Epidermal Growth Factor Receptor (EGFR) tyrosine kinase. Glioma-associated EGFR mutant forms show constitutive kinase activity that chronically stimulates Ras signaling to drive cellular proliferation and migration [1],[2]. Other common genetic lesions include loss of the lipid phosphatase PTEN, which antagonizes the phosphatidylinositol-3 kinase (PI3K) signaling pathway, and activating mutations in PIK3CA, which encodes the p110α catalytic subunit of PI3K [1],[2]. Gliomas often show constitutively active Akt, a major PI3K effector [1],[2]. However, EGFR-Ras or PI3K mutations alone are not sufficient to transform glial cells, rather multiple mutations that coactivate EGFR-Ras and PI3K-Akt pathways are sufficient to induce glioma [2]–[4].
Understanding the interplay of these mutations and the neurodevelopmental origins of these tumors could lead to new insights into the mechanisms of gliomagenesis. The mammalian brain contains multiple glial cell types that maintain proliferative capacities, including differentiated astrocytes, glial progenitors, and multipotent neural stem cells. EGFR-Ras and PTEN-PI3K signaling regulates many developmental processes in these cell types, particularly proliferation and self-renewal, which are also properties of glioma cells [1]. Although recent hypotheses favor that gliomas arise from multipotent stem cells, data from mouse models demonstrate that differentiated glia, glial progenitors, and stem cells can all produce gliomas in response to genetic lesions found in human gliomas [5],[6]. Thus, misregulation of these genetic pathways may confer unrestricted proliferative capacities to a range of glial cell types, but how this occurs remains unclear. While many of the same effectors are utilized by EGFR-Ras and PI3K in both glial development and cancer, constitutive activation of these pathways may deploy distinct outputs, not utilized in development, that allow particular cells to escape normal physiological cues that restrain proliferation and self-renewal. The identity of such outputs remains unclear.
With these issues in mind, we developed a Drosophila glioma model to facilitate genetic analysis of glial pathogenesis. Drosophila offers many tools for precise manipulation of cell-type-specific gene expression and dissection of multigene interactions. Most human genes, including 70% of known disease genes, have functional Drosophila orthologs [7]. Among the most conserved genes are components of major signal transduction pathways, including many gliomagenic genes. Recently, Drosophila has emerged as a model system for human neurological diseases because the CNS shows remarkable evolutionary conservation in cellular composition and neurodevelopmental mechanisms [8]. Similarly, Drosophila have multiple glial cell types that require the EGFR pathway for their normal development, and these cells appear homologous to mammalian glia in terms of function, development, and gene expression [9]. These similarities between flies and humans make Drosophila an attractive system for modeling gliomas.
Since concurrent activation of EGFR-Ras and PI3K signaling in glial precursors induces glioma in the mouse [4], we sought to create mutant phenotypes by hyperactivation of these pathways in fly glia and glial precursors. Drosophila has a single functional ortholog each for EGFR(dEGFR), Raf (dRaf), PIK3CA(dp110), PTEN(dPTEN), and Akt(dAkt), and two functional orthologs for Ras(dRas85D, dRas64B) (www.flybase.org). A diagram of the specific mutant forms of dEGFR used in our assays can be found in Figure S1. We performed glial overexpression assays with the Gal4-UAS system [10], using the repo-Gal4 driver, which gives sustained UAS-transgene expression in almost all glia, from embryogenesis through adulthood. For glial-specific RNAi, we employed UAS-dsRNA constructs [11], which we verified with phenotypic tests and/or antibody staining (see Materials and Methods). Glial morphology was visualized with membrane-localized GFP (CD8GFP) [12]. Cell number was determined with staining for Repo, a homeobox transcription factor expressed by repo-Gal4 positive glia [13].
Glial-specific coactivation of EGFR-Ras and PI3K stimulated glial neoplasia, giving rise to CNS enlargement and malformation, neurologic defects, and late larval lethality. repo-Gal4-driven co-overexpression of activated dEGFR (dEGFRλ) and dp110 (dp110CAAX) induced progressive accumulation of ∼50-fold excess glia (Figure 1A and 1B) [14],[15]. dEGFRλ is a constitutively active dEGFR variant in which a lambda dimerization domain replaces the extracellular domain [14] (Figure S1). Co-overexpression of combinations of dEGFRλ and core components of the PI3K pathway, such as dAkt, induced phenotypes similar to repo>dEGFRλ;dp110CAAX, although phenotypes varied somewhat depending on strength of pathway activation and transgene expression (Figure S2 and Table S1). Dramatic glial overgrowth also occurred upon co-overexpression of constitutively active dRas (dRas85DV12) or its effector dRaf (dRafgof) with dp110CAAX, dAkt, or a dPTENdsRNA, which partially knocked-down dPTEN (Figure S2 and Table S1). Finally, glial overgrowth in repo>dEGFRλ;dp110CAAX larvae was strongly suppressed by co-overexpression of dPTEN or more moderately by dominant negative dRas85D (dRas85DN17) (Figure 1D and 1E), indicating that Ras activity and excess phospho-inositols are essential for neoplasia.
In contrast, glial-specific activation of the EGFR-Ras pathway alone, through overexpression of dEGFRλ or Rafgof, induced 5–10 fold excess glia in the larval brain and later pupal lethality (Figure 1F and Table S1). dRas85DV12 overexpression induced approximately 5–10-fold excess glia, and these glia were smaller than wild-type or dEGFRλ;dp110CAAX glia. (Figure 1G). dRas85DV12 may be more potent than dEGFRλ because dRas85DV12 can activate endogenous PI3K signaling [16]. Overexpression of dEGFRElp, a classical hypermorphic mutant form of dEGFR [17], induced excess glial proliferation and neural morphogenesis defects (Figure S2), but also caused early lethality which precluded examination of dEGFRElp-dp110 interactions. As in mouse models, overexpression of wild-type dEGFR failed to induce excess glia [6],[17], and instead retarded CNS growth (Figure S2). Unlike dEGFRλ, dEGFRWT and dEGFRElp have functional ligand-binding domains (Figure S1), and may cause additional defects by sequestering ligand otherwise required for normal development [17]. Glial-specific activation of the PI3K pathway alone, either by overexpression of dp110CAAX, dp110wild-type, dAkt, or dPTENdsRNA gave viable animals with relatively normal brains (Figure 1H, Figure S2, and Table S1). Therefore, coactivation of the EGFR and PI3K pathways synergize to produce much more severe phenotypes than would be expected if the effects of these pathways were additive.
In repo>dEGFRλ;dp110CAAX brains, excess glia emerged in early larval stages and accumulated over 5–7 days. dEGFRλ;dp110CAAX glia severely disrupt the normal cellular architecture of the larval brain (Figure 1A and 1B and Figure 2A–C), lose normal stellate glial morphologies (Figure 2A–C), and generate multilayered aggregations of abnormal glia throughout the brain (Figure 2A–C); in these ways dEGFRλ;dp110CAAX glia are neoplastic [18]. Like neoplastic epithelial cells, dEGFRλ;dp110CAAX glia ectopically expressed an active form of the matrix metalloprotease dMMP1 (Figure S3), which can confer an invasive potential [19],[20], implying that abnormal dEGFRλ;dp110CAAX glia may be invasive within the brain. Unlike neoplastic epithelia, neoplastic neural cells, such as dEGFRλ;dp110CAAX glia, typically retain expression of genes that regulate neural cell fate, such as Repo [21],[22].
Relative to controls, many dEGFRλ;dp110CAAX glia showed BrdU incorporation, which marks S-phase cells (Figure 2D and 2E), indicating that neoplastic glia arise from overproliferation. repo>dEGFRλ;dp110CAAX animals also showed reduced BrdU in neuronal precursors (Figure 2E, data not shown), demonstrating that neoplastic glia disrupt neuronal development. The cell cycle is governed by CyclinD-Cdk4 and CyclinE-Cdk2 complexes, which phosphorylate and inactivate Rb proteins, to release E2F activators to stimulate G1-S-phase entry [23]. Kip-type (p21/p27/p57) and Ink-type cyclin-dependent kinase inhibitors antagonize proliferation by inhibiting CyclinE-Cdk2 and CyclinD-Cdk4, respectively. Cdc25 phosphatases and mitotic cyclins, including CyclinB, promote G2-M progression. Flies have single orthologs each for CyclinE, Cdk2, CyclinD, Cdk4, CyclinB, and p21/p27/p57 (Dap), E2F activators (E2F1) and two orthologs for Rb (Rbf1 and Rbf2) and Cdc25 (Stg and Twe) but no Ink ortholog [23].
dEGFRλ;dp110CAAX glia showed ectopic expression of dCyclinE and dCyclinB (Figure 2F–I), demonstrating that EGFR and PI3K activity upregulated proteins that promote cell cycle entry and progression. High-grade human gliomas contain highly proliferative anaplastic glia and enlarged pleiomorphic polyploid glia [24]. Similarly, repo>dEGFRλ;dp110CAAX larvae showed accumulation of small, highly proliferative glia that strongly expressed cyclins and labeled with BrdU. repo>dEGFRλ;dp110CAAX larval brains also displayed abnormal polyploid glia, as assessed by DAPI staining (data not shown), and these cells typically expressed only dCyclinE but not dCyclinB, and thereby likely underwent ectopic DNA replication without mitosis (Figure 2G and 2I, data not shown). However, overexpression of dCyclinE-dCdk2, dCyclinD-dCdk4, or dE2F1-dDp complexes and/or Rbf1 knock-down did not cause neoplasia, and instead either doubled glial cell numbers or resulted in embryonic lethality (Figure S4, data not shown).
We next examined negative regulators of the cell cycle. dEGFRλ;dp110CAAX glia expressed Rbf1, but showed little Dap, a result we also observed in wild-type glia (Figure S5). Dap inhibits dCyclinE-cdk2 complexes [25], and is transiently expressed in neural progenitors to promote cell cycle exit as they begin differentiation [26]. Dap overexpression completely suppressed repo>dEGFRλ;dp110CAAX glial overgrowth (Figure 2L), demonstrating that glial neoplasia is cell-autonomous and requires dCyclinE-dCdk2. Similarly, overexpressed Rbf1 and dCyclinE mutations also reduced repo>dEGFRλ;dp110CAAX glial overproliferation (Figure 2M, data not shown). The gross neural morphogenesis defects observed in repo>dEGFRλ;dp110CAAX brains may be secondary to glial overproliferation since these defects were largely prevented by Dap or Rbf1 co-overexpression (Figure 2J–M). In repo>dEGFRλ;dp110CAAX animals, other mutant glia outside of the brain, such as peripheral glia, also became highly proliferative and invasive, and these defects, too, were corrected by Rbf1 or Dap overexpression (data not shown). In controls, Rbf1 or Dap overexpression in wild-type glia inhibited proliferation (Figure S4), reducing numbers of glia by approximately half. Together, these results suggest that repo-Gal4 glia undergo at least one round of cell division, consistent with published studies [27],[28], and this proliferation becomes prolonged by constitutive coactivation of EGFR and PI3K signaling.
The phenotype triggered by coactivation of EGFR and PI3K in glia is distinct from other Drosophila brain-overgrowth mutant phenotypes, which involve accretion of excess neurons or neuroblasts [21],[29]. repo>dEGFRλ;dp110CAAX cells were not transformed into neurons or neuroblasts as they lacked expression of the Elav and Miranda markers (Figure S6). Lineage-tracing with a Su(H)-lacZ neuroblast reporter showed that excess dEGFRλ;dp110CAAX glia did not express LacZ, and thus are not directly derived from larval neuroblasts (data not shown). Moreover, constitutive EGFR-Ras and PI3K signaling does not elicit overgrowth in all neural cell types, as assessed with defined cell-type specific Gal4-drivers (Table S2). For example, dRas85DV12 overexpression in fly neurons causes defects in fate specification, patterning, and apoptosis [30],[31]. Co-overexpression of dEGFRλ or dRas85DV12 with dp110CAAX in neurons (elav-Gal4, scratch-Gal4, OK107-Gal4, and Appl-Gal4) and neuroblasts/neuronal precursors (pros-Gal4, wor-Gal4, and 1407-Gal4) did not induce overgrowth (Figure S7, data not shown), even with increased transgene expression from Gal4 amplification [32]. In fact, broad co-overexpression of dEGFRλ and dp110CAAX in neuroblasts and neuronal precursors (pros-Gal4) reduced brain size, perhaps because signaling through these pathways stimulates precocious cell cycle exit of neuronal precursors, as in the developing eye [33]. Furthermore, transient expression of dRas85DV12 or dEGFRλ and dp110CAAX in embryonic glia (gcm-Gal4) also failed to promote glial overgrowth (Figure S7) [27], demonstrating that sustained activation of these pathways is required for glial overproliferation. Certain glial subtypes, such as oligodendrocyte-like neuropil glia (Eaat1-Gal4), some astrocyte-like cortex glia (Nrv2-Gal4), and peripheral perineurial glia (gli-Gal4) also failed to become neoplastic in response to EGFR-Ras and PI3K (Figure S7, data not shown) [34]–[36]. Therefore, neoplastic proliferation is not a uniform cellular response to EGFR and PI3K.
Since repo>dEGFRλ;dp110CAAX animals die in 5–7 days, we assessed the proliferative potential of mutant glia using an abdominal transplant assay, a classic test of tumorigencity in flies [37]. Brain fragments from repo>dEGFRλ;dp110CAAX and wild-type larvae were transplanted into young adults. Wild-type transplants grew and survived over 1–6 weeks, but produced few glia (Figure 3A and 3C). dEGFRλ;dp110CAAX mutant glia survived and proliferated into massive tumors that filled the hosts' abdomens, often causing premature death (Figure 3B). Tumors were composed of small glial cells with little cytoplasm (Figure 3D–F). Tumors also contained trachea embedded throughout their mass (Figure 3D and 3E and Video S1), suggesting that tumors stimulated growth of new trachea or enveloped existing trachea, perhaps in a process akin to tumor angiogenesis. The leading edges of the tumors harbored individual cells invading nearby tissues, such as the ovary (Figure 3F and Video S2), which is consistent with the ectopic expression of active dMMP1 observed in dEGFRλ;dp110CAAX glia in the larval brain (Figure S3). However, some tissues, such as the gut, did not contain metastases, implying some degree of selective invasion. Thus, once unconstrained by the larval life cycle, dEGFRλ;dp110CAAX glia fail to exit the cell cycle, continue to proliferate, and form highly invasive tumors, all properties of human cancer cells.
To explore the invasive potential of mutant glia, we used FLP-FRT clonal analysis, a technique in which discrete clones of mutant cells are induced in otherwise normal tissues, a situation analogous to somatic tumorigenesis. We used a heat-shock-driven FLP-recombinase to catalyze mitotic recombination between FRT-bearing chromosomes such that a daughter cell (and all of its clonal progeny) initiated expression of GFP and UAS-containing transgenes only in repo-Gal4-expressing cells [12]. Clones were induced late in development, from mitotic founder cells, and were examined in adults. We could not definitively determine if clones were derived from single cell events since our study of these clones was retrospective, but given the frequency of control clone induction, many mutant clones likely originated from single cells.
In wild-type controls, we observed clones in 68% of brains examined (N = 149). Of the brains with clones, 75% had 1–3 clones, and 83% of these clones consisted of 1–3 cells of the same glial subtype (Figure 4A and 4B). Glial clones overexpressing dRas85DV12, dEGFRλ, or dEGFRElp alone typically contained 2-fold more cells than wild-type (dRas85DV12 shown in Figure 4C). To examine PI3K signaling, we used a dPTEN null allele, which became homozygous in FLP-FRT clones. dPTEN−/− glia did not overgrow, but did show aberrant cytoplasmic projections (Figure 4D), perhaps reflecting dPTEN function in the cytoskeleton [38].
To coactivate EGFR-Ras and PI3K in glial clones, dRas85DV12, dEGFRElp, or dEGFRλ was overexpressed within dPTEN−/− glia, using repo-Gal4. We observed overgrown and invasive dRas85DV12;dPTEN−/− and dEGFRElp;dPTEN−/− clones (Figure 4E–G). dEGFR;dPTEN−/− clones were less affected (Figure 4F and 4G), consistent with the larval overexpression of dRas85DV12 giving more severe growth phenotypes. Tumor-like overgrowth was only observed in dEGFR-dRas85D;dPTEN−/− cells, illustrating that chronic EGFR-Ras activation and PTEN loss can cause cell-autonomous over-proliferation. Cells from these double mutant clones appeared to invade the brain, typically following fiber tracts, and sometimes induced the formation of trachea (Figure 4E). Tumor-like growths of dEGFR-dRas85D;dPTEN−/− cells often penetrated deep into the brain, as exemplified by Videos S3 and S4 which show an animated 88 µm thick confocal z-stack of the dRas85DV12;dPTEN−/− clone in Figure 4E compared to a 16 µm thick z-stack of a wild-type control clone in Figure 4A. These phenotypes were reminiscent of invasion and angiogenesis in human gliomas [24]. We more commonly observed smaller dEGFR-dRas85D;dPTEN−/− clones composed of relatively differentiated, enlarged glia with diffusely invasive projections (Figure 4G); these clones likely derive from glia that differentiated prior to achieving sufficient EGFR-Ras transgene expression, and are consistent with findings that not all glial subtypes become neoplastic.
Since neoplastic larval glia were concentrated in the outer anterior central brain and developing optic lobe, they may be derived from glial progenitor cells present in these regions [27],[28],[39]. To create clones from a discrete subpopulation of glial progenitors, we used an eyeless(ey)-promoter driven FLP-recombinase (ey-FLP), which is active in ey-expressing glial progenitors in the optic lobe [27]. Single mutant ey-FLP clones of dPTEN−/−, dRas85DV12, or dEGFRElp cells contained a modest number of excess and abnormal glia relative to wild-type controls (Figure 4H–J and Figure S8). In contrast, the double mutant dRas85DV12;dPTEN−/−, dEGFRElp;dPTEN−/−, or dEGFRλ;dPTEN−/− ey-FLP clones, which emerge as approximately tens of cells in 3rd instar larval brains (Figure 4K), became large invasive tumors composed of hundreds to thousands of cells in adults (Figure 4L and Figure S8).
To address the function of EGFR-Ras and PI3K in glioma, we analyzed the genetic basis of glial pathogenesis in our repo>dEGFRλ;dp110CAAX model, as this model shows robust neoplasia, similarity to human tumor genotypes, and sensitivity to dEGFR and dPTEN gene dosage (data not shown).
EGFR-Ras signaling can promote proliferation through Erk kinase-mediated induction of nuclear targets. In repo>dEGFRλ;dp110CAAX brains, mutant glia showed high levels of nuclear, activated di-phospho-Erk relative to wild-type glia in control brains (Figure 5A and 5B). In flies, Erk activity can induce expression of PntP1, an ETS-family transcription factor encoded by the pointed (pnt) locus [40],[41]. PntP1, which is expressed in embryonic glia and is required for their normal development [40], was upregulated in dEGFRλ;dp110CAAX glia (Figure 5C and 5D). High levels of PntP1 can be detected normally in neuronal progenitors (data not shown), suggesting that it promotes a proliferative progenitor state. In developing eye tissue, EGFR-Ras-Erk signaling induces Pnt proteins to stimulate G2-M cell cycle progression through direct upregulation of Stg (Cdc25 ortholog) expression [41]. Glial-specific RNAi knock down of pnt reduced Stg expression and completely suppressed dEGFRλ;dp110CAAX neoplasia (Figure 5E–H), demonstrating that Pnt proteins are required for both Stg expression and neoplastic overproliferation in dEGFRλ;dp110CAAX glia. Notably, in repo>pntdsRNA;dEGFRλ;dp110CAAX brains, glia maintained their fate, as evidenced by repo-Gal4 and Repo expression.
Stg itself was rate limiting for glial neoplasia. Reduction of stg with a mutation or a stgdsRNA partially suppressed the repo>dEGFRλ;dp110CAAX phenotype, whereas overexpressed Stg synergistically enhanced neoplasia (Figure 5E and Figure S9, data not shown). In contrast, Stg overexpression alone increased glial cell numbers approximately 2-fold, and could not induce neoplasia when combined with PI3K effectors (data not shown). Thus, dEGFRλ;dp110CAAX induces neoplasia via coordinated stimulation of G1-S entry through dCyclinE, and G2-M progression through Stg, both of which are EGFR-Ras dependent outputs [41],[42].
We sought to determine which PI3K effectors contribute to the repo>dEGFRλ;dp110CAAX phenotype. Genetic reduction of dAkt, a major target of PI3K signaling, with a dAktdsRNA or a mutant allele strongly suppressed repo>dEGFRλ;dp110CAAX glial neoplasia (Figure 6A–C, data not shown). Therefore, Akt is necessary for the outcome of EGFR-PI3K coactivation. Many Akt effectors are implicated in glioma, and we tested orthologs of these loci in our model (Table S3).
Tor, a kinase that promotes cell growth and proliferation, is a key Akt target. In glioma models, coactivation of EGFR-Ras and PI3K stimulates Tor, and in humans, Tor activity is correlated with poor patient prognosis [43],[44]. We tested the single Drosophila Tor ortholog, dTor, by genetically reducing dTor activity in repo>dEGFRλ;dp110CAAX larvae with a viable combination of hypomorphic dTor alleles or by co-overexpression of dominant negative dTor 45,46. Both of these manipulations reduced glial overgrowth (Figure 6D, data not shown).
In flies and mammals, Tor exists in two different signaling complexes, TORC1 and TORC2. TORC2, a complex including Tor and the Sin1 and Rictor regulatory proteins, directly phosphorylates Akt, creating a positive feedback loop that fully activates Akt [47]. In mouse, Sin1 and Rictor mutants die early due to extraembryonic defects, but dSin1 and dRictor mutant flies are viable as homozygous nulls, allowing us to remove TORC2 function genetically [47],[48]. dSin1−/−; repo>dEGFRλ;dp110CAAX larval brains showed a near-wild type phenotype (Figure 6E). Results were similar with a dRictor null allele and a dSin1dsRNA (data not shown). dSin1−/− and dRictor−/− mutants display reduced Akt-dependent phosphorylation and inactivation of dFoxO [49], the single Drosophila ortholog of FoxO transcription factors. This suggests that TORC2 loss might antagonize glial neoplasia through dFoxO upregulation. However, excess dFoxOSA, which is resistant to dAkt phosphorylation [45], only partially suppressed repo>dEGFRλ;dp110CAAX glial overproliferation (Figure 6F), arguing that TORC2 has additional roles. Notably, on their own, dSin1−/− and dRictor−/− mutant flies did not show any detectable glial defects (Figure 6G, data not shown). Thus, TORC2 is dispensible for normal glial development, but is necessary for dEGFRλ;dp110CAAX glial neoplasia.
TORC1, a complex including Tor and the Raptor regulatory protein, drives cellular growth by stimulating protein synthesis through its effectors S6 kinase and the eIF-4E translation initiation factor [47]. Akt and Erk stimulate TORC1 through phosphorylation and inactivation of the TSC1-TSC2 protein complex, which activates Rheb, and stimulates TORC1 kinase activity. We tested TORC1 function by glial-specific overexpression of dsRNAs for the single Drosophila orthologs of Raptor (dRaptor), S6-kinase (dS6K), and eIF4E (deIF4E); these all significantly reduced accumulation of dEGFRλ;dp110CAAX mutant glial cells, but only caused mild glial hypoplasia in controls (Figure 6I–K and Figure S10). Co-overexpression of d4EBP, a deIF4E antagonist and dFoxO target gene, also blocked glial neoplasia (Figure 6L). Glial-specific RNAi of dTSC1 enhanced repo>dEGFRλ;dp110CAAX glial overgrowth (Figure 6H). However, overexpression of dTSC1dsRNA, dRheb, activated dS6K (dS6Kact), or deIF4E alone did not produce glial overproliferation, even though these constructs can mimic TORC1 activation (Figure S10, data not shown) [47],[49]. Thus, TORC1 activity is necessary for EGFR-PI3K-driven glial neoplasia, but is not sufficient. Moreover, neither deIF4E nor dS6Kact produced glial neoplasia when co-overexpressed with dEGFRλ (data not shown), illustrating that additional dTor-dependent outputs synergize with dEGFR signaling to drive neoplasia.
dTor coordinates increased translation, mediated by dS6K and deIF4E, with expression of cell cycle regulators and ribosomal components, through dMyc, the single Drosophila ortholog of the Myc bHLH transcription factors [49]. Within developing epithelial tissues, dMyc is required for TORC1-dependent growth and can substitute for dTor activity [49],[50]. Myc protein levels can also be posttranslationally upregulated by EGFR-Ras-Raf signaling [16],[51]. Thus, we suspected that dMyc might mediate signal integration between EGFR-Ras and PI3K.
dEGFRλ;dp110CAAX glia showed high levels of nuclear dMyc compared to wild-type glia (Figure 7A and 7B). dMyc was also highly expressed in wild-type neuroblasts (Figure 7A), suggesting that dMyc promotes a proliferative progenitor state [21]. Genetic reduction of dMyc with a dsRNA or a single loss-of-function allele strongly suppressed dEGFRλ;dp110CAAX glial neoplasia (Figure 7D, data not shown). In fact, some dMyc+/−; repo>dEGFRλ;dp110CAAX animals were rescued to viability, indicating that dMyc is an essential rate-limiting output of EGFR-PI3K coactivation.
Myc proteins activate transcription through heterodimerization with the bHLH Max. Max activity was also required for glial neoplasia; a dsRNA for dMax, the single Drosophila Max ortholog, strongly suppressed repo>dEGFRλ;dp110CAAX overgrowth (Figure 7E). dMyc-dMax heterodimers promote proliferation by activating expression of multiple cell cycle genes, including dCyclinD and dCdk4 [52]. dCyclinD expression, which is high in repo>dEGFRλ;dp110CAAX glia, was inhibited by a dMycdsRNA (Figure 7F–H), suggesting that dMyc reduction suppresses glial neoplasia through reduced dCyclinD-dCdk4 activity. To test this we used loss-of-function mutations in dCdk4, which are viable [23]. dCdk4−/−; repo>dEGFRλ;dp110CAAX larvae showed near complete absence of excess glia and rescue of glial morphogenesis defects (Figure 7I). Moreover, a dCdk4dsRNA and dCyclinDdsRNA suppressed repo>dEGFRλ;dp110CAAX (data not shown). Thus, dCyclinD-dCdk4 is essential for EGFR-PI3K glial neoplasia, although dCdk4 is not required for development (Figure 7J).
dMyc is necessary for glial neoplasia, but is not sufficient when overexpressed alone (Figure 8C). dMyc-overexpressing glia showed polyploidy (Figure 8C, data not shown), indicating that these cells undergo DNA replication without mitosis, but require additional signals for cell cycle progression. In contrast, co-overexpression of dMyc with dEGFRλ produced a phenotype on par with that of repo>dEGFRλ;dp110CAAX (Figure 8D), indicating that dMyc overexpression can substitute for PI3K activation and promote neoplasia when combined with EGFR signaling.
Given that the dMyc targets dCyclinD-dCdk4 are required for dEGFRλ;dp110CAAX neoplasia, we tested whether dCyclinD-dCdk4 overexpression could cooperate with dEGFRλ. Additionally, we tested loss of Rbf1, the only known dCdk4 substrate that controls proliferation [23]. repo>dCyclinD;dCdk4;dEGFRλ and repo>Rbf1dsRNA;dEGFRλ animals showed glial overgrowth, but did not accumulate as many cells as repo>dEGFRλ;dp110CAAX animals (Figure 8E and 8F). Thus, glia likely require additional dp110 or dMyc effectors to undergo full neoplastic proliferation. Other known PI3K and dMyc-dMax target genes that promote proliferation include ribosomal proteins and translation regulators [49],[52], such as eIF4E, which is highly expressed in dEGFRλ;dp110CAAX glia and required for neoplasia (Figure 6 and Figure S11).
Our data imply that PI3K, dMyc, and dCyclinD-dCdk4 exist in a linear system, in which Rbf1 inactivation by dCyclinD-dCdk4 is one direct output of dp110CAAX or dMyc. However, in high-grade glioma, Rb loss co-occurs with EGFR and PTEN mutations [2], implying that these mutations cooperate to promote gliomagenesis. To explore interactions between Rb and EGFR-PI3K, we created the triple mutant repo>Rbf1dsRNA;dEGFRλ;dp110CAAX. These animals displayed exacerbated glial neoplasia, with a substantial increase in small anaplastic-like glia throughout the CNS (Figure 8G). This synergistic interaction likely derives from derepression of dE2F1 upon Rbf1 loss, and concomitant increased expression of dE2F1 target genes, including Stg and dCyclinE [53]. Increased dCyclinE and Stg expression may accelerate cell cycle progression, perhaps through increased dCdk2 and dCdk1 activity and/or truncated G1 and G2 gap phases caused by constant dCyclinE and Stg protein levels [23]. Consistent with this, we observed increased dCyclinE expression in Rbf1dsRNA;dEGFRλ;dp110CAAX glia relative to dEGFRλ;dp110CAAX glia (Figure 8H and 8I), and co-overexpression of Stg or dCyclinE-dCdk2 with dEGFRλ;dp110CAAX synergistically exacerbated glial neoplasia, yielding phenotypes similar to repo>RbfdsRNA;dEGFRλ;dp110CAAX (Figure 5E and Figure S9, data not shown). To assess dCdk2 activity in repo>Rbf1dsRNA;dEGFRλ;dp110CAAX brains compared to repo>dEGFRλ;dp110CAAX brains, we stained for phospho-MPM2 (Figure S12), which detects nuclear foci in cells with active dCyclinE-dCdk2 complexes [54]. Phospho-MPM2 foci were present in glia of both genotypes, although repo>Rbf1dsRNA;dEGFRλ;dp110CAAX brains appeared to have a higher density of glia with phospho-MPM2 foci (Figure S12), suggesting that expanded expression of dCyclinE results in broader activation of dCdk2. Thus, while PI3K and Rbf1 act in a common genetic pathway linked by dCyclinD-dCdk4, Rbf1 loss nevertheless synergizes with mitogenic stimulation from combined EGFR and PI3K signaling, and this synergy emerges from increased expression of dCyclinE and Stg, rate-limiting regulators of the cell cycle.
We show that constitutive coactivation of EGFR-Ras and PI3K signaling in Drosophila glia and glial precursors gives rise to neoplastic, invasive cells that create transplantable tumor-like growths, mimicking human glioma, and mirroring mouse glioma models. This represents a robust organotypic and cell-type specific Drosophila cancer model in which malignant cells are created by mutations in the signature genes and pathways thought to be driving forces in a homologous human cancer. This was not necessarily an expected result since fly and human glia show many biological differences despite displaying important similarities [9],[55],[56]. Through genetic analysis of our model, we identified crucial downstream effectors of EGFR and PI3K signaling, many of which are mutated and/or activated in human glioma. These effectors act in a combinatorial network to coordinately stimulate cell cycle entry and progression, block cell cycle exit, and promote inappropriate cellular growth and migration (Figure 8J). Pathways within this network, while interdependent, act synergistically, rather than additively. Thus, Drosophila shows evolutionary conservation of oncogene cooperation.
At least four pathway circuits are necessary for glial neoplasia initiated by EGFR and PI3K signaling, including dRas and dMyc circuits, which induce dCyclinE and dCyclinD to drive cell cycle entry, a Pnt circuit, which induces Stg to promote cell cycle progression, and a Tor-eIF4E-S6K pathway, which provides protein translation necessary for proliferation and growth (Figure 8J). When activated individually, these pathways fail to elicit glial neoplasia, implying a requirement for coordinated stimulation of multiple effectors and inactivation of negative regulators. Orthologs for many of the genes within these pathways, such as dRictor, are implicated in human glioma, although specific roles for some, such as ETS transcription factors, have not been defined despite their expression in glioma [2],[57],[58]. While many of these genes are known EGFR and PI3K pathway components, we did not necessarily expect them to be required for EGFR and PI3K dependent glial neoplasia. Indeed, we have tested many other pathway components and outputs, such as Jun kinase, that did not significantly suppress repo>dEGFRλ;dp110CAAX phenotypes upon reduced function (unpublished data).
Coactivation of EGFR and PI3K signaling upregulates dMyc, which is necessary for glial neoplasia. This is consistent with findings that, in flies and mammals, EGFR-Ras, PI3K, and Tor signaling upregulate Myc protein levels [16],[49],[51],[59],[60]. Myc oncogenes are well-known to cooperate with RTK-Ras signaling to drive neoplastic transformation [51], and we demonstrate that this property of Myc is conserved in flies. We also observed sensitivity to reduced Myc gene dosage in our glioma model, which has also been recently documented in a mouse model of PTEN-dependent glioma [61]. c-myc is commonly amplified in gliomas [62], implying that Myc is rate limiting, and c-myc amplification may be selected for this reason. D-cyclins, established Myc target genes, and Cdk4 are also commonly amplified and/or overexpressed in gliomas [1],[51]. We observed dMyc-dependent dCyclinD overexpression, and a requirement for dCyclinD-dCdk4 in repo>dEGFRλ;dp110CAAX neoplasia, although dCdk4 itself is not required for normal glial proliferation. Together with our analysis of TORC2, this illustrates that oncogenic EGFR-PI3K co-opts effectors that do not control normal glial development. Similarly, cdk4−/− mutant mice show normal proliferation in many tissues, but are resistant to ErbB-2-driven breast cancers [63],[64]. Our data argue that Cdk4 activity is a key tumor-specific rate-limiting output of EGFR and PI3K signaling in glioma as well.
In contrast to glia, coactivation of EGFR-Ras and PI3K in neuroblasts, which are fly neural stem cells, does not promote unchecked proliferation, despite the fact that neuroblasts express dMyc and are capable of undergoing neoplastic transformation in response to other genetic mutations [21]. Thus, in Drosophila, neither a neural stem cell fate nor Myc activity confer competence to undergo EGFR-PI3K neoplastic transformation. Rather, our results suggest that neoplastic cells arise from committed glial progenitors: dEGFR-dRas85D;dPTEN−/− clones derived from progenitor cells produce large tumors, and anaplastic cells in repo>dEGFRλ;dp110CAAX brains are concentrated in regions enriched for glial progenitors. Notably, regulated developmental signaling through the EGFR pathway promotes proliferation of normal Repo-expressing glial progenitors [27], and our results show that constitutive EGFR and PI3K signaling prolongs this proliferative progenitor state. Further studies of Drosophila glial progenitors and glioma-like cells may illuminate the cellular origins of human gliomas, which are thought to arise from progenitor-like glial cells. Moreover, our results argue that cell-type specific factors govern glial neoplasia. One such factor may be Dap, the single p21/p27 ortholog, which is normally expressed in only 5% of all glia (Figure S4). Perhaps glial progenitors do not express Dap, whereas neuronal progenitors do [26], and this underlies susceptibility to transformation by EGFR-Ras and PI3K. Dap is highly regulated in a cell-type specific manner [26], and studies of Dap regulation in glia may further illuminate the genetic origins of glioma, especially given that lack of p21 expression may underlie the tumorigenic response of mammalian glial progenitors to constitutively active EGFR [65].
While EGFR-Ras and PI3K are commonly upregulated in gliomas and experimental models demonstrate that these pathways are required for tumorigenesis, therapies that target EGFR and PI3K signaling have proven disappointing. This discrepancy between clinical and experimental data has many possible explanations. For example, recent studies have demonstrated that EGFR inhibitors are attenuated by particular mutations found in glioma cells, such as PTEN loss or RTK co-amplification [2]. Addressing these and other possibilities remains a challenge that dictates a need for new experimental models. The results presented here establish Drosophila as a viable model system for the study of glioma, offering a complex organismal system for rapidly identifying and evaluating therapeutic targets using genetic approaches. Such a system may be especially useful for distinguishing those genetic mutations and pathways that drive tumorigenesis from the large number of genes that show mutations and altered expression in glioblastomas uncovered by recent genomic analyses of patient samples [66],[67]. Our studies have already identified key rate-limiting genes, such as dCyclinE, Stg, and dMyc, and genes only required for abnormal neoplastic glial proliferation, such as dSin1, dRictor, and dCdk4, which may represent important therapeutic targets in human gliomas.
Flies were cultured at 25°C. All genotypes were established by standard genetics. To assess larval brain overgrowth phenotypes, embryos were collected for 6–24 hrs, grown for 120–140 hrs, and wandering 3rd instar larvae were selected for dissection.
Stocks were obtained from the Bloomington Stock Center unless otherwise noted. Other than UAS-PTENdsRNA lines from Bloomington, all UAS-dsRNA lines were obtained from the VDRC stock center [11]. The following stocks were obtained from other investigators: UAS-dEGFRλ (T. Schubach), UAS-dEGFRElp, UAS-dEGFRwild-type (N. Baker), UAS-dPTEN, FRT40A dPTEN2L117, UAS-dFoxOSA (S. Oldham), UAS-dap (I. Hariharan), UAS-dp110wild-type, UAS-dCycD, UAS-dCdk4, UAS-dMyc, dMyc4 (B. Edgar), UAS-Rbf1 (N. Dyson), appl-Gal4 (K. Finley), pros-Gal4 (B. Ohlstein), wor-Gal4 (C. Doe), gcm-Gal4 (V. Hartenstein), dTor2L7, dTorl(2)k17004 (R. Bodmer), dRictorΔ2 (S. Cohen), and stgCB03726 (A. Spradling)
UAS-dMycdsRNA, UAS-TSC1dsRNA lines were validated in prior publications [48]. UAS-dsRNA lines were crossed to actin-Gal4, ey-Gal4, and GMR-Gal4 to assess phenotypes. Lines that showed phenotypes inconsistent with known phenotypes for their target genes were excluded from analysis. Gene knock-down in repo-Gal4 glia was verified with immunohistochemical stains for the following constructs: UAS-dMycdsRNA, UAS-dAktdsRNA, UAS-dS6KdsRNA, UAS-eIF4EdsRNA, UAS-Rbf1dsRNA, and UAS-pntdsRNA (Figure S11).
Larval brains were dissected into sterile PBS, washed, and cut into fragments. Abdominal incisions were made in virgin female hosts and single brain fragments were inserted. Hosts were cultured for 1–6 weeks, dissected and fixed in 4% paraformaldehyde, incubated in 10% sucrose and embedded in O.C.T. Thick 50 µm sections were stained as described below.
For hs-FLP clones, genotypes are indicated in figure legends. Flies were initially grown at 18°C or 20°C to minimize spontaneous clones, which occurred at a low frequency during late larval-pupal stages. 3rd instar larvae, 0–48 hr pupae, or 0–2 day old young adults were treated with heat shock to induce clones and subsequently cultured at 25°C for 1–4 weeks. For ey-FLP clones, flies were cultured at 25°C continuously.
Larval tissue was fixed for 30–50 minutes in 1×PBS 4% paraformaldehyde. Adult brains were fixed for 1–2 hr in 1×PBS 4% paraformaldehyde or in PLP with 2% paraformaldehyde. For BrdU labeling, larvae were cultured in food with 1 mg/ml BrdU for 4–6 hrs, and fixed larval brains were treated with 2 N HCl for 30 minutes followed by DNase for 1 hr. Stains were performed in 1×PBS 10% BSA with 0.3% Triton-X100 for larval brains and 0.5% Triton-X100 for adult samples.
The following antibodies were obtained from the Developmental Studies Hybridoma Bank and diluted 1∶5–1∶10: 8D12 anti-Repo, anti-dMMP1, anti-dCyclinB, anti-Elav, and 40-1a anti-lacZ. Larval and/or adult brains were also stained with rabbit anti-Repo (G. Technau, 1∶500), rat anti-dCyclinE (H. Richardson, 1∶100), anti-BrdU (BD, 1∶100), rat anti-Miranda (C. Doe, 1∶100), mouse anti-diphospho-Erk (Sigma, 1∶200), mouse anti-Rbf1 (N. Dyson, 1∶5), mouse anti-Dap (I. Hariharan, 1∶10), rabbit anti-PntP1 (J. Skeath, 1∶500), rabbit anti-eIF4E (P. Lasko, 1∶100), rabbit anti-dMyc (D. Stein, 1∶1000), and anti-phospho-MPM2 (Upstate Biotechnology, 1∶200). Anti-HRP-Cy5 and anti-HRP-Cy3 (Jackson Labs) were used at 1∶250–1∶500. Secondary antibodies were conjugated to Cy3 (Jackson Labs) or Alexa-488 or Alexa-647 (Molecular Probes). Actin was visualized with Rhodamine-labeled phalloidin (Invitrogen).
Brains were imaged as whole mounts on a Zeiss LSM 510 confocal system. Images were analyzed in Zeiss LSM Image Browser and processed in Photoshop CS3. For experiments in which protein levels were compared between genotypes, all sample preparation, histochemistry, imaging, and image processing was performed in parallel in the same manner.
|
10.1371/journal.pcbi.0030228 | A Dynamical Systems Hypothesis of Schizophrenia | We propose a top-down approach to the symptoms of schizophrenia based on a statistical dynamical framework. We show that a reduced depth in the basins of attraction of cortical attractor states destabilizes the activity at the network level due to the constant statistical fluctuations caused by the stochastic spiking of neurons. In integrate-and-fire network simulations, a decrease in the NMDA receptor conductances, which reduces the depth of the attractor basins, decreases the stability of short-term memory states and increases distractibility. The cognitive symptoms of schizophrenia such as distractibility, working memory deficits, or poor attention could be caused by this instability of attractor states in prefrontal cortical networks. Lower firing rates are also produced, and in the orbitofrontal and anterior cingulate cortex could account for the negative symptoms, including a reduction of emotions. Decreasing the GABA as well as the NMDA conductances produces not only switches between the attractor states, but also jumps from spontaneous activity into one of the attractors. We relate this to the positive symptoms of schizophrenia, including delusions, paranoia, and hallucinations, which may arise because the basins of attraction are shallow and there is instability in temporal lobe semantic memory networks, leading thoughts to move too freely round the attractor energy landscape.
| One of the hallmarks of schizophrenia is the complexity and heterogeneity of the illness. We propose that part of the reason for the inconsistent symptoms may be a reduced signal-to-noise ratio and increased statistical fluctuations in different cortical brain networks. The novelty of the approach described here is that instead of basing our hypothesis purely on biological mechanisms, we develop a top-down approach based on the different types of symptoms and relate them to instabilities in attractor neural networks. Schizophrenia is characterized by cognitive, negative, and positive symptoms. We propose which characteristic effects in a dynamical system could cause these symptoms, and investigate our hypothesis in a computational model. We implement an integrate-and-fire network model and focus on the alterations of synaptic channels activated via NMDA and GABA receptors. We found that a decrease in the NMDA receptor conductance could contribute to both the cognitive and negative symptoms by reducing the neuronal firing rates and the stability of the attractor states. A reduction of both NMDA and GABA conductance causes an instability of the attractor states related to the positive symptoms. Overall, we provide a framework to discuss schizophrenia in a dynamical system framework.
| Schizophrenia is a major mental illness, which has a great impact on patients and their environment. One of the difficulties in proposing models for schizophrenia is the complexity and heterogeneity of the illness. We propose that part of the reason for the inconsistent symptoms may be a reduced signal-to-noise ratio and increased statistical fluctuations in different cortical brain networks. The novelty of the approach described here is that instead of basing our hypothesis purely on biological mechanisms, we develop a top-down approach based on the different types of symptoms and relate them to instabilities in attractor neural networks [1].
The main assumption of our hypothesis is that attractor dynamics are important in cognitive processes [2]. Our hypothesis is based on the concept of attractor dynamics in a network of interconnected neurons that in their associatively modified synaptic connections store a set of patterns, which could be memories, perceptual representations, or thoughts [3–5]. The attractor states are important in cognitive processes such as short-term memory, attention, and action selection [6]. The network may be in a state of spontaneous activity, or one set of neurons may have a high firing rate, each set representing a different memory state, normally recalled in response to a retrieval stimulus. Each of the states is an attractor in the sense that retrieval stimuli cause the network to fall into the closest attractor state, and thus to recall a complete memory in response to a partial or incorrect cue. Each attractor state can produce stable and continuing or persistent firing of the relevant neurons. The concept of an energy landscape [4] is that each pattern has a basin of attraction, and each is stable if the basins are far apart and also if each basin is deep, which is caused, for example, by high firing rates and strong synaptic connections between the neurons representing each pattern, which together make the attractor state resistant to distraction by a different stimulus. The spontaneous firing state, before a retrieval cue is applied, should also be stable. Noise in the network caused by statistical fluctuations in the stochastic spiking of different neurons can contribute to making the network transition from one state to another; we take this into account by performing integrate-and-fire simulations with spiking activity, and relate this to the concept of an altered signal-to-noise ratio in schizophrenia [7–9].
Schizophrenia is characterized by three main types of symptom: cognitive dysfunction, negative symptoms, and positive symptoms [10–12]. We consider how the basic characteristics of these three categories might be produced in a neurodynamical system, as follows. Dysfunction of working memory, the core of the cognitive symptoms, may be related to instabilities of persistent attractor states [13,14], which we show can be produced by reduced firing rates in attractor networks in brain regions such as the prefrontal cortex. The negative symptoms such as flattening of affect or reduction of emotions may be caused by a consistent reduction in firing rates of neurons in regions associated with emotion, such as the orbitofrontal cortex [1]. These hypotheses are supported by the frequently observed hypofrontality, a reduced activity in frontal brain regions in patients with schizophrenia during cognitive tasks [15–17]. The positive symptoms are characterized by phenomenologically overactive perceptions or thoughts such as hallucinations or delusions, which are reflected, for example, by higher activity in the temporal lobes [17,18]. We relate this category of symptoms to a spontaneous appearance of activity in attractor networks in the brain and more generally to instability of both the spontaneous and persistent attractor states. We do not try to account for every detail of schizophrenic symptoms, which are diverse and vary among patients, but instead show how these three main categories of symptoms could be related to changes in the stability of dynamical attractor systems in the brain, and how the changes in the stability might be produced by changes at the level of the currents being passed through receptor-activated ion channels at synapses. There are specific symptoms such as aberrant eye movements that cannot be accounted for by this general scheme.
In particular, we were interested in how these symptoms are related. Negative and cognitive symptoms typically precede the first psychotic episode [19,20]. Positive symptoms can be treated in most cases with neuroleptics, whereas negative and cognitive symptoms persist, at least for typical neuroleptics. Can a mapping onto a dynamical system help to understand these relations? After proposing a dynamical systems hypothesis for the different symptoms of schizophrenia, we study a standard neural network model [21] of cortical dynamics specifically in relation to our hypothesis. We were especially interested in how excitation and inhibition implemented by NMDA and GABA synapses affect the network dynamics. Alterations in the efficacies of the NMDA and GABA channels have been identified in the pathology of schizophrenia [22,23], and transmitters such as dopamine influence the currents in these receptor-activated channels [24]. Do NMDA and GABA currents have antagonistic effects, or do they have a special role in the network dynamics? How could this be related to our hypothesis of schizophrenia? Building upon the current body of neural network research, we specifically conduct neural network simulations to substantiate our dynamical systems hypothesis of schizophrenia. While focussing on NMDA and GABA synapses in the Results, in the Discussion we consider how altered transmission at D1 and D2 receptors by modulating NMDA and GABA conductances could not only influence working memory, which has been investigated previously [21,25–27], but could in particular influence the different symptoms of schizophrenia.
We relate the three types of symptoms of schizophrenia to the dynamical systems attractor framework described at the end of the Introduction as follows [1].
The cognitive symptoms of schizophrenia include distractibility, poor attention, and the dysexecutive syndrome [10–12,28]. The core of the cognitive symptoms is a working memory deficit in which there is a difficulty in maintaining items in short-term memory [29,30], which could directly or indirectly account for a wide range of the cognitive symptoms. We propose that these symptoms may be related to instabilities of persistent states in attractor neural networks, consistent with the body of theoretical research on network models of working memory [13]. The neurons are firing at a lower rate, leading to shallower basins of attraction of the persistent states, and thus a difficulty in maintaining a stable short-term memory, normally the source of the bias in biased competition models of attention [5,6]. The shallower basins of attraction would thus result in working memory deficits, poor attention, distractibility, and problems with executive function and action selection [6,31].
The negative symptoms refer to the flattening of affect and a reduction in emotion. Behavioural indicators are blunted affect, emotional and passive withdrawal, poor rapport, lack of spontaneity, motor retardation, and disturbance of volition [10–12]. We propose that these symptoms are related to decreases in firing rates in the orbitofrontal cortex and/or anterior cingulate cortex [1], where neuronal firing rates and activations in functional MRI investigations are correlated with reward value and pleasure [1,32]. Consistent with this, imaging studies have identified a relationship between negative symptoms and prefrontal hypometabolism (i.e., a reduced activation of frontal areas [33,34]).
The positive symptoms of schizophrenia include bizarre (psychotic) trains of thought, hallucinations, and (paranoid) delusions [10–12]. We propose that these symptoms are related to shallow basins of attraction of both the spontaneous and persistent states in the temporal lobe semantic memory networks and to the statistical fluctuations caused by the probabilistic spiking of the neurons. This could result in activations arising spontaneously and thoughts moving too freely around the energy landscape, loosely from thought to weakly associated thought, leading to bizarre thoughts and associations that may eventually over time be associated together in semantic memory to lead to false beliefs and delusions. Consistent with this, neuroimaging studies suggest higher activation, especially in areas of the temporal lobe [17,18,35].
To further investigate our hypothesis, we use an attractor network, as this is likely to be implemented in many parts of the cerebral cortex by the recurrent collateral connections between pyramidal cells, and has short-term memory properties with basins of attraction that allow systematic investigation of stability and distractibility. The particular neural network implementation we adopt includes channels activated by AMPA, NMDA, and GABAA receptors, and allows not only the spiking activity to be simulated, but also a consistent mean-field approach to be used [21].
First, we introduce an analytical approach to the concepts of how changes in transmitters could affect the depth of the basins of attraction in networks in ways that may be related to the symptoms of schizophrenia.
The depth of the basins of attraction can be assessed by calculating the flow toward the attractors using the mean-field reduction (see Methods and [36]). The flow is the force that draws the dynamical system to one of the attractors. Figure 1 shows the flow between the spontaneous and persistent state in a network featuring one selective pool. The curve for (Normal) shows the flow for the reference baseline condition. The crossing of the curve and the 0-flow axis represent the fixed points, which are either stable (negative derivative) or unstable (positive derivative). We use this line as a reference to assess the relative depth of the energy landscape. The system flows either into a spontaneous firing state (of approximately 2 Hz), or a high-firing attractor state.
A reduction of NMDA (−NMDA) shows a stronger flow than the unchanged condition at low firing rates toward the spontaneous attractor (at about 2 Hz). The absolute values of the function are higher compared with the normal condition until the first unstable fixed point (at around 6–7 Hz). The basin of attraction toward the persistent attractor at high firing rates yields the reverse picture. Here, the −NMDA curve is clearly below the unchanged condition, and the flow toward the attractor is smaller. Overall, the basin of attraction is deeper for the spontaneous state and shallower for the persistent state compared with the unchanged condition. This pattern fits with the cognitive symptoms of schizophrenia as proposed in our hypothesis. Also, note that the firing rate of the persistent fixed point is reduced in the −NMDA condition (crossing with the flow = 0-axis), which is consistent with the hypothesis for the negative symptoms.
A reduction of the GABA conductance (−GABA) yields the opposite pattern to that in the reduced NMDA condition. Here, the basin of attraction of the persistent state is deeper. This is not a condition that we suggest is related to the symptoms of schizophrenia.
However, in the condition in which both the NMDA and the GABA conductances are reduced (−NMDA, −GABA), the persistent-state basin of attraction is shallower, and the spontaneous-state basin is a little shallower. This condition corresponds to the proposed landscape for the positive symptoms as considered above. In particular, in the −NMDA, −GABA condition, the system would be less stable in the persistent state, tending to move to another attractor easily, and less stable in the spontaneous state, also tending to move too readily into an attractor from spontaneous activity.
Overall, the flow analysis suggests that both the cognitive and negative symptoms could be related to a decrease in the NMDA conductances. This is consistent with the fact that these two symptoms usually appear together. The flow analysis suggests that the positive symptoms are related to a reduction in both NMDA and GABA. Thus, the transition from the cognitive and negative symptoms to the positive, psychotic symptoms might be caused by an additional decrease in the GABA conductance. It is notable that excitation and inhibition do not cancel each other out as assumed by many models, but have distinct influences on the network dynamics.
The flow analysis provides insight into how the depth of the basins of attraction and the firing rates are influenced by changes in the conductivities of the channels activated via NMDA and GABA receptors. However, the overall stability of the different attractors is affected not only by the depth of the basins of attraction, but also by the breadth and distance apart of the basins and by the statistical noise generated by the randomness of the spiking of the different neurons. These statistical fluctuations play a role in the way in which the system moves from one state to another, for these statistical fluctuations can cause hills in the energy landscape to be crossed stochastically. Since the mean-field analyses do not take these properties into account, we investigate the system further using large-scale integrate-and-fire network simulations and measuring the statistics of the network behavior.
To clarify the concept of stability, we show examples of trials of spontaneous and persistent simulations in which the statistical fluctuations have different effects on the temporal dynamics. Figure 2 shows the possibilities, as follows.
In the spontaneous-state simulations, no cue is applied, and we are interested in whether the network remains stably in the spontaneous firing state, or whether it is unstable and, on some trials due to statistical fluctuations, enters one of the attractors, thus falsely retrieving a memory. Figure 2A shows an example of a trial on which the network correctly stays in the low spontaneous firing rate regime, and another trial (labelled spontaneous unstable) in which statistical spiking-related fluctuations in the network cause it to enter a high-activity state, moving into one of the attractors even without a stimulus.
In the persistent-state simulations, a strong excitatory input is given to the S1 neuronal population between 0 and 500 ms (see Analysis section). Two such trials are shown in Figure 2B. In one, the S1 neurons (correctly) keep firing at approximately 30 Hz after the retrieval cue is removed at 500 ms. However, due to statistical fluctuations in the network related to the spiking activity, on the trial labelled persistent unstable, the high firing rate in the attractor for S1 was not stable, and the firing decreased back toward the spontaneous level, in the example shown starting after 1.5 s. This trial illustrates the failure to maintain a stable short-term memory state, even when no distractor is applied.
In Figure 2, the transitions to the incorrect activity states are caused by statistical fluctuations in the spiking activity of the integrate-and-fire neurons and the depth of the basins of attraction, which has been reduced in the simulations shown by reducing both the NMDA and the GABA currents, as indicated in the caption. We hypothesize that such instabilities are related to the symptoms of schizophrenia. We note that there are two sources of noise in the spiking networks that cause the statistical fluctuations: the randomly arriving external Poisson spike trains, and the statistical fluctuations caused by the spiking of the neurons in the finite sized network. The magnitude of these fluctuations increases as the number of neurons in the network becomes smaller [37].
For our investigations, we selected w+ = 2.1, which with the default values of the NMDA and GABA conductances yielded stable dynamics; that is, a stable spontaneous state if no retrieval cue was applied, and a stable state of persistent firing after a retrieval cue had been applied and removed. To investigate the effects of changes (modulations) in the NMDA and GABA conductances, we chose for demonstration purposes a reduction of 4.5% and 9%, respectively, as these could cause instabilities, as illustrated in Figure 2. However, the exact values are not crucial to observe the effects described. The magnitudes of these reductions are smaller than those that can be produced experimentally [24,26]. A strength of our approach is that we show that even quite small reductions in the synaptic currents can alter the global behaviour of the network, e.g., the stability of its attractors.
We assessed how the stability of both the spontaneous and persistent states changes when NMDA and GABA efficacies are modulated. Specifically, we ran multiple-trial integrate-and-fire network simulations and counted how often the system maintained the spontaneous or persistent state, assessed by the firing rate in the last second of the simulation (2–3 s) of each 3-s trial. Figure 3 shows the stability of the spontaneous and persistent attractors relative to the unmodulated reference state (Normal). A negative percentage means that the system was less stable than in the unmodulated state.
A reduction of the NMDA conductance (−NMDA) reduces the stability of the persistent state drastically, while slightly increasing the stability of the spontaneous state (see Figure 3). We hypothesized that this type of change might be related to the cognitive symptoms, since it shows a reduced stability of the working memory properties. A reduction of GABA shows the opposite pattern: a slight reduction in the stability of the spontaneous state, and an increased stability of the persistent (i.e., attractor) state (see Figure 3).
When both NMDA and GABA are reduced, one might think that these two counterbalancing effects (excitatory and inhibitory) would either cancel each other out or yield a tradeoff between the stability of the spontaneous and persistent states. However, this is not the case. The stability of both the spontaneous state and the persistent state is reduced (see Figure 3). We relate this pattern to the positive symptoms of schizophrenia, in which both the spontaneous and attractor states are shallow, and the system merely jumps by the influence of statistical fluctuations between the different (spontaneous and attractor) states.
To investigate more directly the wandering between spontaneous and several different persistent attractor states, we simulated the condition with decreased NMDA and GABA conductances over a long time period in which no cue stimulus input was given. Figure 4 shows the firing rates of the two selective pools S1 and S2. The high activity switches between the two attractors due to the influence of fluctuations, which corresponds to spontaneous wandering in a shallow energy landscape, corresponding, for example, to sudden jumps between unrelated cognitive processes. These results are consistent with the flow analysis and demonstrate that the changes in the attractor landscape influence the behavior at the stochastic level.
As distractibility is directly related to the symptoms of schizophrenia, we ran simulations specifically to assess this property using persistent and distractor simulations (see Analysis section). A distractor strength of 0 Hz corresponds to the persistent condition described in the preceding section (Stability). Figure 5 shows the stability and distractibility for reductions of NMDA and GABA currents. The reference state is labelled “Normal.” In this state, pool S1 continued to maintain its attractor firing without any distractor (distractor strength = 0 Hz) throughout the delay period on almost 90% of the trials. In both conditions that reduce the NMDA current (labelled −NMDA), the network was less and less able to maintain the S1 attractor firing as the distractor stimulus strength was increased through the range of 0–80 Hz. The stability of the persistent state was reduced, and the distractibility was also increased, as shown by the fact that increasing distractor currents applied to S2 could move the attractor away from S1. The implication, therefore, is that a reduction of the NMDA currents could cause the cognitive symptoms of schizophrenia by making short-term memory networks less stable and more distractible, thereby reducing the ability to maintain attention. Reducing only the GABA currents (−GABA) reduces the distractibility for low distractor strengths and coincides with the reference (Normal) condition at high values of the distractor strengths.
We further investigated the signal-to-noise ratio in relation to the changes in synaptic conductances. The signal-to-noise ratio denotes the level of a signal relative to the level of background noise. In an attractor network, a high signal-to-noise ratio indicates that the network will maintain the attractor stably, as it will be unlikely to be disrupted by spiking-related statistical fluctuations that are the source of the noise in the network. Figure 6 shows the signal-to-noise ratio of a measure related to the functional MRI blood oxygenation level–dependent signal. (This measure described in the caption to Figure 6 and below was used because the experimental data with which we wish to compare the simulation results use functional MRI measures [7–9]. The index we used of the activity of the network was the total synaptic current of selective pool 1 averaged over the whole simulation time of 3 s to take the temporal filtering properties of the blood oxygenation level–dependent signal into account, given the typical time course which lasts for several seconds of the haemodynamic response function [38]. Further, we subtracted the averages of the spontaneous trial simulations that represent the baseline activity from the persistent trial simulation values. The signal-to-noise ratio was calculated from the mean of this index across trials divided by the standard deviation of the index, both measured using 1,000 simulation trials. If the network sometimes had high activity, and sometimes low, then the signal-to-noise measure gave a low value. If the network reliably stayed in the high persistent firing states, then the signal-to-noise ratio measure was high.) As shown in Figure 6, we found that in all the cases in which the NMDA or the GABA conductance, or both, are reduced, the signal-to-noise ratio, computed by the mean divided by the standard deviation, is also reduced. This relates to recent experimental observations which show a decreased signal-to-noise ratio in schizophrenic patients [7–9]. Here, we directly relate a decrease in the signal-to-noise ratio to changes (in this case, decreases) in receptor-activated synaptic channel conductances. Given these results, it would be of interest in future studies to model the exact paradigm used by Winterer et al. [8].
We have proposed a hypothesis that relates the cognitive, negative, and positive symptoms of schizophrenia [10–12] to the depth of basins of attraction and to the stability properties of attractor networks caused by statistical fluctuations of spiking neurons. This assumes that some cognitive processes can be understood as dynamical attractor systems, which is an established hypothesis in areas such as working memory, but has also been used in many other areas [2,5]. Our approach applies this concept to mental illnesses [39]. Due to the diversity of schizophrenic symptoms, our general hypothesis is meant to serve as a heuristic of how the different kinds of symptoms might arise and are related. We investigated the hypothesis empirically in a computational attractor framework to capture an important aspect of cortical functionality. Figure 7 summarizes our hypothesis and its relation to the investigations of an attractor neural network.
The middle column in Figure 7 shows the overview for the cognitive and negative symptoms. The core of the cognitive symptoms is a failure of working memory and attentional mechanisms. Working memory activity is related to the ongoing (i.e., persistent) firing of neurons during the delay period of cognitive tasks [29,30]. This could be implemented by associatively modifiable synapses between the recurrent collateral synapses of cortical pyramidal cells [13,14,40,41]. We propose that the cognitive symptoms of schizophrenia could arise because the basins of attraction of the persistent states in the prefrontal cortex become too shallow. This leads in combination with the statistical fluctuations due to randomness of the spiking activity to either a fallout of an active working memory state or to a shift to a different attractor state, leading to a failure to maintain attention and thereby impairing executive function. The hypofrontality in schizophrenia, that is, less activation in frontal brain regions during working memory tasks [15,42], is in line with our hypothesis, since the firing rates of the persistent state are lower in the reduced NMDA condition (Figure 1), and the system spends on average less time in the persistent state, since it is less stable than in the normal condition (Figure 3). In addition, a reduced signal-to-noise ratio as shown in our simulations (Figure 6) has also been identified in imaging studies [7–9]. Our simulations suggest that a reduction in NMDA conductance at the synaptic level (see Figure 7) can account for this phenomenon. This is in line with previous work on the stability of working memory networks [14,27,43].
A reduction of the NMDA conductance also results in a reduction of the firing rates of the neurons in the persistent state (see Figure 1 and [21]). We relate this, following [1], to the negative symptoms, which include flattening of affect, a reduction in emotion, emotional and social withdrawal, poor rapport, passive withdrawal, lack of spontaneity, motor retardation, apathy, and disturbance of motivation. These symptoms are related to decreases in activity in the orbitofrontal cortex and/or anterior cingulate cortex [33,34], both of which are implicated in emotion [1,32,44]. The emotional states represented in the orbitofrontal cortex and anterior cingulate cortex include states elicited both by rewards and punishers. Our hypothesis is that both would be reduced by the mechanism described. Correspondingly, motivation would be reduced in the same way, in that motivation is a state in which we work to obtain goals (rewards) or avoid punishers [1].
Both the negative and cognitive symptoms thus could be caused by a reduction of the NMDA conductance in attractor networks. The proposed mechanism links the cognitive and negative symptoms of schizophrenia in an attractor framework and is consistent with a close relation between the cognitive and negative symptoms: blockade of NMDA receptors by dissociative anesthetics such as ketamine produces in healthy subjects schizophrenic symptoms, including both negative and cognitive impairments [45,46]; agents that enhance NMDA receptor function reduce the negative symptoms and improve the cognitive abilities of schizophrenic patients [47]; and the cognitive and negative symptoms occur early in the illness and precede the first episode of positive symptoms [12,19,20]. Consistent with this hypothesized role of a reduction in NMDA conductances being involved in schizophrenia, postmortem studies of schizophrenia have identified abnormalities in glutamate receptor density in regions such as the prefrontal cortex, thalamus, and the temporal lobe [22,47], brain areas that are active during the performance of cognitive tasks.
The dopamine D1 receptor has been shown to modulate the performance of working memory tasks [30,48–50]. An increase in D1 receptor activation has been shown to increase the NMDA current [24,26], and modeling studies have shown that this increase is related to the stability of working memory states [21,25,27]. Imaging data also support the importance of the D1 receptor in schizophrenia [51,52]. We therefore suggest that an increased activation of D1 receptors might alleviate both the cognitive and the negative symptoms of schizophrenia [53,54] by increasing NMDA receptor-mediated synaptic currents (Figure 7). Atypical neuroleptics might use this mechanism by not only blocking D2 receptors, but also by increasing the presynaptic release of dopamine, which in turn would increase the activation of the extrasynaptic D1 receptors [48,55].
Taken together, we suggest that the cognitive and negative symptoms could be caused by the same synaptic mechanism, namely a reduction in the NMDA conductance, which reduces the stability and increases the distractibility of the persistent attractors, and reduces the activity (firing rates) of neurons (Figure 7, middle column). The reduced depth of the basins of attraction can be understood in the following way. Hopfield [4] showed that the recall state in an attractor network can be thought of as the local minimum in an energy landscape, where the energy would be defined as
where yi and yj are the firing rates of the ith and jth neurons in the network, which are connected by synaptic weight wij. In general, neuronal systems do not admit an energy function. Nevertheless, we can assume an effective energy function: in fact, the flow picture shown in Figure 1 resulting from the mean-field reduction associated with the spiking network analyzed here can be viewed as an indirect description of an underlying effective energy function. From this equation, it follows that the depth of a basin of attraction is deeper if the firing rates are higher and if the synaptic strengths that couple the neurons that are part of the same attractor are strong. (The negative sign results in a low energy, and thus a stable state, if the firing rates of the neurons in the same attractor and their synaptic coupling weights are high.) If we reduce the NMDA receptor–activated channel conductances, then the depth of the basins of attraction will be reduced both because the firing rates are reduced by reducing excitatory inputs to the neurons, and because the synaptic coupling weights are effectively reduced because the synapses can pass only reduced currents.
The positive symptoms (Figure 7, right column) of schizophrenia include delusions, hallucinations, thought disorder, and bizarre behavior. Examples of delusions are beliefs that others are trying to harm the person, impressions that others control the person's thoughts, and delusions of grandeur. Hallucinations are perceptual experiences that are not shared by others and are frequently auditory, but can affect any sensory modality. These symptoms may be related to activity in the temporal lobes [11,12,56]. The attractor framework approach taken in this paper hypothesizes that the basins of attraction of both spontaneous and persistent states are shallow (Figure 7). Due to the shallowness of the spontaneous state, the system can jump spontaneously up to a high activity state, causing hallucinations to arise and leading to bizarre thoughts and associations. This might be the cause for the higher activations in temporal lobe areas that are identified in imaging experiments [17,18].
We relate the positive symptoms to not only a reduction in NMDA conductance, but also to a reduction in GABA conductance. This is consistent with the fact that the positive symptoms usually follow the cognitive and negative symptoms and represent a qualitative worsening of the illness [12]. Alterations in GABA receptors have been identified in schizophrenia [23,57].
D2 receptor antagonism remains a main target for antipsychotics [58,59]. Dopamine receptor D2 antagonists mainly alleviate the positive symptoms of schizophrenia, whereas the cognitive and negative symptoms persist, especially for the typical neuroleptics [12]. Together with the simulations, our hypothesis suggests that an increase in the GABA current in the state corresponding to the positive symptoms (−NMDA, −GABA) might have the same effect as D2 antagonists. The therapeutic effect of D2 antagonists might thus be caused by an increase in GABA currents. Indeed, it has been found that D2 receptors decrease the efficacy of the GABA system [60,61]. (For example, the application of D2 antagonists prevented a decrease in evoked inhibitory postsynaptic current amplitude produced by dopamine [60].) Thus, D2 antagonists would, in a hypersensitive D2 receptor state [62,63], increase GABA inhibition in the network, and we suggest could increase the stability of attractor networks involved in the positive symptoms of schizophrenia, and thus ameliorate the positive symptoms. Since the concentration of dopamine in the cortex depends on cortical–subcortical interactions [64], the causes of the described changes could also result from subcortical deficits. A detailed analysis of these feedback loops would require specific modeling.
Earlier accounts of the relation of dopamine and schizophrenia in the cortex [24,60] have suggested two distinct states of dopamine modulation. One is a D2 receptor–dominated state in which there is weak gating and in which information can easily affect network activity. The other is a D1 receptor–dominated state in which network activity is stable and maintained. We have proposed a more detailed account for stability and discussed this separately for the spontaneous and persistent attractor states. This allows us to account for the dichotomy between the cognitive/negative and positive symptoms. We emphasize that in biophysically realistic network simulations, excitation and inhibition are not merely antagonistic but implement different functions in the network dynamics. Thereby, our modeling approach provides a missing link between the symptoms of schizophrenia and network models of working memory and dopamine [21,25,27]. This approach is not meant to provide a detailed discussion of specific symptoms of schizophrenia. Further research would be needed to relate specific symptoms to the overall scheme presented here in this paper.
We concentrated in our study on basic effects in local cortical dynamics based on biophysically realistic spiking networks. One can extend the model to study the interaction between large-scale networks involving more than one location in the cortex and/or subcortical regions. This increases the dimensionality of the system and adds to the issues of stability [65,66]. Processes such as bubbling transitions and chaotic attractors also add to the notion of stability [67,68]. For high-dimensional systems, Kaneko uses the concept of a Milnor attractor [69], which offers a way to formally describe the stability of an attractor. This approach measures the stability of an attractor against perturbations by introducing the concept of “return probability.” By sampling over random perturbation and orbit positions, the return probability is defined as the fraction of trials in which the system returns back to the original point. This measures the strength of the attractor. In our work, we wanted to stay close to paradigms of working memory and address mechanisms involved in cognitive processes, attention, and distractibility. Consequently, our quantitative measures of stability are intrinsically related to these paradigms and therefore use concepts such as the escape probability instead of the return probability. Indeed, we introduced these measures in order to relate the quantitative concept of stability more directly with the cognitive symptoms. Nevertheless, in detailed investigations of specific symptoms of schizophrenia, the concepts mentioned above should be kept in mind, as they might play a role in neural dynamics.
Cohen and collaborators focused in a series of computational and experimental studies on the cognitive symptoms of schizophrenia. Their connectionist models try to account for specific experimental data on context-processing deficits using working memory, gating, and attentional selection as mechanisms [70–72]. The context-processing hypothesis is compatible with the hypothesis of working memory deficits in schizophrenia, as context processing relies on stable working memory: a deficit in working memory would consequently lead to a deficit in context processing. Our model is conceptually at another level, featuring biophysically plausible single-neuron properties and specific synaptic currents. It is not intended to account for specific experimental data, but provides insights at a more generic and also biological level. In this sense, the two models complement each other.
We discussed a possible cause for the proposed alterations of the attractor landscape related to schizophrenia, namely changes in NMDA and GABA conductance, as these are directly related to schizophrenia [22,23]. We did not investigate changes in AMPA conductance. In this particular model, the contribution of the AMPA current is relatively small [21]. A more detailed investigation could also include AMPA conductance, especially because it is known to be influenced by NMDA synaptic plasticity [73]. Indeed, if reduced NMDA currents led in turn by synaptic plasticity to reduced AMPA currents, this would amplify the effects we describe. The proposed alterations in the attractor landscape could have a variety of causes at the neurobiological level: abnormalities in glutamate and GABA receptors and signaling, modulations in synaptic plasticity, aberrant dopamine signaling, reduced neuropil, genetic mechanisms, and brain volume reduction [12,30,74,75]. Besides cortical mechanisms, cortical–subcortical dynamics could also cause the proposed alterations in the cortical attractor landscape, for example, via neuromodulatory influences such as dopamine or serotonin or cortical–subcortical feedback loops [64,76]. Our general hypothesis regarding the attractor landscape is meant to describe the aberrant dynamics in cortical regions that could be caused by several pathways. Future work could analyze further how changes of different factors such as regional differences, subcortical–cortical networks, or even more detailed neural and synaptic models might influence the stability of the type of neurodynamical system described here. We envision that our hypothesis could serve as a useful guideline for further theoretical work.
Our aim is to investigate stability and distractibility in a biophysically realistic attractor framework so that the properties of receptors, synaptic currents, and the statistical effects related to the probabilistic spiking of the neurons can be part of the model. We use a minimal architecture, a single-attractor or autoassociation network [3–5,41,77]. We chose a recurrent (attractor) integrate-and-fire network model which includes synaptic channels for AMPA, NMDA, and GABAA receptors [21]. These synaptic receptors are important, since there is evidence that alterations in synaptic currents are related to the symptoms of schizophrenia, and many of the drugs used to treat schizophrenia act on these receptor-mediated currents, either directly or indirectly [22, 23].
Both excitatory and inhibitory neurons are represented by a leaky integrate-and-fire model [78]. The basic state variable of a single-model neuron is the membrane potential. It decays in time when the neurons receive no synaptic input down to a resting potential. When synaptic input causes the membrane potential to reach a threshold, a spike is emitted and the neuron is set to the reset potential at which it is kept for the refractory period. The emitted action potential is propagated to the other neurons in the network. The excitatory neurons transmit their action potentials via the glutamatergic receptors AMPA and NMDA, which are both modeled by their effect in producing exponentially decaying currents in the postsynaptic neuron. The rise time of the AMPA current is neglected because it is typically very short. The NMDA channel is modeled with an alpha function, including both a rise and a decay term. In addition, the synaptic function of the NMDA current includes a voltage dependence controlled by the extracellular magnesium concentration [79]. The inhibitory postsynaptic potential is mediated by a GABAA receptor model and is described by a decay term.
The single-attractor network contains 400 excitatory and 100 inhibitory neurons, which is consistent with the observed proportions of pyramidal cells and interneurons in the cerebral cortex [80,81]. The connection strengths are adjusted using mean-field analysis [21] so that the excitatory and inhibitory neurons exhibit a spontaneous activity of 3 Hz and 9 Hz, respectively [82,83]. The recurrent excitation mediated by the AMPA and NMDA receptors is dominated by the NMDA current to avoid instabilities during the delay periods [84].
Our cortical network model features a minimal architecture to investigate stability and distractibility, and consists of two selective pools, S1 and S2 (Figure 8). We use just two selective pools to eliminate possible disturbing factors. The nonselective pool NS models the spiking of cortical neurons and serves to generate an approximately Poisson spiking dynamic in the model [21], which is what is observed in the cortex. The inhibitory pool IH contains the 100 inhibitory neurons. The connection weights between the neurons of each pool or population are called the intrapool connection strengths w+. The increased strength of the intrapool connections is counterbalanced by the other excitatory connections (w−) to keep the average input constant.
The network receives Poisson input spikes via AMPA receptors that are envisioned to originate from 800 external neurons at an average spontaneous firing rate of 3 Hz from each external neuron, consistent with the spontaneous activity observed in the cerebral cortex [41,83]. A detailed mathematical description is provided in Protocol S1.
Our analysis is targeted to investigate the stability and distractibility with respect to NMDA and GABA receptor modulations. We use two different techniques: multiple trial spiking simulations and mean-field simulations. Spiking trial simulations integrate the complete neural and synaptic dynamics over time, including statistical components of the network model. Therefore, the spiking simulations are needed to assess the stability and distractibility of the dynamical system, for this depends in part on the statistical fluctuations that occur in a network of spiking neurons [85]. This is done by simulating a network configuration for several trials, each run with different random seeds, and running a statistical analysis on the data.
We simulate three different conditions: the spontaneous, persistent, and distractor conditions (see Figure 9).
In spontaneous simulations, we run spiking simulations for 3 s without any extra external input. The aim of this condition is to test whether the network is stable in maintaining a low average firing rate in the absence of any inputs, or whether it falls into one of its attractor states without any external input.
In persistent simulations, an external cue of 120 Hz above the background firing rate of 2,400 Hz is applied to each neuron in pool S1 during the first 500 ms to induce a high-activity state, and then the system is run for another 2.5 s. The 2,400 Hz is distributed across the 800 synapses of each S1 neuron for the external inputs, with the spontaneous Poisson spike trains received by each synapse, thus having a mean rate of 3 Hz. The aim of this condition is to investigate whether the network, once in an attractor short-term memory state, can maintain its activity stably, or whether it falls out of its attractor, which might correspond to an inability to maintain attention.
The distractor simulations start off like the persistent simulations with a 500-ms input to pool S1 to start the S1 short-term memory attractor states, but between 1 s and 1.5 s we apply a distracting input to pool S2 with varying strengths. The aim of this condition is to measure how distractible the network is. The degree of distractibility is measured parametrically by the strength of the input to S2 required to remove the high-activity state of the S1 population. These simulation protocols serve to assess the generic properties of the dynamical attractor system rather than to model specific experimental data obtained in particular paradigms.
We used a mean-field approach (described in Protocol S1) to calculate the stationary attractor states of the network for the delay period [21]. These attractor states are independent of any simulation protocol of the spiking simulations and represent the behavior of the network by mean firing rates to which the system would converge in the absence of statistical fluctuations caused by the spiking of the neurons and by external changes. Therefore, the mean-field technique is suitable for tasks in which temporal dynamics and fluctuations are negligible. It also allows a first assessment of the attractor landscape and the depths of the basins of attraction, which then need to be investigated in detail with stochastical spiking simulations. Part of the utility of the mean-field approach is that it allows the parameter region for the synaptic strengths to be investigated to determine which synaptic strengths will on average produce stable activity in the network, for example, of persistent activity in a delay period after the removal of a stimulus. For the spontaneous state, the initial conditions for numeric simulations of the mean-field method were set to 3 Hz for all excitatory pools and 9 Hz for the inhibitory pool. These values correspond to the approximate values of the spontaneous attractors when the network is not driven by stimulus-specific inputs. For the persistent state, a selective pool was set to a higher initial value (30 Hz) to account for the excitation of these neurons during the preceding cue period.
In addition, we used the mean-field technique to assess the flow between the attractor states. The flow is the force that drives the system toward the attractor given a parameter value in phase space, i.e., the firing rates of the pools. Since we were interested in the depth of a single attractor, we used a setup with just one selective pool. This was done by fixing the value of the firing rate of the selective pool and letting the other values converge to their fixed point. Afterward, we computed the flow with this configuration [36].
|
10.1371/journal.pntd.0002597 | Recombinant Antigens from Phlebotomus perniciosus Saliva as Markers of Canine Exposure to Visceral Leishmaniases Vector | Phlebotomus perniciosus is the main vector in the western Mediterranean area of the protozoan parasite Leishmania infantum, the causative agent of canine and human visceral leishmaniases. Infected dogs serve as a reservoir of the disease, and therefore measuring the exposure of dogs to sand fly bites is important for estimating the risk of L. infantum transmission. In bitten hosts, sand fly saliva elicits a specific antibody response that reflects the intensity of sand fly exposure. As screening of specific anti-saliva antibodies is limited by the availability of salivary gland homogenates, utilization of recombinant salivary proteins is a promising alternative. In this manuscript we show for the first time the use of recombinant salivary proteins as a functional tool for detecting P. perniciosus bites in dogs.
The reactivity of six bacterially-expressed recombinant salivary proteins of P. perniciosus, yellow-related protein rSP03B, apyrases rSP01B and rSP01, antigen 5-related rSP07, ParSP25-like protein rSP08 and D7-related protein rSP04, were tested with sera of mice and dogs experimentally bitten by this sand fly using immunoblots and ELISA. In the immunoblots, both mice and canine sera gave positive reactions with yellow-related protein, both apyrases and ParSP25-like protein. A similar reaction for recombinant salivary proteins was observed by ELISA, with the reactivity of yellow-related protein and apyrases significantly correlated with the antibody response of mice and dogs against the whole salivary gland homogenate.
Three recombinant salivary antigens of P. perniciosus, yellow-related protein rSP03B and the apyrases rSP01B and rSP01, were identified as the best candidates for evaluating the exposure of mice and dogs to P. perniciosus bites. Utilization of these proteins, or their combination, would be beneficial for screening canine sera in endemic areas of visceral leishmaniases for vector exposure and for estimating the risk of L. infantum transmission in dogs.
| The protozoan parasite Leishmania infantum is a causative agent of zoonotic visceral leishmaniasis, an important and potentially fatal human disease. The main reservoir hosts of this Leishmania species are dogs, and the only proven vectors are phlebotominae sand flies, Phlebotomus perniciosus being considered the major vector in the western Mediterranean area. During feeding on the host, sand flies spit saliva into the host skin; hosts develop a specific antibody response directed against sand fly salivary proteins and levels of these antibodies reflect the intensity of sand fly exposure. As the availability of salivary gland homogenate is limited, recombinant salivary proteins have been suggested as antigens suitable for measuring specific antibody levels. In the present work, we expressed six of the most-antigenic salivary proteins, and studied the mice and canine humoral immune responses to these recombinant proteins. We demonstrated that three proteins, a yellow-related protein and two apyrases, are suitable antigens for measuring anti-P. perniciosus antibody levels and estimating the host exposure to this sand fly species.
| Canine leishmaniasis (CanL) is a systemic disease with variable clinical symptoms. Its causative agent, the protozoan parasite Leishmania infantum, is transmitted by phlebotomine sand flies (Diptera: Phlebotominae). CanL occurs frequently around the Mediterranean Basin and in many countries in Latin America, where the prevalence of infection often exceeds 25%. Dogs with inapparent infections often play a role in the circulation of the parasite, as they are able to infect sand flies (reviewed in [1]). New cases of autochthonous leishmaniasis caused by L. infantum have been occurring in various countries, suggesting an expansion of CanL towards new biotopes at higher latitudes and higher altitudes (reviewed in [2], [3], [4]). Importantly, CanL is not just a veterinary problem; infected dogs serve as a reservoir host of human visceral leishmaniasis and there is a correlation between the prevalence of leishmaniasis in the canine population and the human disease in many countries [1].
Two sand fly genera are involved in L. infantum transmission, Lutzomyia in the New World and Phlebotomus in the Old World. Seven species of the genus Phlebotomus, subgenus Larroussius, are proven or probable vectors of CanL in different places around the Mediterranean Basin [5]. Of these, Phlebotomus perniciosus has the widest distribution, with ranges in both the southern and northern parts of the Mediterranean, from Morocco and Portugal in the west to Italy in the east and Germany in the north [5].
Measuring the exposure of dogs to sand fly bites is important for estimating the risk of L. infantum transmission. Recently, it has been demonstrated that experimental exposure of dogs to L. longipalpis or P. perniciosus bites elicits the production of specific anti-saliva IgG (measured by ELISA with whole SGH ) that positively correlates with the number of sand fly bites [6], [7]. The elicitation of IgG antibody levels persists for at least five weeks after last exposure of dogs to P. perniciosus [7] or even nineteen weeks after dogs are exposed to L. longipalpis [6]. Therefore, monitoring canine IgG levels specific for sand fly saliva could indicate their exposure to sand fly bites. Such a monitoring technique would be useful for evaluating the need for, and effectiveness of, anti-vector campaigns [6], [8]. However, obtaining sufficient native antigens through sand fly dissections for the upscaled production of such antigens necessary for ELISA is not feasible; thus there is a need to replace native antigens in the ELISAs by recombinant antigens. To date, recombinant salivary proteins from Phlebotomus papatasi have been used to detect the antibody response in mice experimentally exposed to this sand fly species [9], and L. longipalpis recombinant antigens have been tested for reactivity with the sera of naturally bitten humans, dogs and foxes [10], [11].
Recent studies [7], [12] have shown that the sera of dogs bitten by P. perniciosus recognize up to thirteen antigens in the salivary gland homogenate (SGH) of this species. The most intense reaction has repeatedly been observed against the 43 kDa yellow-related protein PpeSP03B (referred to further in the text as yellow protein rSP03B), the 35.3 kDa PpeSP01B and 35.5 kDa PpeSP01 kDa apyrases (apyrase rSP01B and rSP01, respectively), the 30 kDa antigen 5-related protein PpeSP07 (antigen 5 rSP07), the 29 kDa ParSP25-like protein PpeSP08 (ParSP25 protein rSP08), and the 24.5 kDa D7-related salivary protein PpeSP04 (D7 protein rSP04). The recombinant forms of these six proteins were therefore chosen in the present work as the most promising candidates for markers of P. perniciosus exposure.
SKH1-hr mice were maintained and handled in the animal facility of Charles University in Prague in accordance with institutional guidelines and Czech legislation (Act No. 246/1992 coll. on Protection of Animals against Cruelty in present statutes at large), which complies with all relevant European Union and international guidelines for experimental animals. The experiments were approved by the Committee on the Ethics of Animal Experiments of the Charles University in Prague (Permit Number: 24773/2008-10001) and were performed under the Certificate of Competency (Registration Number: CZU 934/05; CZU 307/09) in accordance with the Examination Order approved by Central Commission for Animal Welfare of the Czech Republic.
Two Phlebotomus (Larroussius) perniciosus sand fly colonies originating from Spain – from Murcia and from Madrid – were used, and maintained under standard conditions described in [13], [14].
Phlebotomus perniciosus salivary glands for immunoblots and ELISA assays were dissected from 3 to 5-day-old females, and pools of 20 salivary glands in 20 µl of Tris-NaCl buffer (20 mM Tris, 150 mM NaCl, pH 7.6) were stored at −80°C. Salivary gland homogenate (SGH) was obtained by disruption of the glands with repeated freezing and thawing cycles. Salivary glands used for isolation of mRNA were dissected from 1-day-old female sand flies, and were stored in groups of 20 salivary glands in 20 µl of RNAlater (Qiagen) and kept at −80°C until use.
Hyperimmune mice sera were obtained by repeated exposure of three SKH1hr mice to bites of uninfected P. perniciosus females; three other mice of the same strain served as non-exposed controls. Canine sera from dogs (beagles) experimentally exposed to P. perniciosus bites were selected from those previously used for studies on SGH [7]. Sera were chosen to cover a wide range of anti-saliva antibody levels: six originated from highly exposed dogs (bitten by 150–190 sand fly females once a week), six from dogs with lower exposure (20–70 females once a week) and six from unexposed dogs (before exposure). The sera of exposed dogs were collected seven days after the fourth exposure.
Six bacterially-expressed recombinant antigens expressed in nine forms were studied: yellow protein rSP03B (KF257369), two apyrases - rSP01B (KF257364, KF257366, KF178455) and rSP01 (KF257365, KF257367), antigen 5 protein rSP07 (KF257368), ParSP25 protein rSP08 (KF178457) and D7 protein rSP04 (KF178456). Coding sequences of SP01 and SP01B (referred to further in the text as rSP01B/1 in pET42b and rSP01/1 in pET42b, respectively), SP07 and SP03B were obtained from salivary glands of P. perniciosus (Murcia colony). After dissection into RNAlater, mRNA was isolated by a Roche High Pure RNA Tissue Kit and transcripted to cDNA by a Roche High Fidelity cDNA Synthesis Kit using polyA primers. The second cDNA strands of SP01B and SP01 were amplified by PCR using specific primers that were synthesized according to the sequences of the mature protein (without signal peptide). These sequences have been published with the following accession numbers [15]: rSP01B/1 - DQ192491 and SP01/1 - DQ192490. Afterwards, we followed the procedure that was described in [9] – briefly, genes were expressed in pET42b (Novagen) with a His tag containing 8 histidines – the E.coli BL21 (DE3) expression system. In parallel, single-stranded cDNA transcripted from salivary glands as described above was sent to Apronex s.r.o. (Prague) for preparation of rSP01, rSP01B, rSP03B and rSP07 proteins in the recombinant form according to sequences published in the cDNA library [15] - rSP03B (DQ150622), rSP01B/2 (DQ192491), rSP01/2 (DQ192490), rSP07 (DQ153101). All four proteins were expressed in the E.coli BL21 (DE3) expression system; apyrases rSP01B/2 and rSP01/2 were expressed in the pET42b vector (Novagen) with 2 His tags - the first containing 6 histidines and the second 8 histidines, and a 1.4 kDa adaptor, while yellow protein rSP03B and antigen 5 protein rSP07 were expressed in the pET28b vector (Novagen) with 1 His tag containing 6 histidines. All proteins were isolated under denaturing conditions with 8M urea.
In addition, salivary coding sequences of SP01B, SP04 and SP08 were obtained from a cDNA library constructed from the salivary glands of P. perniciosus from Madrid – GenBank accession numbers are: SP01B - HE974345.1, SP04 - HE980444.1, SP08 - HE974347.1; in contrast to the other proteins tested, they contain signal sequences (thus they seem to be about 3 kDa heavier on immunoblots). Recombinant proteins rSP01B and rSP04 were expressed in the pQE31 vector (Qiagen) with a His tag containing 6 histidines in E. coli M15 cells, and purified under denaturing conditions with 8M urea. Protein rSP08 was cloned into the pGEX4T3 vector (Amersham Biotech), expressed in E. coli Arctic Express cells (Agilent), and purified by polyacrylamide gel extraction in PBS. As rSP08 was expressed as a fusion protein with gluthathione S-transferase (GST), the latter was also obtained and used as a control in immunoblots and ELISA. In all these cases, protein refolding was performed using dialysis against PBS using SnakeSkin Dialysis Tubing (10 kDa MWCO, Thermo Scientific Goettiengen, Germany).
The concentration of all proteins was quantified by the Lowry method (Bio-Rad) following the manufacturer's protocol.
The immunogenicity of the recombinant proteins was tested by the immunoblot technique. Recombinant salivary proteins were separated by SDS PAGE on 12% polyacrylamide gels under non-reducing conditions using a Mini-protean apparatus (Bio-Rad). Protein concentration was optimized using preliminary experiments; the list of proteins and their quantity per well of the 5-well comb are given in Table 1. Proteins were either stained with Coomassie Blue (Invitrogen) or transferred from the gel to nitrocellulose membranes using an iBLOT dry system (Invitrogen).
Membranes were cut into strips (area corresponding to one well was cut into 5 strips), blocked for 1–2 hours with 5% milk in Tris buffer with 0.05% Tween (Tris-Tw) and then incubated for 1 hour with either mice or canine sera diluted in Tris-Tw. Mice sera were diluted 1∶200 for denatured proteins rSP01B and rSP01 in pET42b and 1∶100 in the case of other proteins; canine sera were used at a dilution of 1∶50. After washing in Tris-Tw, the strips were incubated with peroxidase conjugated anti-mouse IgG (1∶1000, AbD Serotec) or anti-dog IgG antibodies (1∶1000, Bethyl) and reacting protein bands were visualized using the substrate solution with diaminobenzidine.
The ELISA test described elsewhere [7], [16] was modified as follows. Covalink plates (Nunc) were coated with 100 µl of either salivary gland homogenate (40 ng of protein per well, corresponds to 1/5 of a salivary gland) or recombinant salivary protein, both in 0.1 M carbonate-bicarbonate buffer (pH 9.5), overnight at 4°C. The optimal concentrations of the recombinant proteins are given in Table 1. After washing in PBS with 0.05% Tween (PBS-Tw), plates were blocked with 6% milk in PBS-Tw for 1–2 h at 37°C.
Mice and canine sera were diluted in 2% milk PBS-Tw. Dilution of mice sera 1∶1600 was optimal for rSP03B, 1∶400 for apyrases in pET42b with 1 His tag, and 1∶200 for the other proteins tested. All canine sera were used at a dilution of 1∶50. Sera were incubated for 90 min at 37°C. Following washing cycles, plates were incubated with peroxidase-conjugated anti-mouse IgG (AbD Serotec) or anti-dog IgG (Bethyl) and the color reaction was developed in the substrate solution with orthophenylendiamine. Absorbance values (OD) were recorded at 492 nm using a Multiscan RC ELISA reader (Labsystems).
The non-parametric Spearman test was used to assess correlations between total anti-SGH and anti-recombinants antibody levels in GraphPad Prism version 6 (GraphPad Software, Inc., San Diego, CA). Statistical significance was considered when the p-value was <0.05.
The sequences of apyrases rSP01B/1 (in pET42b), rSP01B/2 (in pET42b), rSP01/1 (in pET42b) and rSP01/2 (in pET42b), yellow protein rSP03B (in pET28b) and antigen 5 protein rSP07 (in pET28b) were based on sequences from a published cDNA library of P. perniciosus [15]: rSP01B - DQ192491, rSP01 - DQ192490, rSP03B - DQ150622) and rSP07 - DQ153101. The other sequences were published directly in GenBank: rSP01B (in pQE31) - HE974345.1, rSP04 (in pQE31) - HE980444.1, rSP08 (pGEX4T3 ) - HE974347.1. The expressed sequences were published with the following GenBank accession numbers: yellow protein rSP03B (KF257369), two apyrases - rSP01B (KF257364, KF257366, KF178455) and rSP01 (KF257365, KF257367), antigen 5 protein rSP07 (KF257368), ParSP25 protein rSP08 (KF178457) and D7 protein rSP04 (KF178456).
All recombinant proteins except antigen 5 protein rSP07 and D7 protein rSP04 were recognized by the sera of all three repeatedly exposed mice; control sera and the GST tag were negative (Fig. 1A). A similar reactivity of recombinant antigens was found with the sera of the three dogs repeatedly exposed to P. perniciosus (Fig. 1B): all recombinant proteins except antigen 5 protein rSP07 and D7 protein rSP04 were recognized by the sera of repeatedly exposed dogs. In comparison with mice sera, the reaction of canine sera was less intense for some proteins (yellow protein rSP03B and the apyrases rSP01B/2 and rSP01/2 in pET42b) and fewer nonspecific bands appeared in the immunoblots. Control canine sera were negative (Fig. 1B).
The sera of three bitten and three non-bitten mice were tested by ELISA for the presence of antibodies against the recombinant salivary proteins as well as for the anti-SGH antibodies. Results are summarized in Fig. 2. Bitten mice had a highly elevated antibody response to the following seven recombinant proteins: apyrase rSP01B in all three plasmids, both rSP01 apyrases, yellow protein rSP03B and ParSP25-like protein rSP08. Despite the low number of sera samples tested, five of these seven proteins also showed significant positive correlations with the antibody response to total SGH (rSP01B/1 in pET42b: r = 0.94, p = 0.017; rSP01B in pQE31: r = 0.94, p = 0.017; rSP01/1 in pET42b: r = 0.9, p = 0.033; rSP03B: r = 0.93, p = 0.017; rSP08: r = 1.0, p = 0.003). The sixth and seventh proteins - rSP01B/2 and rSP01/2 in pET42b - showed positive correlations but were not significant (r = 0.77, p = 0.103 for both of them).
The ELISA results of recombinant proteins with eighteen canine sera (covering a wide range of anti-SGH antibody levels) are given in Fig. 3. Highly positive correlations with the reaction against SGH were obtained for the two denatured apyrases rSP01B and rSP01 (for both proteins in pET42b with 1 His tag: r = 0.91, p<0.0001; for rSP01B with 2 His tags r = 0.89, p<0.0001; and for rSP01 with 2 His tags r = 0.91, p<0.0001) and yellow-related protein rSP03B (r = 0.89, p<0.0001) (Fig. 3). The correlations were not significant for the other four proteins tested (Fig. 3, data not shown for antigen 5 rSP07).
In this study, we evaluated the reactivity of six recombinant salivary proteins of P. perniciosus with animal sera using immunoblots and ELISA. We chose the sera of mice as model laboratory animals and the sera of dogs as the natural reservoir host of L. infantum.
In immunoblots, the recombinant antigens reacted similarly with both mice and canine sera: specific reactions were achieved with both apyrases rSP01B and rSP01 (altogether 5 forms tested), yellow protein rSP03B and ParSP25 protein rSP08. On the other hand, antigen 5 protein rSP07 and D7 protein rSP04 were not recognized by any sera. The only difference between mice and canine sera was in the intensity of the positive bands – reactions with mice sera were stronger.
In ELISA, the reactivity of mice and canine sera differed in some aspects; the antibody response to refolded apyrase rSP01B, D7 protein rSP04 and ParSP25 protein rSP08 correlated with anti-SGH response only in mice sera. On the other hand, three denatured recombinant proteins, yellow protein rSP03B and two apyrases, rSP01B and rSP01, correlated significantly with the anti-SGH antibody response using both mice and canine sera. Variations in antigen conformation (denatured vs. refolded) may have led to the exposure of different epitopes. The discrepancy between results found for the refolded rSP01B and rSP08 using ELISA vs. immunoblot could be explained by differences in the exposure of antigens in these techniques; a similar lack of concordance has already been observed between ELISA and immunoblots with mice antibodies against Phlebotomus sergenti saliva [17].
Yellow-related proteins were found in the saliva of all sand fly species studied [18], [19], [20], [21]. They were shown to have hemagglutination and lectin-like properties [22]. They also act as high affinity binders of proinflammatory biogenic amines such as serotonin, catecholamines and histamine, suggesting that these proteins may reduce inflammation during sand fly blood-feeding [23]. In L. longipalpis, yellow-related protein LJM11 has been proven to have immunogenic properties leading to protective cellular immunity in C57BL/6 mice against leishmaniasis caused by L. major [23], [24]. Recombinant yellow-related proteins from P. papatasi and L. longipalpis reacted with the sera of hosts bitten by these sand flies [9], [10], [11]. Similarly, we have shown here that anti-P. perniciosus antibodies also strongly recognize recombinant yellow-related protein from P. perniciosus (Figures 1–3). Thus, yellow-related proteins appear to be, in general, promising markers of sand fly exposure.
Apyrases are nucleoside triphosphate-diphosphohydrolases ubiquitously present in the saliva of blood-sucking arthropods. They hydrolyze ADP and ATP in a Ca2+-dependent manner and inhibit ADP-induced platelet aggregation and inflammation to facilitate the blood feeding [18]. In sand fly host models, mouse and hamster antibodies elicited by P. duboscqi or P. perniciosus saliva recognized bacterially expressed apyrases of P. duboscqi and P. perniciosus, respectively [25], [26].
The three recombinant salivary proteins from P. perniciosus are primarily designed for measuring the canine exposure to bites of this sand fly in endemic areas of visceral leishmaniases, and for estimating the risk of L. infantum transmission to dogs. Seven sand fly species belonging to the subgenus Larroussius are proven or probable vectors of L. infantum in the Mediterranean area, with five of them being the most important: P. perniciosus, P. ariasi, P. perfiliewi, P. neglectus and P. tobbi [5], [27]. Among them, P. perniciosus is the most abundant in the Western Mediterranean at lower altitudes - in Italy, France, Spain and Portugal. In some of these localities, P. perniciosus occurs sympatrically with other Larroussius vectors, namely P. ariasi, P. perfiliewi and P. neglectus [27], [28], [29], [30], [31]. Studies on the cross-reactivity of anti-P. perniciosus antibodies with the saliva of these sand fly species are hampered by difficulties in the maintenance of Larroussius colonies; however, based on studies with other sand flies [16], [17], [32], a certain level of cross-reactivity can be expected only in closely-related species. In this case, such cross-reactivity might be an advantage as all mentioned Larroussius species are known to be L. infantum vectors [5], [27]. It is also important to point out that the specificity of the protein against sympatric sand fly species needs to be studied. Demonstrating sand fly exposure could be pivotal in the discrimination between vector-borne and direct (e.g. congenital, sexual) infections, the latter being hypothesized more and more to justify unexpected autochthonous canine leishmaniasis [33].
In conclusion, we have demonstrated that three denatured recombinant proteins from P. perniciosus saliva, the apyrases rSP01B and rSP01 and yellow protein rSP03B, are novel recombinant antigens with great promise in screening canine exposure to this important L. infantum vector and for estimating the risk of canine leishmaniases in the western Mediterranean area.
|
10.1371/journal.pntd.0002324 | Etiology of Severe Non-malaria Febrile Illness in Northern Tanzania: A Prospective Cohort Study | The syndrome of fever is a commonly presenting complaint among persons seeking healthcare in low-resource areas, yet the public health community has not approached fever in a comprehensive manner. In many areas, malaria is over-diagnosed, and patients without malaria have poor outcomes.
We prospectively studied a cohort of 870 pediatric and adult febrile admissions to two hospitals in northern Tanzania over the period of one year using conventional standard diagnostic tests to establish fever etiology. Malaria was the clinical diagnosis for 528 (60.7%), but was the actual cause of fever in only 14 (1.6%). By contrast, bacterial, mycobacterial, and fungal bloodstream infections accounted for 85 (9.8%), 14 (1.6%), and 25 (2.9%) febrile admissions, respectively. Acute bacterial zoonoses were identified among 118 (26.2%) of febrile admissions; 16 (13.6%) had brucellosis, 40 (33.9%) leptospirosis, 24 (20.3%) had Q fever, 36 (30.5%) had spotted fever group rickettsioses, and 2 (1.8%) had typhus group rickettsioses. In addition, 55 (7.9%) participants had a confirmed acute arbovirus infection, all due to chikungunya. No patient had a bacterial zoonosis or an arbovirus infection included in the admission differential diagnosis.
Malaria was uncommon and over-diagnosed, whereas invasive infections were underappreciated. Bacterial zoonoses and arbovirus infections were highly prevalent yet overlooked. An integrated approach to the syndrome of fever in resource-limited areas is needed to improve patient outcomes and to rationally target disease control efforts.
| The syndrome of fever is caused by a large number of infectious diseases. Malaria is thought to have been declining in the tropics since 2004. Increasing use of malaria diagnostic tests reveal a growing proportion of patients with fever who do not have malaria. While malaria diagnostic tests may be available, healthcare workers have few tools to diagnose causes of fever other than malaria. In order to identify major causes of fever other than malaria in northern Tanzania, we studied 870 patients with fever who were sufficiently ill to require admission to hospital. Malaria was uncommon and over-diagnosed, whereas invasive infections, including bloodstream infections, were underappreciated. Infections associated with animals such as brucellosis, leptospirosis, Q fever, and spotted fever group rickettsioses as well as viral infections transmitted by mosquitoes were common yet overlooked. We recommend that research on the syndrome of fever in resource-limited areas should focus on a wide range of potential causes. Animal-associated infections should be prioritized in patient management and disease control.
| Fever without a localized cause is one of the most common presenting complaints among persons seeking healthcare in many low- and middle-income countries [1], [2]. However, unlike the syndromes of pneumonia and diarrhea that feature in global disease burden estimates and have well coordinated programs integrating efforts across the range of responsible pathogens to avert morbidity and mortality, there has been a lack of a coordinated approach for febrile illness. While illness and death due to some specific infections causing fever, such as malaria [3] and increasingly bacterial sepsis are well quantified [4]–[6], others such as a range of zoonoses and viral infections are uncounted and consequently may be underappreciated.
The various etiologies of febrile illnesses are difficult to distinguish from one another clinically [7], [8]. As clinical laboratory services are often limited in areas where febrile conditions are particularly common [9], [10], clinicians may have few diagnostic tools to establish an etiologic diagnosis. Therefore, clinical management is often driven by syndrome-based guidelines employing empiric treatment [11]–[13]. In the absence of systematically collected data on fever etiology, considerable mismatch between clinical diagnosis, clinical management, and actual etiology may occur resulting in poor patient outcomes [14]. It is increasingly recognized that malaria is over-diagnosed in many areas [14], [15]. To address this problem, the World Health Organization (WHO) malaria treatment guidelines moved away from clinical diagnosis of malaria to treatment based on the results of a malaria diagnostic test such as a blood smear or a malaria rapid diagnostic test. With more widespread availability of diagnostic tests to exclude malaria and apparent declines in malaria worldwide [3], clinicians in resource-limited areas are faced with a growing proportion of febrile patients who do not have malaria and few tools to guide subsequent management.
We sought to describe comprehensively the causes of febrile illness in northern Tanzania among patients sufficiently ill to require hospitalization. Febrile patients admitted to two hospitals were evaluated for a wide range of infectious etiologies using conventional standard diagnostic techniques.
This study was approved by the Kilimanjaro Christian Medical Centre (KCMC) Research Ethics Committee, the Tanzania National Institutes for Medical Research National Research Ethics Coordinating Committee, and Institutional Review Boards of Duke University Medical Center and the CDC. All minors had written informed consent given from a parent or guardian and all adult participants provided their own written informed consent.
Moshi (population, >144 000) is the administrative center of the Kilimanjaro Region (population, >1.4 million) in northern Tanzania and is situated at an elevation of 890 m above mean sea level. The climate is characterized by a long rainy period (March–May) and a short rainy period (October–December) [16]. Malaria transmission intensity is low [17]. KCMC is a consultant referral hospital with 458 inpatient beds serving several regions in northern Tanzania, and Mawenzi Regional Hospital (MRH), with 300 beds, is the Kilimanjaro Regional hospital. Together KCMC and MRH serve as the main providers of hospital care in the Moshi area. In 2008, KCMC admitted 22,099 patients and MRH admitted 21,763 patients.
A study team that was independent of the hospital clinical team identified participants among infants and children admitted to KCMC from 17 September 2007 through 25 August 2008, and among adolescents and adult admitted to KCMC and MRH in Moshi, Tanzania, from 17 September 2007 through 31 August 2008. The methods of these studies have been described in detail elsewhere [7], [8]. In brief, all admitted patients were screened for eligibility by study team members as soon as possible after admission and no later than 24 hours after admission. Infants and children aged from ≥2 months to <13 years, with a history of fever in the past 48 h or an axillary temperature ≥37.5°C or a rectal temperature of ≥38.0°C, and adolescents and adults aged ≥13 years and with oral temperatures of ≥38.0°C were invited to participate in the study. Patients admitted with known malignancy, renal failure, hepatic failure, bone marrow aplasia, trauma or surgery were excluded.
A standardized clinical history and physical examination were performed on consenting patients by a trained clinical officer who was a member of the study team and who worked in parallel with the hospital admitting team. Provisional diagnoses by the hospital clinical team made independently of the study team were recorded and coded using the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) codes. Following cleansing of the skin with isopropyl alcohol and povidone iodine, blood was drawn from adults and adolescents for aerobic blood culture (5 mL) and for mycobacterial blood culture (5 mL) and from pediatric patients for a single aerobic blood culture (4 ml). In addition, blood was drawn for complete blood count, examination for blood parasites, and HIV antibody testing. Acute serum, plasma, and whole blood were archived on all participants. For patients found to be HIV seropositive, CD4-positive T lymphocyte count (CD4 cell count) and serum cryptococcal antigen level were also measured. HIV-seronegative patients were screened for the presence of acute HIV infection by polymerase chain reaction (PCR) for HIV-1 RNA. Urine was collected as soon as possible after admission for detection of urine antimicrobial activity and for antigen detection. A discharge form was completed at the time of discharge from the hospital that captured whether the patient died in hospital, the in-hospital management, and the discharge diagnoses coded using ICD-10 codes. The results of study investigations done in Moshi were provided immediately to the hospital clinical team to inform patient management. The results of investigations done at reference laboratories were provided to the hospital clinical team as they became available. The hospital clinical team was responsible for all aspects of patient management, following clinical judgment and use of locally adapted and developed treatment guidelines. All participants were asked to return to a study clinic 4–6 weeks after enrollment to provide a convalescent serum sample. To promote high levels of follow up, the study team provided a follow up appointment card prior to hospital discharge, made reminder telephone calls to participants during the week prior to the appointment, reimbursed travel expenses of returning participants, and when necessary a field worker made home visits.
Laboratory evaluations were selected to reflect a range of infectious diseases that might occur in northern Tanzania. Priority was given to laboratory evaluations for infectious diseases that might require specific management.
Data were entered using the Cardiff Teleform system (Cardiff Inc., Vista, Ca., USA) into an Access database (Microsoft Corp, Va., USA). When a diagnostic test was not applied to the whole cohort due lack of availability of an acute or convalescent sample, the proportion of confirmed cases in the tested group was extrapolated to the untested group by assuming that prevalence was the same in the tested group as in the untested group. Statistical analyses were performed with SAS version 9.1 software (SAS Inc, Cary, NC).
Figure 1 summarizes participant screening, enrollment, and diagnostic testing. Of 870 febrile admissions to two hospitals in northern Tanzania enrolled in the study 484 (55.6%) were female. Of participants, 467 (53.7%) were infants and children with a median (range) age of 2 years (2 months - 13 years); the remainder adolescents and adults with a median (range) age of 38 (14–96) years. Fifty seven (12.2%) infants and children were HIV-infected compared with 157 (39.0%) adolescents and adults. Among infants and children 34 (7.3%) of 464 with hospital outcome data died; 2 (5.9%) of those who died had invasive infections. Among adolescents and adults, 41 (10.3%) of 398 with hospital outcome data died; 11 (26.8%) of those who died had invasive infections. In hospital deaths could not be attributed to etiologies requiring serologic diagnosis due to the requirement for testing a convalescent serum sample.
Table 1 shows the number of patients with acute and convalescent samples available for testing for each etiologic agent or group of etiologic agents. Not all tests could be applied to all participants because of limited volumes of sample for some participants, and by the lack of availability of convalescent serum for participants who died before the follow up visit or who did not return. The number of confirmed cases in each group is also shown. The proportion of febrile admissions attributed to each etiology is calculated. A complete sample set was available for 243–467 (52.0–100.0%) infants and children and for 207–403 (51.4–100.0%) adolescents and adults.
Of 467 infants and children enrolled, malaria was the clinical diagnosis for 282 (60.4%), but was the actual cause of fever in 6 (1.3%). Bacterial and fungal bloodstream infections described in detail elsewhere [8] accounted for 16 (3.4%) and 4 (0.9%) febrile admissions, respectively, and were underrepresented on admission differential diagnoses. Bacterial zoonoses were identified among 49 (20.2%) of febrile admissions; 5 (2.0%) had brucellosis, 19 (7.7%) leptospirosis, 7 (2.6%) had Q fever, 18 (7.4%) had spotted fever group rickettsioses, and none had typhus group rickettsioses. In addition, 34 (10.2%) of participants had a confirmed acute arbovirus infection, all due to chikungunya (Table 1). No patient had a bacterial zoonoses or an arbovirus infection included in the admission differential diagnosis.
Of 403 adolescents and adults enrolled, malaria was the clinical diagnosis for 254 (63.0%), but was the actual cause of fever in 8 (2.0%). Bacterial, mycobacterial, and fungal bloodstream infections described in detail elsewhere [7] accounted for 69 (17.1%), 14 (3.5%), and 21 (5.2%) febrile admissions, respectively, and were underrepresented on admission differential diagnoses. Bacterial zoonoses were identified among 69 (33.3%) of febrile admissions; 11 (5.3%) had brucellosis, 21 (10.1%) leptospirosis, 17 (7.9%) had Q fever, 18 (8.7%) had spotted fever group rickettsioses, and 2 (1.0%) had typhus group rickettsioses. In addition, 21 (5.7%) of participants had a confirmed acute arbovirus infection, all due to chikungunya (Table 1). No patient had a bacterial zoonosis or an arbovirus infection included in the admission differential diagnosis.
Among all 870 participants, malaria was the clinical diagnosis for 528 (60.7%), but was the actual cause of fever in 14 (1.6%). By contrast, bacterial, mycobacterial, and fungal bloodstream infections accounted for 85 (9.8%), 14 (1.6%), and 25 (2.9%) febrile admissions, respectively, and were underrepresented on admission differential diagnoses. Bacterial zoonoses were identified among 118 (26.2%) of febrile admissions; 16 (13.6%) had brucellosis, 40 (33.9%) leptospirosis, 24 (20.3%) had Q fever, 36 (30.5%) had spotted fever group rickettsioses, and 2 (1.8%) had typhus group rickettsioses. In addition, 55 (7.9%) of participants had a confirmed acute arbovirus infection, all due to chikungunya (Table 1). No patient had a bacterial zoonoses or an arbovirus infection included in the admission differential diagnosis. The proportional etiology of febrile illness among study participants after extrapolating to the untested group is summarized in Figure 2.
We demonstrate among hospitalized febrile patients in northern Tanzania that malaria is uncommon and over-diagnosed, while invasive bacterial, mycobacterial, and fungal infections are underappreciated. At the same time, the bacterial zoonoses leptospirosis, Q fever, and spotted fever rickettsioses, and to a lesser extent brucellosis, and the arbovirus infection chikungunya are common yet unrecognized causes of fever. Our findings point to important mismatches between clinical diagnosis and management with actual diagnoses that have major implications for patient care, disease control and prevention, and for judicious use of antimalarial medications.
While the problem of malaria over-diagnosis has been appreciated for some time [14], [15], studies that comprehensively describe the causes of severe non-malaria fever requiring hospital admission beyond bloodstream infections have been lacking. The over-diagnosis of malaria results in inappropriate use of antimalarial medications and may be associated with higher case fatality rates among patients treated for malaria who do not have the infection [14], [15], [34]. While the underlying causes of the over-diagnosis of malaria are complex [35], the lack of epidemiologic information about the importance of alternative infections and guidance on their management is likely to play a role. Our findings confirm the potential benefits of making reliable malaria diagnostic tests available at healthcare facilities and using the results as the basis for prescription of antimalarial medications [36]. When adopted, such an approach to malaria treatment would support the judicious use of antimalarials and would define the population of patients with nonmalaria fever.
We found that the bacterial zoonoses, leptospirosis, Q fever, and spotted fever group rickettsioses, and to a lesser extent brucellosis, are major causes of febrile illness among patients sufficiently unwell to require hospitalization. That a group of neglected bacterial zoonoses are of major clinical and public health importance in sub-Saharan Africa is a new and paradigm-changing finding. For clinical practice, with the exception of leptospirosis that may be effectively treated with commonly prescribed antibacterials, patients with brucellosis, Q fever, and the rickettsioses are likely to leave hospital without specific treatment. In northern Tanzania where many rely on livestock for their health and economic wellbeing, Leptospira, Brucella, and Coxiella spp. also indirectly affect human health through their impact on animal fertility, growth, and survival. The control and prevention of the neglected bacterial zoonoses is likely to involve interventions that require the collaboration of human health experts with the animal and environmental health disciplines, an approach that is underdeveloped in many parts of the world.
Clinical guidelines for management of febrile patients in low resource areas focus on the identification and treatment of malaria and bacterial sepsis [11]–[13]. Our findings suggest that there is a need to identify and incorporate guidance on when to use a tetracycline for treatment of Q fever or rickettsial infection and when to consider treatment for brucellosis. We have previously demonstrated that features of the clinical history and physical examination do not perform well for identifying fever etiology [7], [8], [21], [26], [30]. Therefore, improvements to treatment algorithms for febrile patients are likely to require the development and incorporation of reliable diagnostic tests that provide timely diagnostic information to clinicians [37]. Unfortunately, many rapid diagnostic tests for infections related to fever management other than malaria and HIV suffer from poor performance characteristics [38], [39].
Lack of coordination among groups working on the various etiologies of febrile illness in low-resource areas has meant that sentinel studies that could provide much more comprehensive information on a wide range of responsible organisms instead have focused on only one or a small handful of etiologies. For example, a clinical trial evaluating the impact of pneumococcal conjugate vaccine on rates of Streptococcus pneumoniae bacteremia in a community has the potential to identify and report all bloodstream infections. Similarly, a study designed to estimate the incidence of typhoid fever to inform vaccine policy could collect acute serum along with the blood culture and, with subsequent collection of convalescent serum, would have the ability to estimate the incidence of leptospirosis and a range of other etiologic agents using conventional serologic methods [40]. However, resources for research have tended to be targeted to specific pathogens and researchers have struggled to leverage additional resources to address a broader range of organisms. Sentinel site studies seeking to understand the infectious causes of febrile illness in low-resource settings have utilized blood culture to highlight the importance of invasive bacterial and fungal infections [4], [41]. Expanding laboratory evaluations to include serologic and molecular approaches to diagnosing infections requiring specific antimicrobial management such as the bacterial zoonoses brucellosis, leptospirosis, Q fever, and the rickettsioses adds considerable value [40]. Detection of infections of public health importance such as those caused by the arboviruses dengue, Rift Valley fever, and yellow fever can inform national control programs. Since considerable etiologic overlap exists between the syndromes of fever, acute respiratory tract infection, and diarrhea [42], [43], addressing these simultaneously in integrated sentinel studies would inform enhancements in empiric treatment guidelines and improvements in the accuracy of syndrome-based disease burden estimates.
Our study had a number of limitations. While we examined a wide range of etiologies of fever, a large proportion of patients were undiagnosed suggesting that we failed to identify potentially important infections. The undiagnosed group is being investigated further using pathogen discovery approaches. Some of the diagnostic tests used in our study are less than 100% sensitive and specific and we did not test for every known pathogen. As a consequence, we probably underestimated the prevalence of some infections while misclassifying others that were falsely positive. Because a number of our diagnostic tests relied on the demonstration of a four-fold rise in antibody titer between the acute and convalescent serum sample, not all enrolled patients returned for collection of convalescent serum to have diagnoses confirmed. It follows that calculation and comparison of case fatality rate was not possible since those who died before the convalescent visit could not be confirmed cases. Incomplete diagnostic information meant that we had to extrapolate prevalence from the tested population to the untested population, potentially introducing bias. Similarly, instances of apparent infection with multiple agents were not accounted for in presentation of pie graphs. Inclusion of a well control group would have allowed the calculation of attributable fractions for individual pathogens, something that should be considered for future febrile illness research, especially in areas where malaria is endemic. Since considerable geographic variation in fever etiology is known to occur, the generalizability of our findings is uncertain.
What is needed to support an integrated approach to the syndrome of fever in resource-limited areas? First, fever should be recognized alongside pneumonia and diarrhea as a major clinical syndrome of public health importance. Achieving this is likely to require leadership from international institutions of public health and reappraisal of the way that the febrile illnesses are approached in burden of disease estimates. This could include estimating total morbidity and mortality from the syndrome of fever as a first step before attributing the associated illnesses and deaths to specific etiologies, much as is done for the other major syndromes [44], [45]. Second, efforts are needed to bring together the diverse groups and disciplines currently working on the febrile illnesses to quantify the morbidity and mortality attributable to each major etiologic agent. Such integration could be facilitated by support for research efforts that study the syndrome of fever comprehensively as well as its etiologies individually, an approach that has been modeled by studies addressing the syndromes of pediatric pneumonia and diarrhea in developing countries [46], [47]. Third, improved diagnostic services are urgently needed to establish disease burden estimates and patient management for the febrile illnesses in resource-limited areas [10]. Conventional diagnostic tests for some infections, such as leptospirosis, are complex. For example, the collection of both acute and convalescent serum samples may be required, and testing services may be available at only a few national or supra-national reference laboratories. Assays relying on convalescent samples cannot be used to estimate case fatality rates [21], [26]. Conversely, simple, rapid tests applied to acute samples may have poor performance characteristics [38]. Finally, clinical studies, including clinical trials, are needed to test and improve clinical management algorithms for febrile patients. The goal should be to target antimicrobial therapy to those who need it and to avoid inappropriate use among patients who will not benefit. In this way, patient outcomes can be improved, health resources can be conserved, and disease prevention and control efforts for febrile conditions can be rationally resourced.
|
10.1371/journal.pgen.1003071 | Construction of a Global Pain Systems Network Highlights Phospholipid Signaling as a Regulator of Heat Nociception | The ability to perceive noxious stimuli is critical for an animal's survival in the face of environmental danger, and thus pain perception is likely to be under stringent evolutionary pressure. Using a neuronal-specific RNAi knock-down strategy in adult Drosophila, we recently completed a genome-wide functional annotation of heat nociception that allowed us to identify α2δ3 as a novel pain gene. Here we report construction of an evolutionary-conserved, system-level, global molecular pain network map. Our systems map is markedly enriched for multiple genes associated with human pain and predicts a plethora of novel candidate pain pathways. One central node of this pain network is phospholipid signaling, which has been implicated before in pain processing. To further investigate the role of phospholipid signaling in mammalian heat pain perception, we analysed the phenotype of PIP5Kα and PI3Kγ mutant mice. Intriguingly, both of these mice exhibit pronounced hypersensitivity to noxious heat and capsaicin-induced pain, which directly mapped through PI3Kγ kinase-dead knock-in mice to PI3Kγ lipid kinase activity. Using single primary sensory neuron recording, PI3Kγ function was mechanistically linked to a negative regulation of TRPV1 channel transduction. Our data provide a systems map for heat nociception and reinforces the extraordinary conservation of molecular mechanisms of nociception across different species.
| Nociception is the perception of noxious, potentially damaging stimuli; and this pain or its equivalent behavioral readout is evolutionarily conserved from fruit flies to humans. Using genetic techniques in the fruit fly, we have been able to evaluate the potential functional contribution of every gene in the fruit fly genome for a role in avoidance of high noxious temperatures (heat pain-like responses). Using this functional genomics data set, we have developed a conserved network map of heat pain/nociception that predicts numerous conserved genes and pathways as novel pain pathways, including phospholipid signaling. Studies in multiple mutant mice confirmed a role for lipid signaling in pain perception, and more specifically we identify the critical lipid kinase (PI3Kγ) as a negative regulator of TRPV1 (receptor for noxious heat and capsaicin, the active component in chili peppers) signaling. This finding shows that our fly-based genetic pain network map is a valuable tool for the discovery of novel “nociception genes” in mammals.
| Although studies in inbred mouse strains and in human twin cohorts have indicated that pain has a strong genetic component [1]–[5], with an estimated heritability of ∼50%, little is known about the specific genes involved in regulating pain sensitivity across phyla. Further, conservation of gene function between species and across evolutionary time acts as a useful tool to develop an understanding of core genetic mechanisms relative to more specialized programs and how they influence behavior [6]. Drosophila is an excellent model organism for characterizing genetic regulators of behavior such as nociception [7]. Use of Drosophila genetics has highlighted a conserved role for multiple genes in the detection and avoidance of noxious heat [8], [9], and recent work on mechanosensation suggests the genetics of this process is likely also highly conserved across phyla [10]. We have previously reported a global in vivo RNAi screen for avoidance of noxious heat in Drosophila, and identification of hundreds of novel genes required in the adult fly, for its manifestation [8]. To interrogate this resource, we have now constructed a global systems network of heat pain. Our goal was to identify potentially conserved genes and pathways involved in pain perception, to provide a tool for focusing research on key pain molecular pathways.
One pathway highlighted in this global systems network was phosphatidylinositol signaling. Phosphatidylinositol signaling is a second messenger cascade involving the sequential phosphorylation of phosphatidylinositol 4-phosphate (PIP) to generate phosphatidylinositol 4,5-bisphosphate (PIP2) via PIP5' kinases (phosphatidylinositol-5-OH kinase; PIP5K), and then phosphorylation of PIP2 via PI3 kinases (phosphatidylinositol-3-OH kinase; PI3K) to generate phosphatidylinositol 3,4,5-trisphosphate (PIP3). In mammalian systems phospholipid signaling has been implicated in regulating pain perception [11]–[13], TRPV1 function [14]–[21], and itch [22], but how phosphatidylinositol signaling is involved in mammalian nociception is controversial, with data suggesting for example that PIP2 may either increase or decrease TRPV1 function [23], [24]. Based on the suggested involvement of phospholipid signaling in our conserved functional pain network, and in context of the controversial role for phosphatidylinositol signaling in pain perception, we employed genetic approaches to evaluate the role of phosphatidylinositol signaling in mammalian nociception.
To construct a global systems network of heat pain we first identified potential mouse and human orthologs of fly candidate pain genes (Figure S1). Of the 580 candidate fly thermal nociception genes we had previously identified [8], 399 had human orthologs (Table S1), many of which are known mammalian pain genes (Table S2). Gene ontology (GO) analyses of the human and mouse orthologs of the fly thermal nociception hits showed a marked enrichment of genes involved in neurotransmission and secretion, housekeeping systems such as mitochondrial structure, ATP synthesis, metabolism, or calcium signaling (Figure S2A–S2C and Table S3). We next generated an interaction map based on first-degree binding partners for the fly thermal nociception hits (Table S4). All binding partners were identified in yeast-2-hybrid screens and reported in the biomolecular interaction network database BIND, i.e. binding partners experimentally confirmed to interact with the candidate genes. Among the first degree binding partners in this interaction network, we found fly allatostatin C receptor 1, which has homology with mammalian opioid receptors, a fly homolog of Lmx1b, which regulates central serotonergic responses to opioids [25], a fly gene related to the nuclear factor-erythroid 2-related factor 2, which has anti-nociceptive effects [26] by inducing upregulation of heme oxygenase-1, and tyrosine hydroxylase, the enzyme required for dopamine and catecholamine production [27]. Thus, our genome-wide functional screen for thermal nociception in flies has generated a human gene network that includes orthologs of several known mammalian pain genes, in addition to numerous uncharacterized genes and pathway not previously associated with pain perception.
To construct a mammalian systems map of thermal nociception, we performed an enrichment analysis (using KEGG pathways and Broad Institute C2 gene sets) on the mouse and human orthologs of the fly pain genes and their first degree binding partners (Figure S1; Table S5 and Table S6). We found significant enrichment of genes (hypergeometric enrichment >90%) involved in mitochondria, metabolism, calcium signaling, inflammation, cell adhesion, RNA processing, and neurotransmission. Finally, to generate a comprehensive conserved network map of thermal nociception, the KEGG pathways from Drosophila, mouse and human were combined with relevant gene sets from the C2 annotations to create a global putative “nociception network” (Figure 1; Table S7). From this combined systems network, we identified gene sets or pathways known to play key roles in many major neural systems.
The connectivity of the entire systems map remains intact even after omitting all binding partners (Figure S3), i.e., expansion of the “pain” network map by including binding partners does not introduce a bias. Moreover, all except one pathway in the network map has >50% representation from the Drosophila heat pain hits alone. To test whether this approach can predict mammalian pain genes, we measured the overlap of the direct candidate pain hits and their binding partners with previous rat microarray expression profiling data generated from pain studies [28] and all “pain” annotated genes from OMIM (Online Mendelian Inheritance in Man, NCBI). Intriguingly, we found a 38.55% overlap of our direct pain hits and a 42.33% overlap of their binding partners with the microarray and OMIM data for pain. In contrast, 100 random gene lists gave a maximum of 6% overlapping genes and a minimum of 0% (Table 1; Table S8). Thus, our hypothesis-free systems map is markedly enriched for genes known to be associated with rodent and human pain. Considering the complexity of the neuronal network involved in generating nociception in the periphery and CNS, these pathways may operate across several neurons; however, our genome-wide functional fly pain screen and in silico data mining provide a road map for conserved molecular components and pathways putatively involved in heat nociception globally across phyla.
We next wanted to validate whether this “nociception network” has the power to identify conserved pathways involved in nociception, and if this pathway information can then help to pinpoint key mammalian pain genes. To this end we focused our first efforts on phosphatidylinositol signaling, one of the major nodes predicted from the pain systems map, and a heat pain precedented pathway. Phosphatidylinositol signaling has been implicated in heat nociception and regulation of TRPV1 by multiple groups [11], [13], [16], [18], however its precise role has been controversial [16], [18], [23], and the specific participation of different phospholipid kinases has never been evaluated genetically, which we now decided to do.
Phosphatidylinositol signaling involves the generation of PIP2 via PIP5K, and then phosphorylation of PIP2 via PI3K to generate PIP3. PIP5Kα is highly expressed in the nervous system but no neuronal function for this kinase has been established [29]. PIP5Kα mutant mice are viable and exhibit an exaggerated anaphylactic immune reaction in response to Fc-receptor engagement [30]. We find that PIP5Kα mutant mice exhibit a significant hyper-responsiveness to radiant heat (Figure 2A) and contact heat, when compared to littermate controls (Figure 2B). In mammals, TRPV1 is the prototypical noxious heat thermo-receptor, and is also the receptor for capsaicin, the active ingredient in chili peppers [31]. We therefore tested whether PIP5Kα mutant mice exhibit exaggerated TRPV1 agonist responses. Indeed, following capsaicin injection, PIP5Kα mutant mice display heightened reactivity compared to littermate controls (Figure 2C) but exhibited no difference in mechanical pain threshold using the von Frey test (Figure 2D).
PI3Kγ, the only G-protein coupled PI3K, is expressed in TRPV1-positive peripheral sensory neurons in both rats and mice [32], [33] and has been implicated in morphine-induced peripheral analgesia [32] and morphine tolerance [33]. Use of non-specific inhibitors, like wortmannin, has suggested a role for PI3' kinases in producing NGF-mediated TRPV1 sensitization [21],[34]. We therefore tested thermal nociception in PI3Kγ (p110γ) mutant mice [35] and found that these mice, like the PIP5Kα null mice, also exhibit an exaggerated behavioral response to radiant heat plantar stimulation using the Hargreaves test (Figure 3A). This enhanced pain sensitivity was confirmed using the hot plate assay (Figure 3B). PI3Kγ mutant mice also exhibit an enhanced pain response to a capsaicin challenge (Figure 3C). Similar to PIP5Kα null mice, the mechanical pain threshold using the von Frey test (Figure 3D), and the behavioral responses to acetone application (a cooling sensation) (Figure 3E) were comparable between control and PI3Kγ−/− littermates. Of note we found a similar thermal hyperalgesia phenotype in a second independent PI3Kγ−/− mutant mouse strain [36] (not shown). Thus, genetic loss of PI3Kγ and PIP5Kα results in enhanced pain responses to heat and capsaicin, providing evidence that phosphatidylinositol signaling acts as a negative regulator of heat pain perception and TRPV1 reactivity in vivo.
Since PIP5Kα and PI3Kγ cooperate to sequentially generate PIP3, and both mutant mice exhibit a similar hypersensitivity phenotype, we focused further on the role of PI3Kγ and PIP3 generation in setting the threshold for heat pain perception. PI3Kγ is highly expressed in haematopoietic cells and functions as key mediator of inflammatory cell migration to the site of injury [35], [37]. We therefore tested PI3Kγ mutant mice for potential defects in inflammation-induced pain sensitization, i.e. thermal hyperalgesia. PI3Kγ−/− and control mice developed comparable levels of thermal hyperalgesia (Figure S4A and S4B) following plantar CFA injection. CFA-induced inflammation, as determined by paw swelling, was also comparable between mutant and control mice (Figure S4C). To further exclude a potential role of haematopoietic cells, we transplanted wild type bone marrow into PI3Kγ mutant mice (WT→KO) and PI3Kγ mutant bone marrow into wild type mice (KO→WT). The presence of a wild type haematopoietic system did not rescue the enhanced sensitivity to thermal pain in the PI3Kγ mutant background (Figure 3F), i.e. the requirement for PI3Kγ in thermal sensing maps to non-haematopoeitic cells. PI3Kγ has been shown to act both in a kinase-dependent fashion, through conversion of PIP2 to PIP3, and in a kinase-independent manner [38]. We therefore tested the behavioral response of PI3Kγ kinase-dead (KD) knock-in mice. PI3Kγ KD mice exhibited a heightened reaction to noxious heat with a reduced thermal nociception latency (Figure 3G), comparable to the enhanced heat pain responses observed in complete PI3Kγ mutant mice. Thus, the kinase activity generating PIP3 modulates heat pain. We also assessed the general neurological phenotypes of PI3Kγ−/− mice, all of which appeared normal (Figure 3H; Figure S5A–S5G). Furthermore, the overall morphology and histology of the central nervous system appeared normal in PI3Kγ−/− mice. These data demonstrate that generation of PIP3 through PI3Kγ negatively regulates pain sensitivity in vivo.
To test if the phosphatidylinositol signaling pathway acts in primary sensory nociceptors, we employed electrophysiology on isolated wild type and PI3Kγ−/− dorsal root ganglion (DRG) neurons. PI3Kγ−/− DRG neurons responded to a thermal ramp (Figure 4A) with a significantly increased steepness in the inward current response to increasing temperature when compared to control neurons. This translated into a substantial increase in the Q10 value (Figure 4B), a measure of temperature-dependent rate change in channel conductivity, indicating that PI3Kγ−/− DRG cells exhibit massive hyper-activation in response to noxious heat, although initiation of this response occurs at a slightly elevated temperature (44.77°C for PI3Kγ−/− vs 42.22°C for wild type DRG neurons, Figure 4A). Since we also observed an enhanced response to capsaicin in PI3Kγ−/− mice in vivo, we directly tested TRPV1 reactivity to capsaicin in sensory neurons in vitro. In accordance with our behavioral data, isolated PI3Kγ−/− DRG neurons exhibited augmented sensitivity to capsaicin (Figure 4C and 4D). Thus, PI3Kγ functions as a negative regulator of TRPV1 responses in nociceptive neurons.
Our data provide a conserved functional systems network map for pain behaviour. This network map revealed many pathways and gene sets previously reported to be involved in mammalian nociception, including multiple genes annotated as candidate pain genes in the human OMIM database. Thus, our systems approach, starting from a functional whole genome fly screen and bioinformatic construction of a conserved pain network map, has the power to identify regulators of mammalian nociception. Our network pointed to a key role for phosphatidylinositol signaling in noxious heat nociception. Positive as well as negative regulatory functions of phosphatidylinositol signaling on the thermal nociceptive sensor TRPV1 have been reported [15], [16], [19], [20], [23], [24]. Our results provide genetic data that the phosphatidylinositol signaling pathway is relevant to heat pain sensitivity in vivo. In particular, we find that the lipid kinases PIP5α and PI3Kγ are involved in regulating heat nociception responses by acting as negative modulators of thermal pain perception and TRPV1 activity.
Our data reinforce the extraordinary evolutionary conservation of the neurobiological mechanisms of nociception, from its manifestation as an acute damage avoidance response in simple organisms like flies to the complex sensation of pain in mammals. When used in conjunction with additional complimentary approaches (e.g. published literature, gene expression profiling, or genetic association studies), this systems network map should be a valuable tool to further pinpoint and prioritize novel candidate nociception genes in mammals.
Identification of fly orthologs in mouse and human was done using pre-computed orthology predictions [39]. Gene Ontology (GO) analysis was performed using GOstat. Binding partner identification was done using GeneSpring GX. Hypergeometric tests were used to identify over-represented gene lists (BROAD Institute) and pathways (KEGG) amongst the pain hits and to generate a conserved systems map. Pain genes and binding partners in the system map that have been annotated as pain genes in the Online Mendelian Inheritance in Man database or by our previous Microarray experiments [28] were also identified.
PI3Kγ (p110γ) knock out [35], [36], kinase dead PI3Kγ knock-in [38], and PIP5Kα mutant mice [30] and have been previously described. Thermal and mechanical sensitivities were assessed using the Hargreaves, hot plate, and von Frey tests. Capsaicin behavior was assessed over 5 minutes following intraplantar injection of capsaicin.
Lumbar dorsal root ganglia (DRG) were harvested as previously reported [40], [41]. Patch-clamp recordings were performed using the whole-cell voltage-clamp configuration of the patch-clamp technique as previously described [40], [41].
All mice were bred and maintained according to an ethical animal license protocol complying with the current Austrian law.
Detailed Materials and Methods are available in Text S1.
|
10.1371/journal.pcbi.1007310 | p13CMFA: Parsimonious 13C metabolic flux analysis | Deciphering the mechanisms of regulation of metabolic networks subjected to perturbations, including disease states and drug-induced stress, relies on tracing metabolic fluxes. One of the most informative data to predict metabolic fluxes are 13C based metabolomics, which provide information about how carbons are redistributed along central carbon metabolism. Such data can be integrated using 13C Metabolic Flux Analysis (13C MFA) to provide quantitative metabolic maps of flux distributions. However, 13C MFA might be unable to reduce the solution space towards a unique solution either in large metabolic networks or when small sets of measurements are integrated. Here we present parsimonious 13C MFA (p13CMFA), an approach that runs a secondary optimization in the 13C MFA solution space to identify the solution that minimizes the total reaction flux. Furthermore, flux minimization can be weighted by gene expression measurements allowing seamless integration of gene expression data with 13C data. As proof of concept, we demonstrate how p13CMFA can be used to estimate intracellular flux distributions from 13C measurements and transcriptomics data. We have implemented p13CMFA in Iso2Flux, our in-house developed isotopic steady-state 13C MFA software. The source code is freely available on GitHub (https://github.com/cfoguet/iso2flux/releases/tag/0.7.2).
| 13C Metabolic Flux Analysis (13C MFA) is a well-established technique that has proven to be a valuable tool in quantifying the metabolic flux profile of central carbon metabolism. When a biological system is incubated with a 13C-labeled substrate, 13C propagates to metabolites throughout the metabolic network in a flux and pathway-dependent manner. 13C MFA integrates measurements of 13C enrichment in metabolites to identify the flux distributions consistent with the measured 13C propagation. However, there is often a range of flux values that can lead to the observed 13C distribution. Indeed, either when the metabolic network is large or a small set of measurements are integrated, the range of valid solutions can be too wide to accurately estimate part of the underlying flux distribution. Here we propose to use flux minimization to select the best flux solution in the13C MFA solution space. Furthermore, this approach can integrate gene expression data to give greater weight to the minimization of fluxes through enzymes with low gene expression evidence in order to ensure that the selected solution is biologically relevant. The concept of using flux minimization to select the best solution is widely used in flux balance analysis, but it had never been applied in the framework of 13C MFA. We have termed this new approach parsimonious 13C MFA (p13CMFA).
| Fluxomics is the omics field that analyses metabolic fluxes (i.e., reaction and transport rates in living cells) which are a close reflection of the metabolic phenotype. As such, quantitative tracking of metabolic fluxes is vital for deciphering the regulation mechanisms of metabolic networks subjected to perturbations, including disease states and drug-induced stress. However, unlike other omics data that can be quantified directly, the fluxome can only be estimated through an indirect interpretation of experimental data[1–3].
There are two main model-based approaches to quantifying metabolic fluxes, Flux Balance Analysis (FBA) and 13C Metabolic Flux Analysis (13C MFA). Both methods use stoichiometric, thermodynamic and experimental constraints to find the range of feasible fluxes across a metabolic network and then find the flux distributions within that space that optimize a given objective function. However, both techniques differ in the type of objective function optimized.
In FBA, the objective function is a set of fluxes to be minimized or maximized. These fluxes must represent a biological objective deemed desirable in the conditions of study (e.g., synthesis of biomass components for proliferating systems)[4]. A significant limitation of FBA is that the choice of objective(s) can significantly influence the predicted flux distributions.
In 13C MFA, the objective function is to minimize the difference between simulated and measured 13C enrichment in metabolites [5,6]. 13C enrichment is quantified in metabolic products and intermediates after incubating samples with metabolic substrates labeled with 13C (tracers) and provides information about how carbons are redistributed along metabolic pathways[7]. Compared to FBA, 13C MFA has a greater capacity to elucidate the fluxes of central carbon metabolism. However, 13C MFA is more complex to solve than FBA due to the non-linear nature of the 13C MFA objective.
A significant limitation of FBA is that there is generally a wide range of optimal flux distributions[8]. This is not usually the case with 13C MFA which can generally determine flux distributions with a high degree of accuracy. 13C MFA achieves this by integrating large sets of measured isotopologue fractions from parallel experiments with tracers optimized for different parts of the network[9–16]. However, when 13C MFA is used in large metabolic networks and with a limited set of measurements, it can also suffer from the same limitation as FBA and result on a wide interval of flux values for part of the metabolic network[5,17–19].
On FBA, an approach to reduce the range of optimal solutions consists in running a second optimization step on the optimal solution range. One of such methods is parsimonious FBA (pFBA)[20]. This approach, which follows the principle of parsimony or simplicity, consists on finding the optimal value of the primary objective function through FBA and then running a second optimization step where the sum of reaction fluxes is minimized while maintaining the optimal primary objective. The GIMME (and its derivative GIM3E) algorithms[21,22] are based on a similar principle as pFBA. Unlike standard pFBA, where all reactions fluxes are minimized with equal weight, GIMME integrates gene expression data to give greater weight to the minimization of fluxes through reactions catalyzed by lowly expressed enzymes. Different to FBA, for 13C MFA, there is currently no approach that relies on a second optimization to reduce the solution space when experimental data is insufficient to constrain the system towards a unique solution.
In addition to model-based approaches (e.g., FBA or 13C MFA), metabolic fluxes can also be analyzed through the direct or semidirect interpretation of 13C data. This approach primarily consists of predicting the contribution of a labeled substrate to the synthesis of a given metabolite (nutrient contribution) and predicting the relative activity of pathways (pathway activity analysis). Pathway activity analysis assumes that the isotopologue fractions used as a surrogate for the pathways of interest are primarily generated through them. This assumption is generally based on the assertion that the pathways of interest are the most direct way to generate such fractions from the labeled substrate used in the experiment[2,7,23–25]. Unlike 13C MFA, direct interpretation of 13C data is generally not able to quantify network-wide flux maps. Instead, it provides a series of qualitative or semiquantitative flux predictions around each analyzed metabolite. Strategies that couple direct interpretation of 13C data to regression and correlation analyses are widely applied to unveil the effect of an external perturbation, such as a therapeutic intervention, on central carbon metabolism[26–30].
Here we present parsimonious 13C MFA (p13CMFA), a new model-based approach to flux estimation. p13CMFA first minimizes the difference between experimental and simulated 13C enrichment in metabolites (13C MFA) and then applies the flux minimization principle to select the best solution among the solutions that fit experimental 13C data. Hence, p13CMFA can be used to select the best flux map in instances where experimental 13C measurements are not enough to fully constrain the 13C MFA solution space. Furthermore, the minimization can be weighted by gene expression allowing seamless integration of 13C with gene expression data (Fig 1).
We have implemented p13CMFA in Iso2Flux, our in-house developed isotopic steady-state 13C MFA software (https://github.com/cfoguet/iso2flux/releases/tag/0.7.2). As a proof of concept, we have applied it to the analysis of the metabolic flux distribution in HUVECs (Human umbilical vein endothelial cells) through the integration of a small set of 13C enrichment measurements and transcriptomics data. Furthermore, we validated the predictive capacity of p13CMFA using data from a published study of HTC116 cells where fluxes had been estimated with a high degree of confidence[14]. Using only a small subset of the measurements from such study, p13CMFA was able to achieve significantly better flux predictions than both 13C MFA and GIMME.
p13CMFA consists of two consecutive optimizations: first, the optimal solution to the 13C MFA problem is identified (Eq 1); secondly, the weighted sum of reaction fluxes is minimized within the optimal solution space of 13C MFA (Eq 2).
The 13C MFA optimization (Eq 1) identifies the flux distribution that minimizes the difference between measured and simulated isotopologue fractions [5,7]:
Xopt=min∑j(Ej−Yj(v)σj)2
(Eq 1)
SubjecttoS.v=0,lb≤v≤ub
where,
v is a vector of flux values describing a valid steady-state flux distribution;
Xopt is the optimal value of the 13C MFA objective;
Ej is the experimentally quantified fraction for isotopologue j;
Yj(v) is the simulated isotopologue fraction for isotopologue j with flux distribution v. Such simulation is performed by solving a complex non-linear system of equations built around isotopologues balances [1].
σj is the experimental standard deviation of the measurements of isotopologue j;
S is the stoichiometric matrix;
lb and ub are vectors defining the upper and lower bounds for flux values. Flux bounds can be used to integrate experimental flux measurements;
Either in large metabolic networks or when small sets of 13C measurements are integrated, the 13C MFA problem can be undetermined and there can be a wide range of possible solutions. Such indetermination emerges from cycles and alternative pathways in the metabolic network, which lead to many possible flux combinations that can result in the measured 13C label patterns. Furthermore, many of the 13C MFA solutions can involve large fluxes through futile cycles, which are usually artifacts of the optimization process as in vivo enzyme activities cannot support such large flux values. Therefore, to select the best solution among the many solutions that fit experimental 13C data, p13CMFA runs a second optimization where the weighted sum of fluxes is minimized (Eq 2):
min∑i|vi|⋅wi
(Eq 2)
subjecttoS.v=0,lb≤v≤ub,∑j(Ej−Yj(v)σj)2≤Xopt+T
where:
wi is the weight given to the minimization of flux through reaction i;
T is the maximum value that the 13C MFA objective can deviate from its optimal value (primary objective tolerance) when fluxes are minimized;
The difference between the optimal 13C MFA objective function value and the objective function value when the total reaction flux is minimized can be assumed to follow a Chi2-distribution with one degree of freedom. Therefore, setting T to 3.84 gives a p13CMFA solution within the 95% confidence intervals of 13C MFA[5].
With p13CMFA, the activity through cycles is minimized to the minimum amount needed to account for experimental measurements. Furthermore, gene expression measurements can be integrated to give greater weight to the minimization of fluxes through reactions catalyzed by lowly expressed enzymes. Then, in instances where multiple pathways can result in similar label patterns, those pathways with stronger gene expression evidence are selected. Hence, p13CMFA reduces the solution space towards a unique solution without requiring a simplification of the metabolic network or additional 13C measurements (Fig 1).
As an example of a potential application of p13CMFA, we applied it to analyze the metabolic flux distribution in HUVECs using a publicly available dataset not large enough to make meaningful flux predictions with conventional 13C MFA.
In this study, available in the MetaboLights repository[31] (accession number MTBLS412), HUVECs were incubated in the presence of the tracer [1,2-13C2]-glucose, and the relative abundance of 13C isotopologues was quantified in glycogen, ribose, lactate, and glutamate. The rates of production/consumption of glucose, glycogen, lactate, glutamate, and glutamine were also quantified. The data were integrated into a stoichiometric model of central metabolism which includes glycolysis, glycogen metabolism, pentose phosphate pathway (PPP), tricarboxylic acid (TCA) cycle, fatty acid synthesis, and energy and redox metabolism (S1 ZIP).
To predict the flux distribution using conventional 13C MFA, 95% confidence intervals were computed for each predicted flux value. From this analysis, the space of flux solutions consistent with the measured 13C enrichment was estimated. The resulting space of solution was still mostly undetermined and, in general, 13C MFA was unable to significantly constraint the flux ranges emerging from the stoichiometric and thermodynamic constraints and the measured extracellular fluxes (Fig 2, S1 Table). For instance, despite integrating measurements of 13C enrichment in ribose, it was not possible to conclude whether the oxidative branch of the pentose phosphate pathway contributed more to de novo ribose synthesis than the non-oxidative branch or vice versa.
Nevertheless, p13CMFA can be applied to select the best solution in the 13C MFA solution space. With this aim, transcriptomic data taken from the literature[32] were used to add additional penalties to the flux through lowly expressed enzymes. Indeed, by applying p13CMFA, we can now conclude that, under the condition of the study, glucose is mostly directed towards lactate production except for a small part going to the TCA cycle through pyruvate dehydrogenase (Fig 2, Fig 3). Glutamine is mainly metabolized to glutamate or directed to glycolysis through the TCA cycle and phosphoenolpyruvate carboxykinase. In the PPP, the non-oxidative branch contributes to roughly 60% of the net ribose synthesis. Only the glycogen phosphorylase/glycogen synthase futile cycle is predicted to be active, while the remaining futile cycles (i.e., the hexokinase/glucose 6-phosphatase, phosphofructokinase/fructose bis-phosphatase, pyruvate carboxylase/phosphoenolpyruvate carboxykinase, and glutaminase/glutamine synthase cycles) are predicted to be inactive. Concerning redox metabolism, most of the reduced NAD+ (NADH) produced in the mitochondria is exported to the cytosol through the malate-aspartate shuttle, where it is used to reduce pyruvate to lactate.
To evaluate the contribution of 13C MFA to the p13CMFA solution, GIMME (i.e., flux minimization weighted by gene expression without integrating 13C data) was also performed (Fig 2, S1 Table). Lacking 13C data, GIMME does not predict any activity in the oxidative branch of the pentose phosphate pathway, nor on the glycogen phosphorylase/glycogen synthase futile cycle. Furthermore, GIMME predicts a significantly larger flux through pyruvate dehydrogenase than p13CMFA. Interestingly, p13CMFA predicts an increased activity of the TCA cycle compared to the GIMME solution. This increased activity is fueled by alternative sources of acetyl-CoA such as fatty acid oxidation or catabolism of ketogenic amino acids. Hence, p13CMFA is able to take advantage of measured 13C enrichments and predict significantly different flux maps than those derived from flux minimization alone.
To validate the p13CMFA method, we used data from a metabolic characterization of the colon cancer cell line HCT 116 published by Tarrado-Castellarnau et al. [14]. In this study, 25 direct flux measurements and 24 sets of isotopologue fractions, measured after incubation with either [1,2-13C2]-glucose or [U-13C5]-glutamine, had been integrated in the framework of 13C MFA. With such a large set of experimental measurements, 13C MFA had been able to estimate the flux through 62 reactions with a high degree of accuracy. In the same study, transcriptomics data were also collected.
From this large data set, we selected a partial data set consisting of 7 experimental flux measurements (the rates of uptake/secretion of glucose, lactate, glutamine, glutamate and, oxygen and the rate of protein and glycogen synthesis) and 4 sets of isotopologue fractions (isotopologue fractions in ribose, lactate, glutamate and glycogen measured after incubation with 1,2-13C2]-glucose). Those are the sets of isotopologues and fluxes that were analyzed in the HUVECs case study with the addition of the rate of protein synthesis and oxygen consumption which Tarrado-Castellarnau et al. described as key determinants of the metabolic phenotype of HCT 116 cells. The partial data set was used to apply pFBA, GIMME, 13C MFA and p13CMFA in the framework of the metabolic network defined by Tarrado-Castellarnau et al. [14] (S2 Zip). p13CMFA was applied both with and without integrating gene expression data (p13CMFA+ge and p13CMFA-ge, respectively). Two complementary metrics, Pearson’s correlation and Euclidian distance, were used to evaluate the similarity between the predicted flux distributions and the flux maps estimated by Tarrado-Castellarnau et al. using the full dataset[14] (Fig 4, S2 Table). The results show that p13CMFA-ge yields a significantly more accurate flux prediction than both pFBA (i.e., flux minimization without integrating 13C data), and 13C MFA. Interestingly, while integrating gene expression significantly enhances the accuracy of p13CMFA (p13CMFA+ge compared to p13CMFA-ge), such effect is less marked than the effect of adding gene expression to standard flux minimization (GIMME compared to pFBA). This is due to the fact that p13CMFA-ge flux predictions have already a remarkable level of accuracy; hence, less information can be gained by adding transcriptomics data. Nevertheless, even if GIMME achieves flux predictions of similar accuracy to p13CMFA-ge, p13CMFA+ge results on flux predictions that are significantly more accurate than those obtained with GIMME. Hence, in instances were only a limited number of 13C measurements are available, p13CMFA is a valid method for obtaining accurate flux estimations, regardless of the availability of gene expression data.
13C MFA is a well-established technique and has proven to be an extremely valuable tool in quantifying metabolic fluxes[9–18]. However, to fully determine fluxes through a large metabolic network, parallel labeling experiments must be performed and 13C propagation must be quantified in many metabolites in the network[19]. Indeed, when applying 13C MFA either with a small set of experimental data or with a large metabolic network, part of the 13C MFA solution space can be too wide to draw meaningful conclusions about the underlying flux distribution. This solution space can be reduced by removing degrees of freedom from the system, for instance, by removing reactions from the network or making reactions irreversible. However, this can introduce an arbitrary bias in the resulting flux distribution.
Here we describe p13CMFA, a new approach for 13C data integration which can overcome these limitations of 13C MFA and estimate a realistic solution within an undetermined 13C MFA solution space. This solution will be the flux distribution within the 13C MFA solution space that minimizes the weighted sum of reaction fluxes. Thus, it will be the most enzymatically efficient solution. In that regard, p13CMFA is partially based on a similar principle as pathway activity analysis (i.e., the assumption that specific fractions of isotopologues are primarily generated through the simplest combinations of pathways). However, unlike pathway activity analysis, p13CMFA is able to integrate all quantified isotopologue fractions and flux measurements (e.g. rates of metabolite uptake and secretion) to generate network-wide flux maps consistent with such data. Furthermore, p13CMFA is highly flexible; for instance, here we show that it can be used to seamlessly integrate gene expression data by giving higher weight to the minimization of the fluxes through lowly expressed enzymes.
As a proof of concept, we exemplified how p13CMFA can be used to estimate flux distributions integrating only limited sets of 13C measurements in a test case where traditional 13C MFA was unable to provide a narrow solution space. Furthermore, we demonstrated that, when a limited set of measurements are integrated, p13CMFA can yield more accurate flux predictions than both 13C MFA and GIMME.
p13C MFA does not aim to be a replacement of 13C MFA; instead, it seeks to supplement it by identifying the more straightforward solution in parts of the network that cannot be uniquely determined. In that regard, it can be used to quantitatively study flux distributions in instances where not enough information can be obtained with conventional 13C MFA. Nor does it aim to replace the direct interpretation of 13C data. The latter is still a suitable technique when the goal of the analysis is to compare the relative activity of well-established pathways across conditions or quantify substrate contributions rather than to generate complete flux maps.
13C data has been widely used to assist in drug discovery. In this regard, tracer analysis coupled with regression and correlation analyses is frequently used to characterize drug response [26–29]. Such approach uses regression and correlation statistics with binary, numeric and visual analysis to integrate drug dosage, time points, as well as all necessary biological variables in order to diagnose disturbed stable isotope labeled matrices[29]. p13CMFA could further expand the role of 13C in drug discovery by allowing the integration of 13C and transcriptomic data in the framework of genome-scale metabolic models. In the framework of such models, drug targets are identified by systematically simulating the effect of reactions or genes knock out to cell function[34]. This is usually attained by applying the ROOM[35] or MOMA[36] algorithms, which take a unique flux solution as input (wild-type flux distribution) to predict the most likely effect of a gene KO. Hence, p13CMFA results could be potentially used as ROOM/MOMA inputs allowing to take full advantage of the flux information derived from both 13C and transcriptomics data to predict new drug targets. With atom mappings now available on a genome-scale[37], the main obstacle to applying p13CMFA at a genome-scale is the high computational complexity of solving the resulting non-linear problem which increases with the size of the network. Hence, the next challenge for p13CMFA will be optimizing its implementation for genome-scale networks.
The flux spectrum[38] (i.e., the feasible range of fluxes for a given set of stoichiometric, thermodynamic and flux boundary constraints) was determined using flux variability analysis [8]. Under this approach, each flux is minimized (Eq 3) and maximized (Eq 4) subject to constraints to find the minimum (vminiFS) and maximum (vmaxiFS) feasible values for each flux:
vminiFS=minvi
(Eq 3)
subjecttoS.v=0,lb≤v≤ub
vmaxiFS=maxvi
(Eq 4)
subjecttoS.v=0,lb≤v≤ub
The 13C MFA solution space is estimated by computing the confidence intervals for each flux. Such intervals are obtained by minimizing (Eq 5) and maximizing (Eq 6) each flux subject to constraints[5].
vmini=minvi
(Eq 5)
subjecttoS.v=0,lb≤v≤ub,∑j(Ej−Yj(v)σj)2≤Xopt+T
vmaxi=maxvi
(Eq 6)
subjecttoS.v=0,lb≤v≤ub,∑j(Ej−Yj(v)σj)2≤Xopt+T
where,
vmini: is the lower bound of the confidence interval for flux i with tolerance T;
vmaxi: is the upper bound of the confidence interval for flux i with tolerance T;
Provided that the same primary objective tolerance (T) is used in computing both the p13CMFA solution and the 13C MFA confidence intervals, the p13CMFA solution will always fall within the boundaries of 13C MFA confidence intervals (vmini≤vi≤vmaxi).
To apply GIMME and pFBA, the sum of fluxes is minimized subject only to network stoichiometry and flux boundaries (Eq 7).
In GIMME, flux minimization weights are derived from gene expression measurements, whereas in pFBA all reactions are given the same minimization weight[20,22].
Transcriptomic data of HUVECs and HCT 116 cells published by Weigand et al.[32] and Tarrado-Castellarnau[14] et al., respectively, were obtained from the Gene Expression Omnibus repository[39]. A Robust Multichip Analysis gene-level normalization[40] was performed with the Oligo package for R[41].
Using gene-protein-reaction rules, normalized transcript intensities were mapped to each enzyme-catalyzed reaction or protein-facilitated transport process. The weight given to the minimization of fluxes was assigned according to the following equation:
wi=1+max(Th−gei,0)
(Eq 8)
where,
gei is the gene expression value assigned to reaction i;
Th is the gene expression threshold. Fluxes through reactions with gene expression levels below this threshold are given additional minimization weight;
Using the same criteria as GIM3E[22], Th was set at the maximum gene expression value found in the set of genes mapped to the metabolic network (Eq 9):
Th=max(ge)
(Eq 9)
Using this threshold, the information gained from integrating available gene expression measurements is maximized. Other Th values were tested in the validation case study[14] and using the maximum gene expression as the threshold was found to yield the most accurate flux predictions (S3 Table).
p13CMFA was implemented in Iso2Flux, our in-house developed 13C MFA software (https://github.com/cfoguet/iso2flux/releases/tag/0.7.2).
Iso2Flux computes steady-state flux distributions as the product of the null space of the stoichiometric matrix and the vector of free fluxes. Reversible reactions are split into forward and reverse reactions. For each reversible reaction, a turnover variable (ti) is introduced defining the flux that is common to the forward (vif) and reverse (vir) reactions. These variables are used to assign values to the fluxes of the forward and reverse reactions as a function of the steady-state net flux (vi).
Iso2flux uses the Elementary Metabolite Unit (EMU) framework[1] to build the 13C propagation model. This framework is based on a highly efficient decomposition method that identifies the minimum amount of isotopologue transitions required to simulate the experimentally quantified isotopologues according to the defined carbon propagation rules. The isotopologue transitions are grouped into decoupled systems based on isotopologue size. Balance equations are built around each isotopologue fraction under the assumption of isotopic steady state (S1 Fig). Using the steady-state flux distribution as an input, systems of equations around isotopologues balances are solved sequentially starting with the smallest isotopologue size [1] using the fsolve function of the SciPy library (https://scipy.org/scipylib/index.html). Solving such system predicts the isotopologue distribution associated with a given steady-state flux distribution (Yj(v)).
The self-adaptive differential evolution (SADE) algorithm from PyGMO (Python Parallel Global Multiobjective Optimizer, https://github.com/esa/pagmo2) was used to find the optimal solution of the 13C MFA (Eq 1) and p13CMFA (Eq 2) problems. SADE was parallelized using the generalized island-model paradigm. Under such implementation, SADE is run in parallel in different CPU processes (islands). After a given number of SADE iterations (generations), the best solutions (individuals) in each SADE process (island) are shared to parallel SADE processes (migrate to adjacent islands). To prevent bias from the starting solutions (starting populations), the islands are seeded through random sampling of all variables. Free fluxes variables are sampled using the optGpSampler implemented into COBRApy[42,43]. Turnover variables are sampled using the random.uniform function built into python. The algorithm was run with 7 islands, each with a population of 60, and with migrations between islands set to occur every 400 generations. For the analyzed 13C MFA and p13CMFA problems, repeated iterations of the algorithm were shown to reliably converge towards the same minimal objective function value.
At the beginning of a 13C experiment, all internal metabolites are unlabeled (m0). Progressively, these products are enriched in 13C, with the subsequent decrease in m0. Isotopic steady state is quickly reached for small pools of metabolites but not necessarily for larger pools such as those of fatty acids, glycogen or metabolites present in large concentrations in the external medium[44]. For these larger pools, unlabeled isotopologues m0 are oversized and might not quickly decrease to the theoretical value that should be reached at steady-state.
However, it is possible to represent the effect of large pools in the framework of steady-state 13C MFA through the addition of a virtual reaction. This reaction replaces labeled isotopologues by unlabeled isotopologues in metabolites with large pools. With p13CMFA, the flux through this virtual reaction can be minimized. Effectively, this allows correcting steady-state 13C simulations for large pools while identifying the solutions that require the minimum amount of correction.
The statistical significance of the difference between correlation coefficients was evaluated using the Fisher r-to-z transformation[33]. Following this approach, Pearson’s correlation coefficients (r) can be converted to a z-score (r’):
r′=12·Ln(1+r1−r)
(Eq 12)
The variance of z (Sz) will depend only on the sample size (n):
Sz=1n−3
(Eq 13)
From Eq 12 and Eq 13, the significance of the difference between two correlation coefficients (r1 and r2) can be evaluated by computing the z score corresponding to such difference (Eq 14) and its associated p-value.
Human Umbilical Vein Endothelial Cells (HUVECs-pooled, Lonza) were maintained on 1% gelatin-coated flasks at 37°C in a humidified atmosphere of 5% CO2 and 95% air in MCDB131 (Gibco) medium, supplemented with the recommended quantity of endothelial growth medium (EGM) SingleQuots (Lonza), 10% fetal bovine serum (FBS) (Gibco), 2 mM glutamine (Gibco) and 0.1% Streptomycin (100 μg/mL)/Penicillin (100 units/mL) (S/P) (Gibco). 1 × 106 HUVECs were seeded in 1% gelatin-coated cell culture plates for 6h, and then the maintenance medium was replaced with the MCDB131 basal medium, supplemented with 2% FBS, 2 mM glutamine and 0.1% S/P and cells were incubated overnight for nutrient deprivation. After nutrient deprivation, the medium was replaced with a restricted medium containing MCDB131 medium supplemented with 2% FBS, 2 mM glutamine and 0.1% S/P with 10 mM of 50% [1,2-13C2]-glucose (Sigma-Aldrich) and cells were incubated for 40h in a humidified atmosphere with 5% CO2 and 1% O2 at 37°C. Both at the beginning (t = 0h) and the end (t = 40h) of incubation, media and pellets were collected. On the one hand, media and cell pellets were used for analyzing isotopologue abundances for glucose, lactate, glutamate, RNA ribose and glycogen. Raw data are publicly available in the MetaboLights repository at http://www.ebi.ac.uk/metabolights [31], with accession number MTBLS412. Isolation, derivatization and analysis details are described in MetaboLights. Glucose, lactate, glutamate, and glutamine concentrations were determined in media samples for estimation of secretion or uptake rates of these metabolites using spectrophotometric methods[45]. Also, the net rate of glycogen re-utilization into glucose was estimated by quantifying glycogen content at initial and final time points using [U-13C-D7]-glucose as recovery standard[46]. All biochemical data were normalized by cell number, and by incubation time (h). The resulting rates–expressed in micromoles of metabolite consumed/produced/transformed per hour per million cells (μmol·h-1·million-cells-1)–were 0.463, 0.099, 0.050 and 1.169 for glucose uptake, glutamine uptake, glutamate secretion, and lactate secretion, respectively, and a net transformation of glycogen of 0.000175.
|
10.1371/journal.pbio.1000538 | Quantitative Control of Organ Shape by Combinatorial Gene Activity | The development of organs with particular shapes, like wings or flowers, depends on regional activity of transcription factors and signalling molecules. However, the mechanisms that link these molecular activities to the morphogenetic events underlying shape are poorly understood. Here we describe a combination of experimental and computational approaches that address this problem, applying them to a group of genes controlling flower shape in the Snapdragon (Antirrhinum). Four transcription factors are known to play a key role in the control of floral shape and asymmetry in Snapdragon. We use quantitative shape analysis of mutants for these factors to define principal components underlying flower shape variation. We show that each transcription factor has a specific effect on the shape and size of regions within the flower, shifting the position of the flower in shape space. These shifts are further analysed by generating double mutants and lines that express some of the genes ectopically. By integrating these observations with known gene expression patterns and interactions, we arrive at a combinatorial scheme for how regional effects on shape are genetically controlled. We evaluate our scheme by incorporating the proposed interactions into a generative model, where the developing flower is treated as a material sheet that grows according to how genes modify local polarities and growth rates. The petal shapes generated by the model show a good quantitative match with those observed experimentally for each petal in numerous genotypes, thus validating the hypothesised scheme. This article therefore shows how complex shapes can be accounted for by combinatorial effects of transcription factors on regional growth properties. This finding has implications not only for how shapes develop but also for how they may have evolved through tinkering with transcription factors and their targets.
| A major challenge in developmental biology is to understand how patterns of gene activity are translated into complex three-dimensional forms, like hearts, wings, or flowers. Addressing this problem has not been easy, partly because of the difficulties in quantifying the effects of genes on shape and also because we lack frameworks that allow hypotheses about underlying mechanisms to be evaluated. Here we address this issue through a combination of experimental and computational approaches, using the Snapdragon flower as a model system. By quantifying the shapes of these flowers in a range of mutants with reduced or increased activity of particular genes, we show how the complex floral shape depends on the way genes act in combination in each petal region. The proposed interactions were tested by incorporating them into a computational model of the growing flower. Quantitative comparisons reveal a good agreement between the shapes generated by the model and those observed experimentally, confirming our underlying hypothesis. The Snapdragon flower, with its tightly fitting upper and lower petals, has evolved as a specialised mechanism for targeting pollinators. Our article shows how the development and evolution of such forms may have arisen by natural tinkering with the local effects of genes on growth.
| Although major progress has been made in the genetic dissection of organ and appendage development, the process whereby gene activities lead to particular tissue shapes is still poorly understood. For example, wing morphogenesis in Drosophila is one of the best defined developmental systems [1], yet little is known about how regional gene activities in the imaginal disc are translated into final wing shape [2]. Addressing this problem has not been easy for several reasons. First, genes that modify shape are normally identified through their overall phenotypic effects, making it difficult to establish how particular regions of the tissue are affected. Second, shape is often described in qualitative terms like “rounder” or “more elongated,” making it difficult to quantify and compare the effects of different gene combinations. Third, we lack modelling frameworks that allow hypotheses for how genes control morphogenesis to be evaluated quantitatively.
Here we combine molecular genetic and morphometric approaches to address these issues, using the Snapdragon (Antirrhinum majus) flower as a model system. A key advantage of choosing a plant system is that the lack of cell movement means that morphogenesis arises mainly through differential growth. Shape changes can therefore be described in terms of genes modifying rates of growth in particular orientations [3]. So far, this approach has been applied to studying the effects of genes on overall growth rates of an organ [4]. However, it should be possible to extend this principle to the subregions within an organ, thus allowing final shape to be dissected into genetically determined modulations in the local rates and orientations of growth.
The Antirrhinum flower is particularly suitable for this approach as specific shapes can be generated through inactivation or over-expression of key transcription factors. Each flower comprises two upper petals (dorsals) and three lower petals (laterals and ventral) that together form the corolla (Figure 1A–D). The petals are united proximally to form a tube while the distal regions form five lobes. The shapes of the upper and lower petals are precisely matched at the boundary between tube and lobe, termed the rim, so that the overall structure forms a closed mouth hinged at its edges.
The distinctive shapes of the upper and lower petals depend on the activities of four dorsoventral genes: CYCLOIDEA (CYC), DICHOTOMA (DICH), RADIALIS (RAD), and DIVARICATA (DIV) [5]–[9]. CYC and DICH encode TCP transcription factors that are expressed from an early stage in the dorsal domain of the flower bud. Mutants lacking both CYC and DICH have flowers with all petals resembling the ventral petal of wild type. RAD and DIV encode Myb-like transcription factors. RAD is switched on by CYC and DICH and promotes dorsal identity, while DIV is active in lower petals and promotes ventral identity. DIV is initially expressed throughout the corolla, but RAD is thought to antagonise its activity, preventing DIV from acting in dorsal petals. At later developmental stages, DIV expression becomes restricted to lateral and ventral petals through the action of the dorsally expressed genes. A cis-acting dominant mutant of CYC (backpetals) has been characterised in which CYC is ectopically expressed, leading to lower petals acquiring dorsal identity [9]. However, it is unclear whether the phenotype is a result of ectopic expression of CYC and/or its target gene RAD.
The changes in shape resulting from inactivation or over-expression of genes may be quantified using morphometric methods. Such methods have been applied previously to genetically controlled shape variations, such as mandible shape in vertebrates, wing shape in Drosophila, and leaf shape in plants [10]–[13]. This approach involves placing landmarks at key positions on the organ, aligning the resulting points, and then using multivariate methods to extract major trends in variation. The advantages of taking a quantitative approach are that average shapes for each genotype can be extracted and the main features under genetic control can be highlighted. Additionally, this approach potentially allows quantitative comparisons to be made between experimentally generated shapes and shapes generated by computational modelling, enabling hypotheses about morphogenesis to be evaluated.
Here we show that the genetic control of flower shape can be accounted for by a combination of region-specific effects. We quantify these effects through shape analysis of previously described mutants and of lines in which RAD is over-expressed in a range of genetic backgrounds. The shapes observed for multiple genotypes can be summarised with a scheme in which dorsoventral transcription factors act in combination with gene activities along the proximodistal and mediolateral axes to modulate the length or breadth of each petal region. Morphogenetic hypotheses for how these phenotypic effects might arise were evaluated using a modelling framework in which genes modify local polarities and specified growth rates [14],[15]. The petal shapes generated by the resulting model show a good quantitative match with those observed experimentally for each petal from 10 different genotypes, thus validating the underlying hypothesis. Our results suggest that evolution of shape involves a process of “tinkering”, through which size and shape of regions is adjusted by piecemeal modification of local growth properties under the control of transcription factors.
As a first step towards evaluating the effects of different genes on organ shape, the corolla was subdivided into several regions along its proximodistal axis. Most proximal is a continuous cylinder of tissue, the proximal tube. Beyond this region, the tube tissue extends to form the upper and lower palate (Figure 1E–H). The palate ends distally with a boundary called the rim, which acts as a line of transition between the tube and the lobes. The proximal region of the lobes comprises the lip, over which the lobes of adjacent petals are united (yellow dotted lines in Figure 1E–H). The lip is greatly reduced at the junction between the dorsal and lateral lobes, creating a hinge that allows the corolla to be opened by pollinators. The lobes are separate over the remaining distal region of the lobes.
To quantify the effects of dorsoventral genes on shape, the outline and size of the various regions of the corolla were captured. First, the 3-D structure of the flower was converted into a series of 2-D shapes. To achieve this conversion, the upper and lower sections of the corolla were separated by making cuts along the junction between lateral and ventral petals. The resulting petal sections were then flattened (Figure 1I–L). Second, the outlines of the regions for each petal were captured using a series of landmarks. Eight primary landmarks (green dots in Figure 1J,L) were located at recognisable morphological features, such as where the lobes become separate or where the tube rim and petal junctions intersect. Cell type patterns, which vary along the proximodistal axis of the tube, were also used to define primary landmarks for internal boundaries such as those between ventral and lateral petals. In cases where there were no discernable palate or lip regions, the landmarks bounding these regions were overlaid. The remaining 47 secondary landmarks (yellow dots in Figure 1J,L) were spaced evenly along the outlines of each region between the primary landmarks.
Taken together, the coordinates for the 55 landmarks summarise the shape and size of the regions for each petal. These coordinate values will vary in a correlated manner between petals depending on how the shapes and sizes of the regions are influenced by genotype and petal identity. The main trends or correlations can be captured using Principal Component Analysis (PCA) [16]. To implement this procedure, 110 coordinate values (from 55 landmarks) were determined for dorsal, lateral, and ventral petals from wild type as well as the various genotypes described below. Dorsal, lateral, and ventral petals were sampled from five different flowers for each genotype. Petal shapes were aligned by translation and rotation (Procrustes alignment). The average position for each landmark gave the mean petal shape and region outlines for the population. The major trends of variation about this mean were then determined by PCA on the covariance. This analysis showed that 94% of the variance in coordinate positions could be captured with four principal components (PCs).
PC1 accounts for 56% of the variance and captures variation in palate and lip size (Figure 2A). Increasing the value of PC1 gives longer petals with extended lip and palate regions, while reducing PC1 gives shorter petals with a reduced lip and palate. PC2 accounts for 23% of the variance and captures petal asymmetry (Figure 2A). Increasing the value of PC2 gives asymmetric petals with shorter lip and palate regions and a longer distal lobe on one side, while reducing PC2 gives bilaterally symmetrical petals. PC3 accounts for 11% of the variance and captures variation in distal lobe size: increasing the value of PC3 gives a smaller distal lobe, while reducing PC3 gives a larger distal lobe. PC4 accounts for 4% of the variance, with an increase in the PC4 value giving a petal that twists in one direction and a decrease giving a petal twisting the opposite way.
To determine the contribution of each PC to the specification of petal shapes, average PC values for wild-type dorsal, lateral, and ventral petals were determined and then used to reconstruct the petal shapes (Table S1, Figure 2B). If all four PC values were used for reconstruction, the resulting shapes closely resemble the observed shapes (compare top row of Figure 2B with Figure 1J,L). This result is expected because these four PCs capture 94% of the variance in petal shape. A good match was also obtained using just PC1 and PC2, showing that these two PCs are sufficient to capture the main features of the regional shapes. This finding allowed the main shape variations to be represented within a 2-D space that has PC1 and PC2 as its axes. This space will be referred to as the DorsoVentral (DV) space (Figure 2C). Each petal sample corresponds to a point in DV space. The origin of DV space, where all PC values are set to 0, corresponds to the mean petal shape. Samples of the same petal type (e.g., dorsal) form a cloud of points clustered around the mean for that petal type (Figure 2C). The dorsal and lateral clouds are near each other but well separated from the ventral cloud. This clustering reflects the similarity in overall shape and asymmetry of the dorsal and lateral petals and the difference in shape and symmetry of the ventral petals.
To determine the effect of the four dorsoventral genes on the ventrally positioned petal, we analysed its shape in several mutant backgrounds. The only dorsoventral gene expressed in the wild-type ventral petal is DIV. The ventral petal of the div mutant therefore expresses no dorsoventral genes and can be considered to represent a ground state. Relative to the wild-type ventral petal, that of div has a reduced palate, is wider, and is not bent back at the rim (Figure 3B). The reduced palate corresponds to a lower value of PC1 (PC1≈0). The div mutant is therefore shifted to the left in DV space relative to the wild-type ventral petal (Figure 3K, arrowed). The position of the div ventral ground state will be shown in all further DV spaces as a common point of reference. In wild type, expression of DIV in the ventral petal throughout development leads to a longer palate and narrower petal than the ground state. Additionally, the wild-type ventral petal bends back at the rim. These observations indicate that DIV acts to increase palate length, reduce petal width, and promote bending back at the rim.
CYC, DICH, and RAD are not expressed in the lower corolla section, so we would not expect these genes to have much effect on ventral petal shape. Consistent with this expectation, the shapes of the cyc dich and rad mutant ventral petals are similar to wild type (Figure 3C,D) and map to similar positions in DV space (Figure 3K). In contrast, the ventral petal of backpetals is markedly different from wild type, showing a reduced lip (Figure 3E). The reduced lip size correlates with a leftward shift in DV space (Figure 3K). Additionally, the distal lobe region of backpetals is larger than wild type, particularly along its lateral edges (giving a low value of PC3; Table S1). Also, similar to the ground state, the ventral lobe does not bend back at the rim in backpetals. Backpetals is a semidominant CYC allele that expresses CYC and its downstream target RAD ectopically in the ventral and lateral petals [9]. The effect of backpetals on ventral petal shape may therefore reflect the action of CYC or RAD or the combined action of both genes.
To separate the contributions of CYC and RAD, we generated plants that expressed RAD ectopically, by introducing RAD under the control of the 35S promoter. The ventral petals from these transgenic plants should express RAD but not CYC. Three transgenics were obtained, two of which showed strong petal phenotypes (Figure 4). No phenotypic effects were observed in leaves, even though RAD expression was detected by RT-PCR of the transgenics but not in wild type (unpublished data).
The most noticeable effect of ectopically expressing RAD in the ventral petal was reduction of both the lip and palate regions (Figures 4 and 3F). This reduction resulted in the 35S::RAD point cloud mapping to a similar position to backpetals in DV space (with a low value of PC1) (Figure 3K). Also, like backpetals, the 35S::RAD ventral petal lobe does not bend back. Thus, RAD can exert an autonomous effect on petal shape in the absence of CYC. However, the phenotype of 35S::RAD is not identical to that of backpetals. Unlike 35S::RAD, backpetals has a slightly enlarged medial palate and a large distal lobe (compare Figure 3E,F), indicating that CYC acts partly independently of RAD to increase the length of these regions.
To explore interactions between the dorsoventral genes further, 35S::RAD was introduced into several mutant backgrounds (Figure 3G–J). Analysis of the ventral petals showed that the tube of 35S::RAD div resembled that of the div ground state, having a reduced palate (compare Figure 3B with Figure 3G). This result is consistent with previous proposals that a major effect of RAD is to antagonise DIV [7],[8]. Additionally, the 35S::RAD ventral lip is greatly reduced compared to div, and the palate is also further reduced (the PC1 value for 35S::RAD is much less than for div). This finding indicates that RAD acts independently of DIV to reduce lip and palate length. The phenotype of 35S::RAD in ventral petals resembles that of 35S::RAD rad and 35S::RAD cyc dich. This result is expected because RAD, CYC, and DICH are not normally expressed in ventral petals. In a backpetals mutant background, 35S::RAD had little effect on ventral petal shape, also expected as RAD is already expressed ectopically in the backpetals mutant.
We next analysed the effect of dorsoventral genes on dorsally positioned petals (Figure 5). Wild-type dorsal petals express CYC, DICH, and RAD and also DIV at early stages. The main difference between wild-type dorsal petals and the ground state is the increased value of PC2, reflecting a marked asymmetry in petal shape. This asymmetry involves a reduced lip on one (lateral) side of the petal and an extended palate on the other (dorsal) side (Figure 5A). Extension of the palate on the dorsal side of the petal most probably reflects DICH activity, as palate asymmetry is not observed in the ventral petal of backpetals (Figure 3E), which only differs from wild-type dorsal petals in not expressing DICH. Reduction of length on the lateral side of the wild-type dorsal petal depends on RAD activity. In the rad mutant, lip length is restored to this side, reducing the degree of petal asymmetry (Figure 5D). The rad dorsal petals remain asymmetric because DICH activity increases palate length on the more dorsal side.
In cyc dich mutants the dorsally positioned petals are fully ventralised (Figure 5C). The petals are bilaterally symmetric because they lack both DICH and RAD expression (activation of RAD depends on CYC and DICH). The absence of RAD also leads to ectopic DIV activity in cyc dich dorsal petals (RAD normally antagonises DIV), accounting for the extended palate and higher value of PC1 relative to the ground state (Figure 5K). If RAD is ectopically expressed in cyc dich dorsal petals (35S::RAD cyc dich), the PC1 value drops below that of the ground state, as lip and palate regions both become reduced (Figure 5H). This result is consistent with RAD reducing lip and palate length and also further reducing palate length by antagonising DIV.
The div mutation does not affect dorsal petal development (Figure 5B), presumably because DIV activity is normally blocked in dorsal petals by expression of RAD. Dorsal petal development is also not affected by the backpetals mutation (Figure 5E), as expected because backpetals does not modify gene expression in the dorsal domain. 35S::RAD also had little or no effect on dorsal petals in wild-type, div, or backpetals backgrounds (Figure 5G,J). Again this result was expected because the endogenous RAD gene is expressed in dorsal petals. 35S::RAD rad dorsal petals have a wild-type phenotype, showing that the transgene complements rad in dorsal regions. This result demonstrates that the shape of the wild-type dorsal petal does not depend on spatial regulation of RAD expression within the dorsal petal.
We next analysed laterally positioned petals in various genetic backgrounds. Similar to the wild-type dorsal petal, each wild-type lateral petal is asymmetric with a reduced lip and palate on one (lateral) side and extended lip and palate on its other (ventral) side (Figure 6A). This morphology places lateral petals in a similar position to dorsal petals in DV space. However, in lateral petals asymmetry of the palate depends on DIV rather than DICH. In the div mutant, the palate is shortened on its ventral side, leading to a more symmetric shape (lower PC2 value, Figure 6B,K). The div lateral petals are still asymmetric because lip and palate length is reduced on the more lateral side of the petal. This reduction involves RAD. In rad mutants, the lateral petal becomes bilaterally symmetrical, with extended lip and palate regions (Figure 6D). The extended palate mainly reflects ectopic DIV activity (DIV is no longer antagonised by RAD), while the extended lip reflects lack of RAD activity. As RAD is not normally expressed in the lateral domain, the reduction of lateral lip growth in wild-type lateral petals involves a non-autonomous effect of RAD expression from the adjacent dorsal domain. If RAD is expressed ectopically in the lateral petal, as in 35S::RAD genotypes, the length of the lip and palate regions becomes negligible and the petal bilaterally symmetrical, with a low PC2 value, similar to that of the ground state (Figure 6F). The value of PC1 value drops below the ground state, reflecting RAD antagonising DIV and also reducing lip length (Figure 6K). Lateral petals of 35S::RAD backpetals are bilaterally symmetrical, like 35S::RAD, but have a partially extended medial palate (Figure 6J). This suggests that expressing CYC counteracts the effect of RAD on reducing palate length in medial regions.
Analysis of petal phenotypes in wild-type, mutant, and transgenic backgrounds reveals that the dorsoventral genes have several region-specific effects on shape. These effects on local shape can be accounted for by a scheme in which the dorsoventral genes interact combinatorially with a pattern of gene activities along the proximodistal and mediolateral axes (Figure 7). Candidate genes for the proximodistal gene activities are the LIP1 and LIP2 genes, which encode AP2-like transcription factors that increase palate and lip length [17], and CIN, which encodes a TCP transcription factor that increases lip length [18]. These genes may play an equivalent role to proximodistal systems involved in animal limb development [19]. Less is known about mediolateral systems in plants [20], although a notable feature in our scheme is that it involves graded changes, allowing lengths to be increased or decreased smoothly. This pattern may be similar to the way graded mediolateral information is provided by Dpp during Drosophila wing development [21],[22]. The scheme also involves graded effects for RAD activity, which spreads non-autonomously from the dorsal into the lateral domain to restrict DIV function. This spread may reflect direct movement of the RAD protein, as described for other small plant Myb proteins [23], or more indirect spreading mediated by signalling molecules.
Although the scheme in Figure 7 can account for the observed phenotypes through combinatorial effects on the shape and size of regions, it does not define the morphogenetic processes through which shapes are generated. To generate phenotypic outcomes, such as an increase or decrease in length of a petal region, genes presumably modify rates of growth along particular orientations within the region as it develops. However, predicting the consequences of particular hypotheses for growth control can be difficult for several reasons. One is that local orientations may become deformed through differential growth, dynamically modifying the principal orientations in which a region grows. Secondly, the extent to which a region grows may be mechanically constrained by neighbouring regions; so specified growth need not be the same as resultant growth. To address these issues, a computational modelling approach for growing tissues, called the GPT-framework (Growing Polarised Tissue framework), was used to determine the consequences of particular hypotheses [24]. The petal was modelled as a growing material sheet of tissue that can deform in 3-D, incorporating the combinatorial interactions described in Figure 7 [14]. Dorsoventral genes such as CYC and DICH were assumed to be expressed uniformly throughout development within their domains. According to the GPT-framework, genes influence shape by modifying tissue polarity and specified rates of growth (rates of extension along axes defined by the local polarity). For example, the combination DIV·PAL increases palate length by promoting specified growth parallel to the local polarity. Tissue polarity is established through three organisers (proximal, central, and distal), from which polarity signals propagate through the tissue. The activity of these organisers is also influenced by dorsoventral genes [14]. Figure 8 shows the output from the growth model for wild type, from the starting shape of a small lobed cylinder of tissue (Figure 8A,B) through to the final shape (Figure 8C,D).
To test the hypotheses underlying the computer model, the various genotypes described in this article were generated by setting the relevant gene activity in the model to 0 (null mutants) or to 1 everywhere (over-expression lines). The resulting corollas showed a good qualitative match to observed flowers (Figure 8G–O). To give a more quantitative comparison, petals from each model corolla were computationally flattened (e.g., Figure 8F) and their outlines processed in the same way as the observed petal data. The PC values from the model were then compared to the PC values observed experimentally for the corresponding genotype and petal (Table S1; Figure 8P–S). As can be seen in Figure 8P,Q, there is a strong correlation between model output and observational data for PC1 (R2 = 0.87, p<0.0001) and PC2 (R2 = 0.91, p<0.0001). This result shows that the model captures the main relationships between genes and shape for each petal and thus provides quantitative validation of the proposed combinatorial interactions between the dorsoventral genes proposed in Figure 7. Values for PC3 also show a significant correlation between observed and modelled (R2 = 0.56, p<0.0001; Figure 8R), suggesting that the model also captures this aspect of petal shape variation. However, PC4 showed little correlation (R2 = 0.04, p = 0.28; Figure 8S), which is not surprising because this PC captures only minor shape variations.
In the growth model, each dorsoventral gene has several region-specific effects on rates or orientations of growth. This hypothesis is consistent with these genes encoding transcription factors that act in combination with other factors to influence a variety of target genes. These interactions may have been elaborated during the evolution of the Antirrhinum lineage, leading to the formation of a corolla with a closed mouth, hinged at its edges. Such evolutionary tinkering [25] would have included promotion of dorsal and ventral palate growth, by DICH and DIV, respectively, repression of lip growth at the lateral petal boundaries by RAD to create a hinge, and promotion of tissue polarity organisers at particular locations. Thus, the close match between upper and lower petals depends on a history of multiple regional modifications. Similar principles may underlie the close match between the upper and lower jaws of vertebrates, illustrated by mutants in which the lower jaw protrudes or recedes [26]–[28]. The evolution of matched tissue shapes can be compared to the way protein domains may evolve to match each other [29]. In both cases shape-matching arises through tinkering, involving either a sequence of adjustments in regional growth properties and polarities as described here or a series of modifications to protein shape through piecemeal amino acid changes.
Plants of JI 7 (wild type), JI 98 (wild type), JI 726 (rad-726), JI 609 (rad-609), JI 721 (cyc-721), JI 608 (cyc-608), JI 705 (backpetals-705), JI 13 (div-35 [5]), and JI 718 (cyc-608 dich-719) were grown in the greenhouse as described previously [30] and recurrently crossed with 35S::RAD transgenic Antirrhinum majus lines. Stocks JI 7 and JI 98 were used as the standard wild type for comparison with the mutants.
The 35S::RAD construct was cut from a pGREEN0029 [31] vector and transformed into a binary vector pBIN 19 [32]. This expression vector was transformed into Agrobacterium strain GV3101 and used to transform Antirrhium majus as described by [33]. Three kanamycin resistant shoots were obtained and analysed by PCR using a set of primers for the kanamycin resistance gene (Neomycin phosphotransferase II), 5′-GATGGATTGCACGCAGGTTC-3′ and 5′-GTGGTCGAATG GGC AGGTAG-3′. A strong phenotype 35S::RAD transgenic line and a weak phenotype 35S::RAD transgenic line were crossed with each of the mutants listed above.
The back-crossed plants were screened on MS medium containing 50 mg/l kanamycin and genotyped using the primer sets described below. Genotyping of mutant alleles was performed by PCR using combinations of gene-specific or transposon-specific primers. Primers were 5′-aggttttatgcgacgaattttg3′ and 5′-aggttttatgcgacgaattttg-3′ for rad-726; 5′-atgtttgggaagaacacata-3′ and 5′-ctaattgatgaacttgtgct-3′ for cyc-721; 5′aggttctgactatctgcgcc-3′ and 5′-gtccagtcctttgtcacgtg-3′ for backpetals-705; 5′-atggcttcgactcgtggttc-3′ and 5′-taaggaagcttcgggtccgg-3′ for rad-609; 5′-atgtttgggaagaacacata-3′ and 5′-gtgacccatgcactcttgg-3′ for cyc-608; and 5′-gggtgttccttggacagaag-3′ and 5′-tcatgcgttcggaaagtgaag-3′ for div-35. The div mutant allele was detected by sequencing PCR products.
To detect RAD and transgene expression, total RNA was extracted from young leaves using an RNeasy Plant Mini Kit (Qiagen, UK). First-strand cDNA was synthesised using the SuperScript III First-Strand Synthesis System for RT-PCR (Invitrogen) on 5 µg of total RNA treated with a TURBO DNA-free kit (Ambion). RT-PCR was carried out using specific primer sets: 5′-atggcttcgactcgtggttct-3′ and 5′-gaattttgagatttctgaacc-3′ for RAD expression; 5′-agatggattgcacgcaggttc-3′ and 5′-gtggtcgaatgggcaggtag-3′ for NPTII expression; and 5′-attggtgctgaggttgaga-3′ and 5′-acaactgactccagcaaacg-3′ for ubiquitin expression. PCR was performed for 4 min at 94°C and then 30 cycles consisting of 40 s at 94°C, 40 s at 61°C and 60 s at 72°C, followed by 10 min at 72°C.
Flower samples were collected from eight individual plants each from mutant and transgenic lines, when flowers were fully opened. Each flower was dissected by cutting in a proximodistal direction along the tube conjunction of dorsal and lateral petals, using a razor. The upper petals (including two dorsal petals) and lower petals (including two lateral petals and a ventral petal) were flattened by gluing onto paper and photographed using a Nikon Coolpix 995 digital camera. All images were normalised to 4000 pixels/cm2 using an ImagePrep tool written in Matlab. Fifty-five landmarks (eight primary landmarks and 47 secondary landmarks) were fitted to each of the dorsal, lateral, and ventral petals (Figure 1J,L) to build the shape model using the AAMToolbox (http://fizz.cmp.uea.ac.uk/wiki/DArT_Toolshed/index.php/Main_Page) in Matlab (version: 7.2), as described in [12]. A statistical PCA model of flower petal shape and size was generated from the petal point models of the mutant and transgenic plant dataset, projected to a morphospace defined by PC1 and PC2.
|
10.1371/journal.pgen.1002069 | Natural Allelic Variation Defines a Role for ATMYC1: Trichome Cell Fate Determination | The molecular nature of biological variation is not well understood. Indeed, many questions persist regarding the types of molecular changes and the classes of genes that underlie morphological variation within and among species. Here we have taken a candidate gene approach based on previous mapping results to identify the gene and ultimately a polymorphism that underlies a trichome density QTL in Arabidopsis thaliana. Our results show that natural allelic variation in the transcription factor ATMYC1 alters trichome density in A. thaliana; this is the first reported function for ATMYC1. Using site-directed mutagenesis and yeast two-hybrid experiments, we demonstrate that a single amino acid replacement in ATMYC1, discovered in four ecotypes, eliminates known protein–protein interactions in the trichome initiation pathway. Additionally, in a broad screen for molecular variation at ATMYC1, including 72 A. thaliana ecotypes, a high-frequency block of variation was detected that results in >10% amino acid replacement within one of the eight exons of the gene. This sequence variation harbors a strong signal of divergent selection but has no measurable effect on trichome density. Homologs of ATMYC1 are pleiotropic, however, so this block of variation may be the result of natural selection having acted on another trait, while maintaining the trichome density role of the gene. These results show that ATMYC1 is an important source of variation for epidermal traits in A. thaliana and indicate that the transcription factors that make up the TTG1 genetic pathway generally may be important sources of epidermal variation in plants.
| Among the goals of modern evolutionary biology is to identify the molecular genetic sources of natural variation. Although genetic mapping has led to an increased understanding of the genetic architecture of natural variation, there are surprisingly few cases where the molecular source of the variation has been identified. Here, we utilize previous mapping results to identify the gene and ultimately a polymorphism that underlies natural variation for a dynamic trait in Arabidopsis thaliana: trichome density. We show that plants carrying a knock-out of the bHLH transcription factor ATMYC1 have a reduced trichome density phenotype; this is the first reported function for ATMYC1. Using traditional and molecular genetic approaches, we identify a single base change in natural alleles of ATMYC1, which leads to an amino acid replacement that qualitatively alters protein–protein interactions with known partners, presumably altering the trichome cell fate pathway. In a broad screen for molecular variation in ATMYC1, we identify a dense block of amino acid replacements that differentiates two high-frequency allele types. Although this block of variation does not appear to affect trichome density, based on paralogs of ATMYC1, we propose that this variation has arisen due to directional selection on another epidermal trait.
| Understanding the origins, maintenance, and loss of natural variation remain important goals of evolutionary biology; ideally, we should like to know what types of molecular genetic changes generate the variation that natural selection acts on. For most traits, variation is distributed continuously in natural populations, a product of polymorphisms at many loci, environmental effects, and genotype by environment interactions [1], [2]. Common first approaches to characterizing the genetic bases of natural variation include quantitative trait locus (QTL) mapping (see reviews [3]–[5]) and, more recently, genome-wide association mapping (e.g., [6]–[8]). While these methods provide many genetic insights, mapping results largely remain hypotheses regarding the molecular nature of biological diversity. To identify the genes and ultimately the polymorphisms that underlie natural variation still require detailed gene-by-gene analysis [9].
Information about the molecular changes that underlie natural variation within and among species provides important insights into the mechanisms that drive local adaptation, morphological evolution, and speciation. For example, molecular data have revealed a good deal about the evolution of flowering time in Arabidopsis thaliana [10]–[13], morphology in various groups [14]–[16], and speciation in Drosophila [17], [18]. Despite progress for specific traits, few general patterns have emerged regarding the molecular bases of natural variation. For example, King and Wilson [19] proposed the concept of “evolution at two levels” more than three decades ago, yet we still know little about the relative roles of coding versus regulatory mutations in evolution [20]. Such patterns may ultimately prove difficult to identify because they vary according to the taxonomic level of comparison, nature of the trait, or life history, but more data are required.
For plants, there are only ∼100 cases where the gene underlying natural variation has been identified and fewer than that for the causal polymorphism (reviewed in [21]). Perhaps further complicating the search for natural molecular evolutionary patterns, these data are heavily biased toward crops; however, roughly a third of the data are reported from work on the model flowering plant Arabidopsis thaliana. Like many model species, A. thaliana has a high degree of intra-specific phenotypic variation (reviewed in [3]) and a substantial functional genetic infrastructure [22], which make it an ideal system for pursuing the genes that underlie natural variation [23]. Indeed, studying highly variable traits with well-described molecular genetic underpinnings may represent our best opportunities to identify genes of interest and ultimately elucidate broad molecular evolutionary patterns. Epidermal cell fate in Arabidopsis thaliana represents one such system.
The interaction between an organism and its environment plays a critical role in the evolution of morphology and local adaptation [24], [25]. For individual plants, which cannot migrate away from sub-optimum conditions, this interaction is all the more important and is mediated by organs such as stomata [26], [27], root hairs [28], [29], trichomes [30]–[32], anthocyanin producing cells [33], [34], and seed coats [35]. Collectively these organs make up the plant epidermis, an elaborate skin that serves as the interface between the organism and its environment. In A. thaliana, epidermal cell fate is largely regulated by the TTG1 genetic pathway [36], which is mainly comprised of many pleiotropic and epistatic transcription factors and the scaffold protein, TTG1. Among the epidermal traits regulated by this pathway, trichome density is known to play a dynamic defensive role in A. thaliana [32]; increased trichome density under herbivorous conditions results in a fitness advantage, but individuals with higher trichome density in the absence of herbivorous insects have been shown to incur a fitness cost. While this suggests that environmental heterogeneity may maintain genetic variation for trichome density (within or between populations), we know little about the molecular nature of this variation. To date, only one QTL for trichome density has been identified [37]; ETC2 encodes a single repeat MYB protein known to be a repressor of the trichome cell fate. This leaves the molecular nature of most trichome density variation within A. thaliana unexplained.
Previous QTL mapping results for trichome number [38], [39] and trichome density [40] have identified multiple QTL in A. thaliana. One QTL mapped by Symonds et al. [40], TDL5, was localized to the top of chromosome four independently in each of four Recombinant Inbred Line (RIL) populations (Figure 1). Estimates of the physical position of this QTL and the similar magnitude of effect for TDL5 across mapping populations suggested that the same locus was mapped independently in each population. In an initial screen of the region, no gene with a known trichome phenotype was discovered; however, the search did reveal a bHLH transcription factor, ATMYC1, three paralogs of which [41], [42] have reduced trichome density phenotypes when knocked out [43]–[45]. ATMYC1 is expressed in both leaves and seeds [46] but over-expression of the gene has yielded no observable phenotype [44]. More recently, Zimmerman et al. [47] demonstrated protein-protein interactions between ATMYC1 and several R2R3 MYBs with known roles in epidermal cell fate. Here, we present genetic, molecular, and protein-protein interaction data that demonstrate that ATMYC1 is involved in epidermal cell fate and is a Quantitative Trait Gene (QTG) that underlies natural variation for trichome density. The results further reveal a complex pattern of protein evolution at ATMYC1 with as yet undetermined origin and effects.
Previous QTL mapping results for trichome density in A. thaliana localized a QTL to the top of chromosome four in four independent mapping populations [40]. Although no known trichome regulator was apparent in this region, ATMYC1, a paralog of three genes with known roles in trichome initiation was discovered. To determine if ATMYC1 has a role in trichome initiation, we examined TDNA insertion (knock-out) lines. A homozygous TDNA insertion line for ATMYC1 (SALK_057388) in a Col-0 background was determined to have a significantly different number of trichomes/first true leaf and trichome density phenotype on fifth true leaves relative to the wildtype Col-0 accession (Figure 2). The atmyc1 mutant produced fewer trichomes than wildtype on first true leaves and had a lower trichome density on fifth leaves. The trichome phenotype of atmyc1 has since been verified in two additional independent TDNA insertion lines of the gene (Figure S1).
To test for a functional difference between the Col-0 and Ler-0 (hereafter, Col and Ler) alleles of ATMYC1, quantitative genetic complementation analyses were performed by comparing the trichome densities of Col, Ler, a homozygous atmyc1 mutant in a Col background, and pairwise F1s among them (Figure 2). Germination rates were variable across genotypes in each experiment, resulting in sample sizes ranging from 11–18 and 8–14 for first and fifth leaf phenotypes, respectively. A one-way ANOVA revealed that both traits were found to differ significantly across the compared groups (first leaf phenotype: F (5, 81) = 42.455, p<0.001; fifth leaf phenotype: F (5, 56) = 20.63, p<0.001) and Tukey-Kramer post-hoc tests, which account for sample size variation, were revealing in several ways. The test cross of Col x atmyc1 showed little to no evidence of a gene dose effect (Figure 2). That is, the Col x atmyc1 genotype does not differ significantly from that of the Col wildtype genotype for first and fifth leaf trichome phenotypes, showing that a single Col allele of ATMYC1 is sufficient to complement the reduced trichome phenotype of the mutant to near wildtype levels. In contrast, the Col x Ler genotype has trichome phenotypes significantly higher than the atmyc1 x Ler genotype, showing that a single copy of the Ler ATMYC1 allele does not complement the atmyc1 mutant phenotype, indicating that Ler contains a nonfunctional (with regard to trichome initiation), recessive allele of ATMYC1.
In a screen of 72 A. thaliana accessions, considerable sequence variation was discovered among natural alleles of ATMYC1 with a total of 28 (inferred) cDNA haplotypes discovered (GenBank accession #s: JF801957-JF802028). Median-joining analyses yielded a network that is split into two diverged clusters (Figure 3); these have been labeled as Type I and Type II, with 16 and 12 haplotypes, respectively. Alleles of these two Types consistently differ by 25 substitutions, which translate to 17 amino acid replacements. Interestingly, nearly all of this variation (24 of 25 substitutions and all 17 replacements) is in exon six (Figure S2).
Both allele types are at high frequency. Of the 72 accessions for which full-length ATMYC1 sequence was obtained, 31 possess a Type I allele and 41 have a Type II allele; however, no obvious geographic pattern was evident. With regard to the four RIL mapping populations in which TDL5 was mapped previously [40], it is interesting that the six parental accessions possess five different alleles (Figure 3; only the allele carried by the four parents of the mapping populations that include Ler as a parent are labeled). Perhaps most interesting among these alleles is that which the Ler accession carries, as this allele consistently conferred lower trichome density in previous QTL mapping experiments. Three natural accessions possess this same allele, one of which is La-0 (cs6765), a wildtype accession from the same region as the progenitor of Ler; the other two are Dra-1 (cs6686) and Sg-2 (cs6859).
An analysis of the 72 A. thaliana alleles of ATMYC1 yielded overall levels of nucleotide diversity and polymorphism (π & θw; Table 1) that are somewhat higher than genome-wide average values reported for A. thaliana genes [48], [49]. A sliding window analysis revealed high localized levels of nucleotide diversity (Figure 4), the highest of which was detected within exon six.
Because regions of high nucleotide diversity corresponded with divergence between Type I and Type II alleles, we wanted to characterize the nature of this molecular variation. To explore this, we used a sliding window method to study rates of non-synonymous (KA) and synonymous (KS) divergence between Types I and II alleles across the entire 1.58 kb coding region. These analyses revealed evidence of alternative forms of selection that are gene region-specific (Figure 4). Across most of the gene, it appears that purifying selection has acted to constrain the amino acid sequence (ratios <<1); however within exon six, extremely high rates of amino acid replacement are evident between the two Types. As a KA/KS ratio greater than one is often cited as a conservative cut-off for positive selection [50], [51], values approaching 30 are exceptional. Even the more rigorous approach using the gene-wide average KS value resulted in a KA/KS ratio greater than eight in this region. Outside of exon six, no other region of ATMYC1 showed evidence of positive selection. Interestingly, most of the divergence between the two A. thaliana ATMYC1 Types falls between two indels that differentiate all A. thaliana alleles from two distant outgroup alleles (Figure 4 and Figure S2).
Of the 93 ecotypes that were screened for trichome density, 50 possessed a Type I allele and 43 possessed a Type II allele. Although broad-sense heritability was relatively high for the experiment (H2 = 0.71), there was no significant difference for trichome density detected between ecotypes carrying the two alternative ATMYC1 allele Types according to a Kruskal-Wallis test (data not shown). Although variation segregating at other loci may overwhelm the effects of alternative ATMYC1 Types, it appears that the observed sequence variation in exon six has little to no effect on trichome density. Given the sample sizes and standard deviations, a power analysis indicated that a trichome density difference of at least three units should have been detectable as significant.
Because a QTL was mapped for trichome density near ATMYC1 in the Col x Ler RIL population and quantitative genetic complementation tests revealed a functional difference between the Col and Ler ATMYC1 alleles, we examined polymorphisms between these two alleles. The Col and Ler accessions possess different ATMYC1 Types; however, the variation in exon six that distinguishes the two Types has no detectable effect on trichome density. Therefore, other polymorphisms between the Col and Ler alleles were investigated. The Col and Ler proteins differ at just four other positions: A13T, E83Q, P189A, and P343H (Col:aa position:Ler). Of these polymorphisms, only position 189 is highly conserved across proteins and taxa. Out of 100 homologs, representing monocots and dicots, retrieved through a protein-BLAST search of the Col ATMYC1 protein sequence, all 100 shared the Col state (proline) at ATMYC1 amino acid position 189 (data not shown). This position is also of interest as it resides within an undefined, but known MYB interaction domain in the amino end of close paralogs of ATMYC1 [44], [45]. The other three polymorphic positions were found to be far less conserved. Based on these results, yeast-2-hybrid experiments focused on the highly conserved position 189 and the non-conserved position 13 as a control.
We investigated the effects of two Col/Ler ATMYC1 polymorphisms on protein-protein interactions using binding assays with known partners, TTG1 and GL1 [47], [52]. The results are clear. ATMYC1 encoded by the native Col allele interacts with TTG1 and GL1 and the ATMYC1 protein encoded by the native Ler allele does not. Reciprocal replacements at position 13 for the Col and Ler alleles had no effect on binding, while reciprocal replacements at position 189 qualitatively altered binding for both proteins. Specifically, when the Col allele was changed to match the Ler allele at position 189, the protein no longer bound to TTG1 or GL1 and when the Ler allele was changed to match the Col allele at position 189, the resulting protein then bound with TTG1 and GL1 (Figure 5).
Identifying the genes and ultimately the molecular changes that underlie natural morphological variation remain important, but largely elusive goals. Here we have taken a candidate gene approach based on QTL mapping results and have identified a function for the transcription factor, ATMYC1, in which natural molecular variation affects trichome density in A. thaliana. Furthermore, our results reveal a strong signal for directional selection restricted to one exon of ATMYC1 that has no detectable effect on trichome density but may suggest a pleiotropic role for the gene.
An initial TDNA insertion line of ATMYC1 in the Col background was found to have a reduced trichome density phenotype relative to wildtype Col (Figure 2); subsequent examination of additional independent insertion lines have confirmed this trichome initiation phenotype (Figure S1). The trichome phenotype is consistent with high sequence similarity between ATMYC1 and close paralogs GL3, EGL3, and TT8 [41], [42], which also have trichome phenotypes upon knock-out [43]–[45]. However, this finding is somewhat surprising, as GL3 and EGL3 have been shown to be sufficient to fulfill the bHLH role in trichome initiation; a gl3/egl3 double mutant is completely glabrous [45], suggesting that ATMYC1 may be an enhancer of GL3 and EGL3. While the precise genetic role that ATMYC1 plays in the TTG1 pathway requires more work to elucidate, the trichome cell fate function is clear. Subsequent quantitative genetic complementation tests showed that the Ler allele of ATMYC1 cannot recover the atmyc1 trichome phenotype, while the Col allele recovers it completely (Figure 2), indicating that natural molecular variation in ATMYC1 alters trichome density.
Sequence variation between Col and Ler (parents of the one of the mapping populations where TDL5 was mapped) ATMYC1 alleles prompted a broad survey of ATMYC1 from 69 ecotypes and three lab strains of A. thaliana, which revealed a striking pattern of high frequency polymorphism. The coding region of the two primary Types observed consistently differ by 25 substitutions, which translate into 17 amino acid replacements; 24 of the 25 substitutions and all 17 replacements reside within exon six. This amounts to >10% amino acid replacement within exon six; no other region of the gene has a high rate of replacement. Allelic dimorphism has been reported for some, but not all, other loci in A. thaliana (e.g., [37], [53]–[55]) and is likely the result of diverfgence between two long-isolated populations of A. thaliana with subsequent break-down of population subdivision and admixture. Regardless of the origin of the Types, the ATMYC1 results are striking due to the high frequency of amino acid replacement and because nearly all of the variation resides within one relatively small region of the gene. That the Col and Ler alleles possess different ATMYC1 Types initially suggested that the molecular variation in exon six might explain the functional difference revealed by quantitative complementation tests; however, neither association tests nor yeast 2-hybrid experiments (Figure 5) support this hypothesis.
Outside of the block of variation in exon six that differentiates the two Types, only four other changes differentiate the coding regions of the Col and Ler alleles. Yeast-2-hybrid experiments to test the known interactions between ATMYC1 and TTG1 and ATMYC1 and GL1 showed that the P189A (Col:aa position:Ler) replacement has a qualitative effect, completely eliminating these interactions for the protein encoded by the Ler allele of ATMYC1. The proline at this position is conserved, even among distant homologs of ATMYC1, and likely resides in a known, but undefined, protein-protein interaction domain identified in close paralogs of ATMYC1 [44], [45]. Indeed, simply replacing the proline for an alanine at this position in the Col allele eliminates the interaction with TTG1and GL1, while the reciprocal change, replacing the alanine for a proline in the Ler allele, restores these interactions (Figure 5). Although this may not have been the first or the only mutation in the P189A ATMYC1 allele to eliminate gene function and reduce trichome density, these data show that the P189A replacement is sufficient to explain the functional difference between the Col and Ler alleles, presumably by altering trichome initiation, thereby decreasing trichome density.
We conclude that ATMYC1 is a Quantitative Trait Gene (QTG) for trichome density in A. thaliana and that the mutation at DNA position 565 is a Quantitative Trait Nucleotide (QTN) for the trait. An interesting point to emerge here is that a single base substitution has lead to a qualitative breakdown in protein-protein interaction, which has a quantitative phenotypic effect; based on sequence similarity, this is likely due to functional redundancy between ATMYC1, GL3, EGL3, and TT8. The nature of the Ler mutation suggests that this was the same QTG mapped for trichome density by Symonds et al. [40] in two other populations that have Ler as a parent: No-0 x Ler and CVI x Ler; CVI and No-0 possess Type I and II ATMYC1 alleles, respectively, but share the proline at amino acid position 189 with Col, further supporting the conclusion that the variation differentiating the Types has little or no effect on trichome density while the replacement at position 189 underlies the mapped effect. Furthermore, QTL mapping results show that the Ler allele at TDL5 consistently confers lower trichome density than the alternative allele from Col, CVI, and No-0 and the additive effect of TDL5 was nearly identical in all three mapping populations [40].
The ATMYC1 allele carried by Ler is shared by three natural ecotypes, La-0 (Poland), Dra-1 (Czech Republic), and Sg-2 (Germany). In our sample, this allele is at a frequency of ∼5%. Because it is unknown if ATMYC1 is pleiotropic, we cannot yet address whether or not the replacement at position 189 affects other traits. However, the 189A allele shows no signs of degradation to pseudogene status in any of the three natural ecotypes. This could be due to one or both of the following: (1) the protein has other functions that are not affected by the mutation at position 189 and is maintained by purifying selection and (2) this mutation is relatively recent and there has not been sufficient time for other mutations to accumulate. With regard to the second hypothesis, the 189A allele has at least persisted long enough for migration to increase its presence to multiple populations.
Association tests showed no obvious trichome density difference for the two high frequency ATMYC1 Types and yeast-2-hybrid experiments suggest that the divergence between the two Types has no effect on known protein-protein interactions. If the variation in exon six has no effect on trichome density, then what explains the clear signature of divergent selection between the allele Types? There would seem to be three logical explanations. First, the association test results could reflect confounding factors, such as segregation of variation at other loci that have larger effects on trichome density and essentially swamp out a potential ATMYC1 Types effect. If this were true, the Types effect would have to be in addition to and much weaker than that found for the replacement at position 189.
Second, divergence could have been in response to selection on a trait other than trichome density; indeed, paralogs of ATMYC1 (GL3, EGL3 and TT8) are all pleiotropic for several epidermal traits [43]–[45], ATMYC1 has been shown to interact with several MYB transcription factors that coordinate other epidermal fates [47], and an ATMYC1 homolog from Vitis vinifera (Vitaceae) was recently shown to have an epidermal (anthocyanin) phenotype [56]. ATMYC1 is most highly expressed in “seeds” [46], therefore, it may be expected to be involved in testa development as well; however, we have observed no differences in comparisons between a TDNA knock-out line of ATMYC1 in the Col-0 background and Col-0 for three seed coat traits: mucilage production, condensed tannin synthesis, and seed coat cell morphology (data not shown).
Finally, the two Types may have evolved independently in response to deleterious indels. Comparisons with outgroup homologs of ATMYC1 show that the divergence between the two A. thaliana allele Types resides between or near two indels (relative to outgroup taxa) of 18 and 15 bp after coding DNA positions 705 and 927 (in Col-0 sequence), respectively (Figure 4 and Figure S2; outgroup sequence data not shown). Specifically, rather than diverging from one another, the two A. thaliana Types may have independently diverged away from a common nonfunctional ancestral copy of the gene. Although at this stage we cannot determine the origins of the indels, recombination and transposable elements seem likely candidates. Whatever the origins, in A. thaliana, isolated populations may have acquired independent compensatory mutations that became fixed in each lineage. Because trichome density is dynamic, with the fitness of a given density being relative to the environment [32], such mutations may persist for long periods, thus allowing time for compensation. All A. thaliana alleles share indel states at these two positions with A. lyrata relative to the more distant outgroups, Capsella bursa-pastoris and Crucihimalaya himalaica. Further functional and analytical tests will be required to resolve the origins and potential effects of the divergence around these indels.
Trichome density in A. thaliana is likely to be under alternating forms of selection, depending on the particular environment in which a plant resides. The TTG1 genetic pathway, which contains multiple and various types of transcription factors, many of which are functionally redundant, would seem a likely reservoir of genetic variation for epidermal traits and a prime pathway for “genetic tinkering” [57] with potentially a low risk of permanent unidirectional trait change. Indeed, we have shown here that a low frequency polymorphism that results in a simple amino acid replacement in ATMYC1 reduces trichome density in natural ecotypes of A. thaliana, thereby ascribing the first function to ATMYC1. Our results also revealed a high frequency block of amino acid replacements in ATMYC1 with as yet unknown effects. ATMYC1 is the second gene in the TTG1 pathway recently identified to affect natural quantitative variation for trichome density; interestingly, for the single-repeat MYB, ETC2, high frequency polymorphisms do affect trichome density [37], while a similar pattern in ATMYC1 does not seem to alter trichome density. Clearly patterns that define the types of mutations and classes of genes that underlie natural variation may be difficult to identify; however, the TTG1 pathway is quickly emerging as a good place to search.
A TDNA insertion line (SALK_057388) for the ATMYC1 locus (At4g00480) in the Col-0 background was obtained from The Arabidopsis Biological Resource Center (ABRC; http://signal.salk.edu/cgi-bin/tdnaexpress). The initial batch of seed was screened using a standard PCR protocol to identify a lineage homozygous for the TDNA insertion, which resides in the first exon of the gene. Based on initial observations of a trichome density phenotype for the atmyc1 mutant, first leaf trichome number and fifth leaf trichome density phenotypes were then measured in replicates of Col-0 and atmyc1as described in the following section.
To test the hypothesis that variation at the ATMYC1 locus underlies trichome density variation mapped to TDL5 in previous QTL studies [40], quantitative complementation tests were performed. In the QTL mapping studies, the Ler allele at TDL5 was shown to confer lower trichome density than the alternative parents' (Col-4, CVI, and No-0) alleles in each mapping population. However, because the available atmyc1 knock-out mutation is in the Col-0 background, the most direct comparison that could be made (with regard to QTL mapping results) was between Col and Ler. Crosses were made by transferring pollen from flowers of the Ler accession onto the stigmatic surface of emasculated flowers of the atmyc1 mutant and of Col wildtype. To control for potential cytoplasmic variation among accessions, all crosses were made with Col or atmyc1 as the pollen recipient. Therefore, the resulting F1s differ only at the atmyc1 locus. This allowed for comparisons between individuals with a Col/Ler and an atmyc1/Ler genotype at ATMYC1, while holding the rest of the genome constant. That is, the only difference between the two sets of progeny is the replacement of a copy of the Col allele with a null (mutant) atmyc1 allele. To test for a dosage effect, Col was crossed to atmyc1, which yields a Col individual with a single functional ATMYC1 allele (atmyc1/Col genotype). Twenty replicates of each F1 genotype, parental accession, and the atmyc1 mutant were potted and the pots were randomized across four flats. All seed were vernalized in the dark for four days at 4°C, and subsequently moved to a fluorescently lit 20°C growth chamber. Sixteen days after emergence the number of trichomes on each of the first two true leaves of each seedling were counted under 50X magnification on a dissecting microscope; this is referred to as the “first leaf” trichome number phenotype. For the “fifth leaf' trichome density phenotype, the same experiment was set up as described above and trichome density was measured on the fifth true leaf at 21 days after emergence, as described by Symonds et al. [40]. The mean for each trait was then calculated from these data for each genotype. The genetic contribution to trichome number and trichome density variation was evaluated for first and fifth leaf phenotypes independently by ANOVA and unplanned pairwise comparisons between genotypes following the Tukey-Kramer method as described by Sokal and Rohlf [58].
DNAs were isolated from 69 natural accessions and three lab strains of A. thaliana acquired from the ABRC (Table S1), following a modified CTAB method [59]. Primers were designed from the Col-0 ATMYC1 sequence (GenBank accession #NC003075) to PCR amplify the open reading frame plus ∼200 bp up- and down-stream of the start and stop codons, respectively. Primers corresponding with the first and last 21 bp of the Col-0 ATMYC1 cDNA sequence were used to amplify the ATMYC1 homolog from outgroup taxa (all Brassicaceae): Arabidopsis lyrata, Crucihimalaya himalayica, and Capsella bursa-pastoris. All primers were used with manufacturer-supplied 1X Taq buffer, 1U AccuPrime High-Fidelity Taq polymerase (Invitrogen Inc.), and ∼20 ng genomic DNA in 20 uL reactions. PCR samples were checked for amplification success on 0.7% agarose gels stained with ethidium bromide, and were subsequently purified in Multiscreen PCR clean-up plates (#MANU03050, Millipore). Approximately 100 ng of each purified PCR product were then used in each of seven sequencing reactions using primers designed to anneal at staggered internal positions, providing a minimum of two overlapping sequences across the entire gene. Allelic contigs were constructed for each ecotype and sequence editing and validation were performed using sequencher v.4.2.2 (Gene Codes Corp.). Full-length genomic sequences of ATMYC1 for all accessions were aligned initially using clustalx v.1.83 [60], and subsequently corrected by hand. To generate inferred cDNA sequence alignments, introns were identified using the published Col-0 cDNA sequence as template (Arabidopsis Genome Initiative) and non-coding DNA sequence was removed from the genomic alignment in bioedit v.5.0.9 [61].
Independent cDNA haplotypes were identified using dnasp v.4.00 [62] and exported in rgl format. A haplotype network was constructed in network (Fluxus Technology, Ltd.) using the median-joining option and redrawn using indesign (Adobe, Inc.). A high level of divergence between two sets of alleles revealed by the haplotype network was the basis for identifying two Types of ATMYC1 alleles; these Types (I and II) are referenced in other sections.
To examine nucleotide diversity and molecular evolution of the ATMYC1 locus, the sequence analysis software dnasp v.4.00 [62] was used. The common nucleotide diversity indices, π [63] and θw [64], were measured across the entire genomic alignment for all sequences and independently for Type I and II data sets. To assess intra-gene variation for nucleotide diversity, a sliding window analysis was run along the full-length (start to stop) cDNA sequence alignment of the 72 A. thaliana alleles; window length was set at 45 bp and moved along the alignment at 3 bp intervals.
Because of initial observations of high levels of diversity and divergence between two apparent Types of ATMYC1 alleles, we tested the null hypothesis of neutral molecular evolution at this locus by measuring the nonsynonymous substitution rate (KA) and the synonymous substitution rate (KS). By examining the ratio of KA/KS, one may identify signals indicative of positive or purifying selection [50]. KA/KS ratios near one are thought to indicate a neutrally evolving gene or region of a gene, values <<1 are expected to be under purifying selection, and values >>1 indicate positive selection. Because different regions of a gene may experience different forms of selection, a sliding window analysis was used to examine sequence divergence (KA/KS) between Types I and II ATMYC1 alleles; the window size was set at 45 bp, and was moved in 3 bp increments along the length of aligned (inferred) cDNA sequences. These ratio plots were generated in two ways: (1) using local KA over local KS measures and (2) using local KA values over the gene-wide KS value. While the former method is the convention, the latter has been suggested as an alternative to deal with false or misleading positives caused by very low local KS values [65]. For each window of sequence the KA/KS ratio was calculated using both methods and the results were plotted using sigmaplot (Systat Software, Inc.).
The finding of two highly diverged allele types at the ATMYC1 locus prompted an investigation of the potential effect of this sequence divergence on trichome density. To this end, trichome density was scored on fifth true leaves for a set of 96 ecotypes of A. thaliana (details on this set of ecotypes can be found in [48]) following the methods of Symonds et al. [40]. The ecotypes were screened for ATMYC1 Type using a PCR scheme with Type-specific forward and reverse primers that terminate on multiple sites that are polymorphic between the two Types; that is, only one set of primers yields a product for each ecotype, thus distinguishing the two Types. The trichome data were partitioned into the two allele classes and because the data showed a bimodal distribution, a Kruskal-Wallis rank sum test was performed using mapqtl [66] to test for a significant difference in trichome density between the two groups. Although association mapping in A. thaliana ecotype collections is potentially confounded by false positives due to population structure [67], [68], we didn't subsequently correct for population structure given our initial negative result.
Outside of the variation that distinguishes the two ATMYC1 Types (Col and Ler possess alternative Types), four amino acid replacements differentiate the Col and Ler alleles. To assess the conservation of these four positions, the Col ATMYC1 protein sequence was submitted to a protein BLAST search and the top 100 hits were aligned and conservation at each of the four sites that differ between the Col and Ler alleles was evaluated in this alignment.
Based on the placement and conservation of polymorphisms between the Col and Ler alleles, two amino acid positions were selected to test for interaction effects with known partners, TTG1 and GL1: A13T and P189A (Col:aa position:Ler). ATMYC1 cDNAs were cloned by Reverse Transcription and PCR amplified using start to stop gateway primers and recombined into pDONR/Zeo (Invitrogen). These clones were then modified using Stratagene's QuikChange XL Site-Directed Mutagenesis Kit as recommended by the manufacturer. The Col cDNA was modified to make a version with a T13A change, one with a P189A change and one with both changes. The Ler cDNA was modified to make a clone with an A13T change, one with an A189P change and one with both. Each of these clones was then recombined into the yeast two-hybrid DNA binding vector pGBT9 RFB. The WDAD (TTG1A) and SRV6 (pGL1A) activation domain vectors were described previously [44]. All clones were sequenced in their entirety. The ATMYC1 yeast vectors were transformed into the Y190 yeast strain. WDAD and SRV6 were then transformed into each of the ATMYC1 yeast lines. The yeast two-hybrid assay was performed as previously described [44] using X-gal as a substrate for β-galactosidase activity and growth on histidine dropout media as interaction markers.
|
10.1371/journal.pcbi.1005696 | Cost-efficient vaccination protocols for network epidemiology | We investigate methods to vaccinate contact networks—i.e. removing nodes in such a way that disease spreading is hindered as much as possible—with respect to their cost-efficiency. Any real implementation of such protocols would come with costs related both to the vaccination itself, and gathering of information about the network. Disregarding this, we argue, would lead to erroneous evaluation of vaccination protocols. We use the susceptible-infected-recovered model—the generic model for diseases making patients immune upon recovery—as our disease-spreading scenario, and analyze outbreaks on both empirical and model networks. For different relative costs, different protocols dominate. For high vaccination costs and low costs of gathering information, the so-called acquaintance vaccination is the most cost efficient. For other parameter values, protocols designed for query-efficient identification of the network’s largest degrees are most efficient.
| Finding methods to identify important spreaders—and consequently protocols to identify individuals to vaccinate in targeted vaccination campaigns—is one of the most important topics of network theory. Earlier studies typically make some assumption about what information is available about the contact network that the disease spreads over. Then they try to optimize an objective function—either the average outbreak size in disease simulations, or (simpler) the size of the largest connected component. For public-health practitioners, gathering the network information cannot be detached from the decision process—their cost function includes the costs for both the vaccination itself and mapping of the network. This is the first paper to evaluate the cost efficiency of vaccination protocols—a problem that is much more relevant and not so much more complicated, than the oversimplified objective functions optimized in previous studies. We find a “no-free lunch” situation, where different protocols proposed in the past are most efficient at different cost scenarios. However, some methods are never cost efficient due to the amount of information they need. What protocol that is the best depends on network structure in a non-trivial way. We use both analytical and simulation techniques to reach these conclusions.
| Infectious disease is a major burden to global health. Infections spread from person to person over human contact networks. The propagation speed is an emergent property of both the pathogenesis in the infected individual and the contacts between people. By understanding the contact networks, we should thus be able to better predict and mitigate disease outbreaks. These are the premises of network epidemiology [1, 2]—one of its most active questions being how to exploit the contact network in targeted vaccination campaigns [3, 4]. Until now, targeted vaccination has mostly been a theoretical topic. The medical practice of network-based immunization has been very limited to both few cases and simple methods—the most famous being “ring vaccination” [5]. This strategy was used to eradicate smallpox and works by vaccinating all network neighbors of an infectious person [6]. Nevertheless, network immunization could be important in future disease control, especially for sexually transmitted infections (where the network links are evident) [7] or livestock diseases (where one node is a farm and links are connections by transport) [8].
In the theoretical literature, the problem of targeted vaccination has typically been formulated as follows. Given some knowledge of the contact network, identify the individuals that are potentially most important for disease spreading. To carry out a targeted vaccination campaign, one would first need to gather information about the contact network, then use this information to vaccinate (or otherwise reduce the impact of the important individuals). There are thus three major costs involved in such an endeavor: the cost of the disease itself (that we use as our base unit), the cost of gathering the information about the network cinfo (in units of the cost of a person getting the disease) and the cost of vaccinating cvacc. We can thus evaluate the cost efficiency of a vaccination protocol by measuring the net saving χ per person in units of the cost of sick individuals
N χ ( f ) = Ω - Ω ′ ( f ) - N f c vacc - n ( f ) c info (1)
where Ω and Ω′ are the expected outbreak sizes (number of individual who had the disease after it became extinct) respectively without and with using vaccinations, N is the number of individuals, f is the fraction of individuals to vaccinate and n is the number of inquiries needed to obtain information. Obviously, Ω corresponds to the no-vaccination scenario and thus does not depend on f. Both n and Ω′ depend on the specific vaccination protocol, but we drop this information in Eq (1) for brevity.
By reformulating the vaccination problem as a cost-optimization problem, one can evaluate the protocols proposed in the literature in a way more useful for decision makers. In this paper, we use this approach to evaluate eight vaccination protocols for many kinds of cost scenarios and underlying networks. We use eight different empirical networks of human contacts (representing sexual interaction or proximity). We also use the configuration model—a popular method to generate synthetic uncorrelated random networks given a degree sequence.
Before proceeding to the details of our approach we will give a brief overview of the recent analytical advances on the vaccination problem. The simplest vaccination protocol is just to vaccinate random individuals—the Random (R) protocol—which often serves as a baseline in the literature, see e.g. Refs. [9–12]. In a seminal paper, Cohen, Havlin and ben Avraham [9] proposed the more effective Acquaintance (A) vaccination. In their approach, one also starts by randomly selected individuals, but does not vaccinate these, rather, asks them to name someone they met (in such a way that contagion could occur). In an uncorrelated network, the probability of meeting a node of degree k in such an approach, is proportional to k. It is important to vaccinate high-degree nodes, not only because they have more people to spread the disease to, but also more people to get the disease from.
Let fc denote the fraction of population that must be vaccinated in order to prevent a global outbreak. Formally, as N → ∞, fc = inf{f: Ω′(f)/N = o(1)}, and we will use a superscript for fc to denote a specific vaccination protocol. It was shown numerically in Refs. [9, 10] that f c A < f c R. An implicit analytical expression for f c A in uncorrelated networks (configuration model) was derived in Ref. [10]. Similar results were obtained in Ref. [11] for a more general model of infection spreading, in Ref. [13] for imperfect vaccine, and in Ref. [14] for the weighted configuration model, where weights of the edges represent contact probabilities.
A large empirical study based on the 2006 census of the Greater Toronto Area [12] suggests that vaccination of top-degree nodes—the Degree (D) vaccination protocol—is most effective. However, such strategy requires information about the entire network, which makes it hard to implement. For analytical results on degree-based vaccination and an implicit expression for f c D, we refer to Ref. [11]. In this paper by optimizing Eq (1) rather that fc, we confirm that the Degree protocol is never most efficient—in all scenarios, the cost of the complete knowledge does not justify the gain in Ω′. Ref. [15], like us, considers vaccination as a cost problem but does not consider the cost of information gathering. They find that the picture of Ref. [12] needs to be modified if one balances the cost of vaccination and treatments so that it is beneficial to vaccinate lower degree nodes.
In addition to the Acquaintance protocol, we consider two strategies, recently developed for quick detection of high-degree nodes: the Random walk (RW) strategy [16], and the Two-step heuristic (TSH) [17]. We also consider two other protocols that require complete knowledge of the network—Coreness and Collective influence (CI). See below for a complete description of all protocols.
In this section we introduce the methods, data sets and network models we use.
We assume that an infectious disease is spreading over a static contact network represented as a graph G = (V, E). V is the set of N vertices, or nodes, representing individuals; E is the set of M undirected edges representing pairs of individuals between whom the disease may spread. The nodes are, at any given time, in one of three states—susceptible (S), infectious (I) or recovered (R). Susceptible nodes do not have the disease, but can get it. Infectious nodes have the disease and they can spread it. Recovered nodes do not have the disease and cannot get it. We assume a disease outbreak starts at time t = 0. At the beginning all nodes are susceptible, except a randomly chosen node that is infectious. If an edge is between one susceptible individual and one infected individual, then the susceptible becomes infectious at rate β. Every infectious recovers at rate ν. In this setting, the infection and recovery times are independent exponential random variables, and an infectious node transfers a disease through an edge before getting recovered with probability β/(β + ν).
The SIR model is essentially determined by the ratio between β and ν. In the well-mixed, differential equation version of the SIR model, this ratio is called R0. The actual values of β and ν are only needed to calculate the time to reach the peak prevalence, extinction etc. In this paper, we set ν = 1 which is equivalent to saying that we are measuring the time in units of 1/ν. In order to simulate this model, it is efficient to perform one infection or recovery event every iteration of the algorithm. The probability of the next event being an infection is
β M SI β M SI + N I , (2)
where MSI is the number of edges between infectious and susceptible individuals, and NI is the prevalence (number of infectious individuals [18]). The time increment since the last iteration is, on average, 1/(βMSI + NI). If an infection event is not performed, one performs a recovery event. In an infection event, the S-I edge is chosen randomly among all S-I links. Similarly, in case of a recovery event, the infectious individual (to recover) is selected uniformly at random among all infectious individuals.
For all contact networks and parameter values, we use 300,000 or more runs of the SIR model for averages. Since each run represents an independent realization of the (random) costs of the entire process, we used the normal approximation of the sample average, to verify with 99% confidence intervals that our evaluations of mean costs were very accurate. The exact values of the confidence intervals are not informative for our purpose and thus are omitted. We use β = 1/32, 1/16, 1/8, 1/4, 1/2, 1, 2, 4, 8, 16, 32 and (as mentioned) ν = 1.
We compare the performance of seven vaccination protocols—five of these have been analyzed in the literature, and two are proposed by us in this work (but derived from a cost-efficient way of finding the highest degree vertices). The vaccination protocols range from simple to complex and use different amount of information about the network.
Ideally, the underlying network of our study should be as realistic as possible (given a pathogen). Our knowledge of the structure of contact networks is advancing, and there are some datasets available. We use the ones that record actual contacts between people and disregard those where contacts are inferred from interaction on social media, etc. [24]. To better understand how the size of the network, and higher-order structures, affect the performance of the algorithms, it is desirable to have models able to generate contact networks. We study one of the simplest such models—the configuration model—not because it is able to generate a network with very realistic structure, but because it enables us to compare the result to other studies, in particular analytical ones.
We start by evaluating the vaccination protocols in some detail for the Colorado Springs data set. Then we proceed to take a cruder look at all the data sets to see how network structure affects the results.
We have discussed how to make theoretical studies of targeted vaccination more practically useful for decision makers. Instead of evaluating vaccination protocols for some scenario about what is known about the network, we evaluate methods based as a cost-benefit problem. From this starting point, we have evaluated the cost efficiency of seven network-based vaccination methods. There is not one universally best method. Rather, depending on the network structure and relative vaccination and information costs, the best method (at least for the network and parameters we explore) seem to be one of four—Acquaintance, TSH, CI and Random walk. We make this point both by analytical calculations and simulations.
Acquaintance vaccination is almost always the most efficient for low cvacc and large cinfo. It is the protocol that uses second least network information after Random. For very high cinfo, Random will trivially be the most efficient (keep in mind that cinfo can, in principle, be larger than one), but we never observe this. TSH dominates the region of large cvacc and low cinfo, for denser networks (for very sparse networks CI could also be most efficient). Random walk dominates intermediate values of cvacc and cinfo. Something that we find hard to rationalize and leave to future investigations.
CI performs well for very sparse networks with few triangles, especially in the region of large cvacc and low cinfo. Degree is never most efficient, meaning that vaccinating exactly in order of degree is not so important that it is worth obtaining all the network information. Furthermore, Coreness is also never most efficient, supporting Refs. [39] and [23] (but disagreeing with Ref. [21]).
In practical applications, one would in principle need to know the parameters, both for the SIR model and to calculate the cost [4]. For e.g. sexually transmitted diseases, this is not impossible. If one, would base a pilot HIV pre-exposure prophylaxis campaign on mapping a sexual network like Ref. [28] (which, in addition to the network itself, could give the contact rates), then one could assume a per-contact transmission probability of 1–2% [40]. Furthermore, the societal cost for a positive HIV case is well understood [41]. With these parameters at hand, it should be possible to narrow down the protocols to one or two.
To proceed towards increasing realism and applicability, one would also need to take social mechanisms into account. Parallel to the targeted immunization problem, there is an emergent field studying vaccination as a social-psychological problem. One issue being that for voluntary vaccination it is irrational to become vaccinated if almost everyone else is vaccinated (the diseases would not spread anyway, and there are side-effects and discomfort associated with being vaccinated). Conversely, it is irrational not to vaccinate if almost nobody is vaccinate, leading to a typical game theoretical dilemma [42]. Another issue in this direction discusses how the awareness of a disease spreading affect the contact networks, and subsequently the spreading dynamics [43]. Or how vaccination and awareness diffusion can create synergistic effects [44]. Other papers study how social influence affects the decision to vaccinate ones children (e.g. Ref. [45]). To make theoretical vaccination studies fully realistic and most useful to decision makers, one would need combine such social aspects with the cost-benefit approach of this paper.
|
10.1371/journal.pntd.0007271 | Rapid immunochromatographic tests for the diagnosis of chronic Chagas disease in at-risk populations: A systematic review and meta-analysis | Despite of a high disease burden, mainly in Latin America, Chagas disease (CD) is underdiagnosed and undertreated. Rapid diagnostic tests (RDTs) might improve the access to diagnosis. The aim of this study is to review the accuracy of commercially available RDTs used in field conditions for the diagnosis of chronic CD in populations at risk, in endemic and non-endemic countries.
We undertook a comprehensive search of the following databases: PubMed, SCOPUS, LILACS (last up-date on the 01st July, 2017), without language or date limits. Non-electronic sources have been also searched. This review included clinical studies with cohort recruitment of individuals at risk of T. cruzi exposure, without age limits; adequate reference standards for the diagnosis of CD. We excluded case-control studies and those testing RDTs during acute CD. Data on test accuracies were pooled through a bivariate random-effects model. Only one index test was evaluated separately. Geographical area, commercial brand, disease prevalence, study size, and risk of bias were explored as possible source of heterogeneity. Values of sensitivity and specificity were computed to obtain summary positive/negative likelihood ratios, and summary diagnostic odds ratio. Ten studies were included on six different immunochromatographic RDTs. The pooled sensitivity and specificity of the RDTs resulted 96.6% (95% CI 91.3–98.7%) and 99.3% (95% CI 98.4–99.7%), respectively. Test accuracy was particularly good in endemic areas (98.07%/99.03% of sensitivity/specificity, respectively). One test (Stat-Pak) showed an overall sensitivity of 97% (95% CI 87.6–99.3) and specificity of 99.4% (95% CI 98.6–99.8).
RDTs demonstrated to be sufficiently accurate to recommend their use for screening in endemic areas, even as stand-alone tests. This approach might increase the accessibility to the diagnosis. However, an additional confirmatory test in case of positive result remains a prudent approach.
| Chagas disease (CD) is a parasitic disease that can affect seriously the health status of affected individuals. People with CD, mainly living in remote areas of Latin America, often face major barriers to the disease recognition, diagnosis and treatment. The World Health Organization recommends the combined use of two tests for diagnosis of the disease in the chronic phase, but this approach is expensive, has time-constraints, and requires well-equipped laboratories, among others. Rapid diagnostic tests (RDTs) are easy-to-use, cheaper and less time-consuming than classical techniques. Hence, their large-scale use could contribute to increase the access to diagnosis, improve treatment coverage, and reduce disease transmission. We reviewed the existing studies on the accuracy of RDTs for the diagnosis of chronic CD. The RDTs under study demonstrated sufficiently good to recommend their use for screening in endemic areas (particularly the Gran Chaco), even as stand-alone tests. On the other hand, not enough evidence has been retrieved on the use of RDTs in other settings. The use of RDTs might increase the access to the diagnosis, particularly in the Gran Chaco area of Latin America.
| Chagas disease (CD) is a parasitic disease affecting more than 8 million people and causing 806,170 DALYs lost, annually, in the endemic countries of Latin America (LA)[1]. It is caused by the protozoan parasite Trypanosoma cruzi, generally transmitted by insect vectors. Following international migration, the disease has spread also to non-endemic countries, where it can be transmitted congenitally or through organ or blood donation[2]. It has been recently estimated that in Northern America (Mexico, United States and Canada) from 1.3 to 7 million people could be affected[3].
CD has been associated to poverty as it causes relevant morbidity and mortality in working-age people and predominantly affects disadvantaged populations[1]. Moreover, the transplacental transmission[4] causes abortion, stillbirth and complications in newborns. In 30–40% adults, it evolves towards potentially fatal complications after decades of silent progression[2]. Infections in the acute phase and, to a lesser extent the chronic one, can be treated with the aim of cure and interruption of transmission, or at least a reduction in the risk of morbidity[5].
Globally, the disease is largely under-diagnosed (an estimated 90% affected people are unaware of their infection and thus at risk of transmitting it and suffering complications[6]) and under-treated (less than 1% of affected individuals have access to treatment[7]). The World Health Organization (WHO) recommends that the diagnosis of chronic CD should rely on concordant results of at least two different serological tests based on different antigens[8]. Traditionally, conventional tests based on crude antigens/parasite lysate (enzyme-linked immunosorbent assay, ELISA; immunofluorescence test, IFAT; indirect hemagglutination test, IHA) are paired with non-conventional ones (mainly ELISAs) based on recombinant antigens[9]. Blood culture and polymerase chain reaction (PCR) are not considered sufficiently sensitive for the diagnosis during the chronic phase due to the intermittent and low-level peripheral parasitemia found throughout this period[10].
Rapid diagnostic tests (RDTs) are easy-to-use and less technically and time demanding than classical serological techniques. Remarkably, many of them can be performed on serum or with a very little volume of whole blood, and they can be stored on the shelf for longer than a year. Their large-scale use could contribute to increased access to diagnosis, better treatment coverage, and a reduction of disease transmission. Yet, despite having commonly been used for field surveys, RDTs are not recommended by the WHO[11–20].
The aim of this study is to review the accuracy of RDTs in field conditions for the diagnosis of chronic CD in populations at risk living in endemic and non-endemic countries.
The protocol was registered with Prospero International prospective register of systematic reviews (record: CRD42016025990) on May 6th, 2016.
We searched PubMed, SCOPUS, LILACS on 26th November 2015 and up-dated the search on 01st July 2017, without language or date limits. Original search strategy is available in Prospero (https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=25990). Non-electronic sources have been also searched, like references listed in included studies or non-published data from expert in the field.
a) clinical studies with cohort recruitment (phase III studies) in field conditions; b) the presence of adequate reference tests (from now called "Reference Standard", RS) for the diagnosis of CD, namely a combination of two (or more) serological tests based on different techniques (either ELISA, IFAT or IHA) and antigens according to current WHO recommendations[8], or one or more high specificity test such as radioimmunoprecipitation analysis—RIPA or immunoblot or western blot, or the use of latent class analysis (LCA)-based reference standard; c) studies conducted on individuals (adults or children) with epidemiological risk of exposure to T. cruzi such as living in endemic area for at least one month, receiving blood transfusion in an endemic country or being born to a Latin American mother. We classified studies, on the basis of the sampling method, as being consecutive or non-consecutive. Case-control studies and those testing RDTs during the acute infection phase were excluded.
Two authors independently selected the studies, on the basis of the inclusion criteria. In case of discordant opinion, a third author was involved. Data were extracted from selected studies and risk of bias was assessed through the QUADAS-2 tool[21]. As possible sources of heterogeneity, we explored: geographical area, commercial brand of index test, type of RS, disease prevalence, study size, and risk of bias.
The values of sensitivity and specificity were automatically computed in RevMan 2014 (Version 5.312). Individual study results were graphically expressed by plotting the estimates of sensitivity and specificity and their 95% confidence intervals (CIs) through both forest plots and receiver operating characteristics (ROC) space. We assessed heterogeneity by visual inspection of forest plots of sensitivity and specificity, and through visual examination of ROC plot of the raw data. Heterogeneity was further investigated using a bivariate random-effects model[22] to obtain estimates of the between-studies variation in sensitivity and specificity and the correlation between the two. The same bivariate model was used to assess the operating point sensitivity and specificity of the diagnostic tests under scrutiny, together with likelihood ratios and summary diagnostic odds ratio (DOR), taking both heterogeneity and threshold effect into account.
All analyses were performed using all articles first, then they were repeated splitting the studies into two main subgroups: studies conducted in endemic areas (continental LA), and studies in non-endemic areas (other continents). This was considered the primary analysis. Based on the results of included studies, we further conducted a secondary analysis on datasets evaluating the RDT most frequently used, i.e. Stat-Pak. All analyses were performed using Stata IC 13.0.
The electronic search identified 4574 records. The study flow is summarized in Fig 1.
Amongst the 151 articles included for full text evaluation, we excluded from the analysis studies on RDTs not based on immunochromatographic technique, because the 25 identified studies using other techniques were either evaluating old, out-of-market tests or were old studies for which we could not get access to the full text article[23–47]. Moreover, 22 papers were excluded on the base of the study design[48–69]. Eighty-two articles were excluded because they did not evaluate RDTs accuracy. Three papers were not eligible for inadequate target population (not chronic CD)[70–72], whereas 8 papers declared an inadequate reference standard for inclusion in the review[73–80].
“Eventually, 9 and 10 studies were included in the quantitative and qualitative analyses, respectively”. Two studies evaluated more than one single RDT, hence each dataset from those studies was considered separately[18,19]. One study[81] reported the results of a test conducted on a relatively small number of patients (114 children) that only had true negatives, providing no information on sensitivity. Thus, as we intended to model sensitivity and specificity jointly, we decided to exclude it from our analyses.
Finally, we included in the analysis 12 datasets, comprising a total of 6123 participants (from 101 to 1913 individuals tested in single studies). Six different RDTs were evaluated; Stat-Pak was assessed on more than half of the overall population included in the analysis (4 studies comprising 3347 participants, 53.3% of individuals included in our meta-analysis). Studies evaluating Stat-Pak were quite heterogeneous in terms of age-range of the selected population: one study included participants with more than 16 years of age; Bern et al. enrolled adults, Roddy et al. children and adolescents, and Eguez et al. individuals of all ages[11,18,19]. However, the participants resulted rather homogeneous in terms of origin, as three studies were conducted in Bolivia, and the study implemented in Europe comprised 47.4% of immigrants of Bolivian origin (474 of 999 participants)[12].
Among the other RDTs, one (Simple CHAGASWB, Operon) was evaluated by two studies with a total of 377 participants[15,20]. Both studies were conducted in Spain, and the reported prevalence of CD was 15.9% in the study by Navarro et al. (all ages, 76.4% of participants coming from Bolivia) and 5.9% in the work by Lopez-Chejade et al. (Latin American adults). Simple CHAGASWB showed a sensitivity/specificity of 88%/94.2% and 100%/96.8%, respectively, in Navarro et al. and Lopez-Chejade et al. studies.
Eguez and colleagues assessed the accuracy of a combination of two RDTs (Stat-Pak and InBios) against conventional tests (namely, IHA, lysate-antigen ELISA, and recombinant antigen ELISA)[19]: for the purpose of this review the study was split into two datasets, each one evaluating one RDT, while the combination of Stat-Pak and InBios was not included in the analysis.
Table 1 shows data and characteristics of each RDT and the reference standard for each study.
Four studies were conducted in non-endemic areas (namely, Spain, Switzerland, and Italy)[12,14,15,20]; all studies conducted in LA were carried out in Bolivia, but one that was conducted in Argentina[13]. A couple of studies were conducted in a cohort of children while the others included either adults or individuals of all ages.
The qualitative evaluation, in terms of rating for each study finally included in the analysis, and their overall methodological quality are shown in Fig 2a and 2b, respectively.
In general, the risk-of-bias and applicability concerns of all studies analyzed were considered low. Patient selection was assessed as “unclear” in four cases: the main reason was that the papers did not specify methods for the enrollment of patients (consecutive recruitment or random inclusion). Moreover, one study[18] did not clearly state if the results of the index tests were interpreted without knowledge of the results of the RS, hence the risk of bias in relation to the index test was assessed as unclear[18].
Fig 3 shows the accuracy of the RDTs according to each dataset. Notably, heterogeneity among results of different studies was low, particularly in terms of specificity. Namely, the variance of the logit of the sensitivity resulted 1·82 (95% CI: 0·55 to 5·00), whereas the variance of the logit of specificity was 1·01 (95% CI: 0·29 to 3·41). The correlation between logit of sensitivity and logit of specificity resulted 0·34 (95% CI: -0·50 to 0·86). Globally, the accuracy of all RDTs resulted in: sensitivity = 96.6% (95% CI: 91.3–98.7%) and specificity = 99.3% (95% CI: 98.4–99.7%) (Table 2).
The RDTs showed better accuracy when used in endemic areas (Table 2): 98.1% and 99.3% respectively averaged sensitivity and specificity, whereas in non-endemic areas their sensitivity resulted lower: about 90%. The overall sensitivity of Stat-Pak was 97% (95% CI 87.6–99.3) and its specificity 99.4% (95% CI 98.6–99.8).
Globally, the sensitivity of the RDTs examined was good (higher than 95%) and the specificity was excellent (>99%), regardless their use in endemic or non-endemic regions. The sensitivity was basically higher in endemic (namely Bolivia) than in non-endemic areas. However, it must be noted that the latter data was obtained from four studies comprising 2063 individuals (around one third of the whole study population), with a lower proportion of Bolivians (35,9%, excluding Lopez-Chejade study, where the origin of migrants is not detailed) [12,14,15,20]. A previous study comparing 11 marked-available RDTs found out 8 tests which were considered valuable for clinical purpose (performances generally >90%). However, this was a case-control study based on selected serum samples, hence the accuracy of the tests could be overestimated [66].
In our work, among all RDTs Stat-Pak could be evaluated individually thanks to the high number of individuals tested. This test showed high accuracy, and its functionality with a little volume (10 μl) of whole blood further supports its use for screening purposes, as well as in field surveillance of the disease.
On the other hand, the number of studies (and participants included) addressing the other RDTs was too low to allow a separate meta-analysis. It must also be considered that T. cruzi population is characterized by a genetic polymorphism that might account at least in part for its variability in pathogenicity and transmission. Currently, seven genetic lineages or Discrete Typing Units (DTU) have been characterized, TcI-TcVI and Tcbat[82,83]. Their distribution varies geographically. Most studies included in this work concerned nationals of Bolivia or neighbouring countries (mainly the Gran Chaco area), where the the TcV prevails[83]. Different DTUs are prevalent in other regions (i.e. Mexico and Central America), with different antigenic features, tissue tropism, and pathogenicity profile. Therefore, our results cannot be automatically transferred to other CD epidemiological contexts. Moreover, all studies, both in endemic and non-endemic countries, concerned populations with high prevalence of T. cruzi infection. In populations with lower prevalence, our findings may not be entirely applicable.
On the other hand the Reference Standards, although based on different tests, were generally similar across selected studies: for all 12 datasets at least two paired tests were used (in two studies three tests were used and case definition was based on at least two concordant results, however)[18,19]: this permitted an accurate comparison between studies. We formally assessed risk of bias through the QUADAS-2 tool, and most of the selected studies received high scores, which further contribute to the robustness of our analysis. Finally, the assessment of variation in sensitivity and of the degree of correlation between sensitivity and specificity provided evidence of limited heterogeneity among studies. The utilization of statistical techniques that consider heterogeneity and threshold effect for the estimation of summary measures, such as the bivariate model suggested by Reitsma et al.[22], allowed the achievement of exhaustive and robust estimates.
All studies included in our review were conducted under real-life conditions on populations at risk of having chronic CD. The practical implications of this statement are better summarized in Table 2. The use of a RDT would appear more appropriate for endemic than non-endemic areas, as in the latter a RDT-based screening would miss about 2 out of 19 infected subjects among 100 individuals tested. In contrast, in the endemic areas the proportion of infected subjects was higher (36 subjects with CD out of 100 tested), but less than one of the 36 would be missed by the RDT. On the other hand, in all contexts, the number of non-infected subjects erroneously found positive would be absolutely negligible, and the only significant consequence (in case a second, confirmatory test was not performed) would be to propose an unnecessary treatment. Considering only Chagas Stat-Pak test, the figures would be very similar, although in this case a comparison between endemic-non endemic areas was not possible.
While no RDT is sensitive enough to recommend its use for blood/organ donor screening, the technique appears to be sufficiently accurate for the screening of individuals at risk who could benefit from treatment. Considering that a strategy based on a single RDT would be much easier and cheaper to implement than the classical strategy based on two serological tests, it is very plausible to assume that the few cases missed (RDT false negatives) would be amply compensated by a larger population screened. This is especially valuable in rural areas of the endemic countries where access to diagnosis may be problematic. On the other hand, all positive individuals to a RDT should be submitted to a confirmatory test whenever possible, in order to avoid the side effects of an unnecessary treatment, as well as the stigma associated to the infection, for a false-positive subject. Alternatively, one of the studies included in this review[19] suggested the combined use of two RDTs as a strategy to increase the accuracy for screening purpose. The combination reached a near-perfect sensitivity (considering at least one positive out of two) and specificity (both positive).
The accuracy of all RDTs under study can be considered sufficiently good to recommend their use in endemic settings, particularly in the Southern Cone of LA, in order to increase access to diagnosis. The Stat-Pak test can be recommended for use in screening surveys when the expected prevalence is moderately high or high, in the setting of Southern Cone or for migrants from that area in case of non-endemic countries[12]. On the other hand, the pooled sensitivity of all RDTs studied resulted too low to recommend them as stand-alone tests for detection of CD affected individuals in a non-endemic context, as a negative result cannot rule out a T. cruzi infection with reasonable certainty.
The WHO’s principle which states the need of diagnosis confirmation through another serologic test remains a prudent approach that should be followed at least for the confirmation of positive results.
Still, further studies conducting head-to-head comparisons of different available RDTs are needed, and it would be particularly important to extend these studies to the Andean countries, the Amazon basin, Central America and Mexico. Similarly, further studies will be required to ensure the applicability of RDTs in non-endemic settings. Robust evidence from studies of high quality is also needed to advocate adequate control policies and quality assurance in endemic countries, mainly in those with lower prevalence of the disease.
|
10.1371/journal.pgen.1002387 | The Fission Yeast Stress-Responsive MAPK Pathway Promotes Meiosis via the Phosphorylation of Pol II CTD in Response to Environmental and Feedback Cues | The RRM-type RNA-binding protein Mei2 is a master regulator of meiosis in fission yeast, in which it stabilizes meiosis-specific mRNAs by blocking their destruction. Artificial activation of Mei2 can provoke the entire meiotic process, and it is suspected that Mei2 may do more than the stabilization of meiosis-specific mRNAs. In our current study using a new screening system, we show that Mei2 genetically interacts with subunits of CTDK-I, which phosphorylates serine-2 residues on the C-terminal domain of RNA polymerase II (Pol II CTD). Phosphorylation of CTD Ser-2 is essential to enable the robust transcription of ste11, which encodes an HMG-type transcription factor that regulates the expression of mei2 and other genes necessary for sexual development. CTD Ser-2 phosphorylation increases under nitrogen starvation, and the stress-responsive MAP kinase pathway, mediated by Wis1 MAPKK and Sty1 MAPK, is critical for this stress response. Sty1 phosphorylates Lsk1, the catalytic subunit of CTDK-I. Furthermore, a feedback loop stemming from activated Mei2 to Win1 and Wis4 MAPKKKs operates in this pathway and eventually enhances CTD Ser-2 phosphorylation and ste11 transcription. Hence, in addition to starting meiosis, Mei2 functions to reinforce the commitment to it, once cells have entered this process. This study also demonstrates clearly that the stress-responsive MAP kinase pathway can modulates gene expression through phosphorylation of Pol II CTD.
| Hundreds of genes are newly expressed during meiosis, a process to form gametes, and the control of meiosis-specific gene expression is not simple. The master regulator of meiosis in fission yeast, Mei2, blocks an RNA destruction system that selectively degrades meiosis-specific mRNAs, highlighting the importance of post-transcriptional control in meiotic gene expression. Here we present another example of unforeseen regulation for meiosis. Ste11 is a key transcription factor responsible for the early meiotic gene expression in fission yeast. The ste11 gene is transcribed robustly only when serine-2 residues on the C-terminal domain (CTD Ser-2) of RNA polymerase II are phosphorylated. We show that the stress-responsive MAP kinase cascade transmits the environmental signal to stimulate CTD Ser-2 phosphorylation. Sty1 MAP kinase appears to phosphorylate and activate the catalytic subunit of CTDK-I, which in turn phosphorylates CTD Ser-2. We demonstrate further that Mei2, expression of which depends on Ste11, can activate the MAP kinase cascade, forming a feedback loop. Thus, we clarify here three important issues in cellular development: the physiological role of CTD Ser-2 phosphorylation, the molecular function of the stress-responsive MAP kinase pathway, and the presence of positive feedback that reinforces the commitment to meiosis.
| The cell cycle programs for mitosis and meiosis appear to be strictly segregated from each other, although they are likely to have molecular mechanisms in common. Analyses in lower eukaryotes have shown that factors required exclusively for meiosis, generated through the transcriptional activation of meiosis-specific genes, are largely responsible for the segregation of these two processes [1], [2]. In addition, we have reported previously in fission yeast that meiosis-specific mRNAs transcribed at the wrong time during the mitotic cell cycle are removed selectively by nuclear exosomes, thereby preventing the inappropriate expression of the meiotic program in mitotic cells [3], [4]. The master meiotic regulator in fission yeast, Mei2, an RNA-binding protein with three RRM domains [5]–[7], suppresses the function of this selective removal system by sequestering a key component Mmi1, which is an RNA-binding protein of the YTH family [3]. Mei2 thus ensures full expression of meiosis-specific genes and facilitates execution of the meiotic program (reviewed in [8]). However, it is unlikely that the function of Mei2 in meiosis is confined to the tethering of Mmi1 as the artificial inactivation of Mmi1 does not induce the full meiotic program, whereas the experimental induction of the activated form of Mei2 does so [3], [6]. The mechanisms and pathways by which Mei2 promotes the entire meiotic program is therefore a subject of considerable interest.
To identify possible upstream or downstream effectors of Mei2, we devised a new screening system and found that a subunit of CTDK-I, which is a CDK-like kinase complex that phosphorylates the C-terminal repeat domain of the largest subunit of RNA polymerase II (Pol II CTD) [9], [10], could genetically interact with Mei2. More specifically, the phosphorylation of Pol II CTD by CTDK-I was found to affect the expression of ste11, which encodes a key transcription factor that regulates the mei2 gene. Pol II CTD serves as a binding scaffold for a variety of nuclear factors, and its phosphorylation status has been implicated in regulation of an ever-increasing number of functions necessary to execute complex transcriptional processes [9], [10]. Our aforementioned findings indicate that the phosphorylation of Ser-2 residues on Pol II CTD in fission yeast is unique in that it is required mainly for the meiotic program, via the activation of ste11 transcription, but is not absolutely necessary for the mitotic program. Essentially the same conclusions have been reached independently by others, through global gene expression analysis [11]. Here we further show that the stress-responsive MAP kinase cascade is crucial for the phosphorylation of Ser-2 residues under nutrient starvation, which is a condition suitable for meiosis. We also show that artificially activated Mei2 has the potential to promote the phosphorylation of Ser-2 residues on Pol II CTD via the stress-responsive MAP kinase cascade, irrespective of the nutrient conditions.
Taken together, the results of our present study demonstrate a new regulatory paradigm for meiosis by Mei2 in fission yeast, i.e., that this master meiotic regulator ensures the commitment to meiosis by strengthening the transcription of ste11 via a feedback loop comprising the stress-responsive MAP kinase cascade and the phosphorylation of Pol II CTD by CTDK-I.
The haploid fission yeast strain JV312 harbors the mei2-L-SATA allele driven by the authentic mei2 promoter. This allele contains a combination of two mutations, mei2-L and mei2-SATA. The former mutation confers temperature-sensitivity to the Mei2 protein (our unpublished results), whereas the latter activates this gene constitutively, overriding the inhibitory phosphorylation by Pat1 kinase [6]. JV312 cells arrest during vegetative growth and induce ectopic meiosis at 25°C because the Mei2-L-SATA protein is functional at this temperature. However, these cells continue vegetative growth at 32°C because Mei2-L-SATA is then inert and does not interfere with cell growth pathways. To identify novel upstream regulators or downstream effectors of Mei2, we screened for suppressor mutants that could grow at 25°C by insertional mutagenesis of JV312 (see Materials and Methods). Several suppressor mutants were thereby isolated, one of which was found to contain an insertion in SPBC4B3.08, which is annotated in the fission yeast database (http://old.genedb.org/genedb/pombe/) to encode a homologue of the γ subunit of RNA polymerase II C-terminal domain kinase I (CTDK-I). CTDK-I belongs to the CDK family, but in addition to the catalytic subunit α and the cyclin-like regulatory subunit β conserved among these family members, it contains a third γ subunit [12], [13]. In fission yeast, the lsk1 and lsc1 genes encode the α and β subunits of the CDK proteins, respectively [14], [15]. Hereafter, we designate SPBC4B3.08 as lsg1.
Because the level of homology between fission yeast Lsg1 and Saccharomyces cerevisiae CTDK-I γ (CTK3) was found not to be high (a 24% amino acid identity; Figure S1), we examined whether Lsg1 was indeed a functional homolog of CTDK-I γ. We constructed the lsg1-deletion strain by replacing the entire lsg1 ORF with a drug-resistant cassette, and compared its phenotype with that of lsk1Δ and lsc1Δ. The lsg1Δ strain exhibited no significant defects in mitotic growth, like the lsk1Δ and lsc1Δ strains previously analyzed [14], [15] (Figure 1A). The doubling time in liquid YE medium at 30°C was 2.1 h for the wild-type, 2.2 h for lsg1Δ, 2.3 h for lsk1Δ, and 2.2 h for lsc1Δ, respectively. However, lsg1Δ cells showed hypersensitivity to Latrunculin A, an inhibitor of actin polymerization, which was a phenotype reported previously for lsk1Δ and lsc1Δ [14], [15] (Figure 1A). In addition, both lsk1Δ and lsc1Δ could suppress the growth defect of mei2-L-SATA at 25°C as efficiently as lsg1Δ (Figure 1B). These observations confirmed that lsg1 indeed encodes the CTDK-I γ subunit, and indicated that loss of CTDK-I activity is responsible for the suppression of mei2-L-SATA.
Although deletion of the gene encoding each CTDK-I subunit led to no obvious defect under normal growth conditions, these deletion mutants all showed impairments in conjugation and sporulation under starved conditions. Under these conditions, haploid lsg1Δ, lsk1Δ or lsc1Δ cells conjugated at a lower frequency than wild-type cells, and diploid lsg1Δ, lsk1Δ or lsc1Δ cells underwent azygotic meiosis and sporulation at a lower frequency than wild-type cells (Figure 1C). We further found that the progression of the meiotic cell cycle was significantly retarded in the CTDK-I subunit mutants. Fluorescence-activated cell sorting (FACS) analysis indicated that diploid lsg1Δ, lsk1Δ or lsc1Δ cells began to arrest in G1 phase as late as eight hours after the shift to nitrogen starvation and showed minimal premeiotic DNA synthesis even after 24 hours. In contrast, wild-type cells began to arrest in G1 phase after two hours and completed premeiotic DNA synthesis at between 2 and 6 hours (Figure 1D).
Our observations that the CTDK-I deletion mutants were defective in sexual development and could suppress growth deficiency, evoked by the mei2-L-SATA allele, led us to speculate that the expression of ste11, which encodes an HMG-family transcription factor, might be impaired in these mutants. Our reasoning was that 1) Ste11 regulates the transcription of many genes essential for sexual development, including mei2 [16]; 2) the deletion of ste11 has been shown to suppress ectopic meiosis induced by the pat1 mutation and restore vegetative growth, by blocking the expression of mei2 [17], [18]; and 3) we had noticed that ste11Δ cells show G1 arrest retardation under conditions of nitrogen starvation, even more extensively than lsg1Δ, lsk1Δ or lsc1Δ cells, while mei2Δ cells are not so much affected (Figure S2A). We thus analyzed the transcription of ste11 in lsg1Δ cells and found that it was significantly suppressed (Figure S2B). Because requirement of CTDK-I for the expression of ste11 has been independently discovered and already reported by Hermand and his colleagues [11], we briefly summarize our data that confirm their conclusions in Figures S2 and S3. We tested whether the forced expression of ste11 could recover sexual development in the CTDK-I deletion mutants. The overexpression of ste11 from the nmt1 promoter, which is roughly four to five times as strong as the physiological expression, effectively recovered conjugation and subsequent meiosis in lsg1Δ, lsk1Δ and lsc1Δ homothallic haploid cells (Figure S2C), indicating that the loss of ste11 expression is a major cause of the mating and sporulation deficiency in the CTDK-I mutants. We then determined the range of genes whose expression is regulated by CTDK-I, by comparing the gene expression profiles between lsg1Δ and wild-type cells starved of nitrogen for 2.5 hours. Genome-wide microarray analysis indicated that the expression of 64 genes was downregulated more than two-fold in the lsg1Δ mutant, whereas 22 genes showed upregulation by more than two-fold in the mutant (Figure S3A). Notably, 33 out of the 64 downregulated genes identified, including ste11 itself, have been shown previously to be controlled by Ste11 [19]. These genes are listed in Table S1. In contrast, the expression of atf1, pcr1, rst2, and other genes that also encode an upstream regulator of ste11 transcription [20]–[25], was not significantly affected by the deletion of lsg1 (Figure S3B), suggesting that CTDK-I may exert its effects on ste11 transcription directly.
Previous work has shown that Lsk1 is involved in the phosphorylation of Ser-2 residues within the heptad repeats of the carboxy terminal domain (CTD) of RNA polymerase II [15]. To determine whether the Pol II CTD phosphorylation status might be changed by the induction of sexual development, we analyzed phosphorylation of Ser-2 and Ser-5 residues within the CTD before and after the shift to nitrogen-depleted medium. Extracts were prepared from wild-type and lsg1Δ homothallic haploid cells, either growing or shifted to nitrogen-free minimal medium, and the phosphorylation of CTD was examined using monoclonal antibodies that recognize either phospho-Ser-2, phospho-Ser-5, or unphosphorylated CTD. As shown in Figure 2A, the phosphorylation of Ser-2 residues on the CTD repeats was increased by nitrogen starvation in wild-type cells, but not in lsg1Δ cells. The level of phospho-Ser-5 was unaffected by nitrogen starvation in both strains. These results suggest that nitrogen starvation induces the phosphorylation of CTD Ser-2 residues by CTDK-I.
We next evaluated the possibility that the insufficient phosphorylation of CTD Ser-2 residues in the CTDK-I mutants underlies their sexual development deficiency. For this purpose we examined the phenotypes caused by two rpb1 alleles (reported by J. Karagiannis and kindly provided to us), namely rpb1-12×CTD and rpb1-12×S2ACTD. The former allele produces Rpb1 carrying a CTD that consists of 12 copies of the authentic heptad repeat (YSPTSPS), whereas the latter produces Rpb1 with 12 copies of a mutant heptad repeat in which Ser-2 is substituted by alanine (YAPTSPS) [15]. Wild-type Rpb1 carries 29 repeats of the heptad [26], but the previous work has shown that 12 repeats are sufficient for cell viability [15]. Cells carrying the rpb1-12×S2ACTD allele were impaired severely in terms of conjugation and sporulation (Figure 2B), and the transcription of ste11 was greatly reduced in them (Figure 2C). Furthermore, the sterility of the rpb1-12×S2ACTD strain was rescued, although not completely, by the overexpression of ste11 (Figure 2D). These results strongly suggest that CTDK-I facilitates the transcription of ste11 by phosphorylating Ser-2 residues on Pol II CTD. In general, the rpb1-12×S2ACTD strain showed severer phenotypes than the CTDK-I mutants with regard to sexual development, probably because CTD Ser2 could also be phosphorylated supplementarily by Cdk9 [11].
We wished to determine the mechanism by which nitrogen starvation caused the increased phosphorylation of CTD Ser-2 by CTDK-I. The concentration of CTDK-I subunits per cell was not found to be significantly altered upon nitrogen starvation (Figure S4A). We also measured the levels of Fcp1, a phosphatase that has been shown to preferentially remove phosphate groups from synthetic CTD peptides phosphorylated on Ser-2 [27], [28]. However, the levels of this protein were also not changed significantly upon nitrogen starvation (Figure S4B).
It has been reported in S. cerevisiae that CTD Ser-2 phosphorylation increases both upon heat shock and during the diauxic shift [29]. The phosphorylation of CTD Ser-2 is also known to be elevated by an exposure to hydroxyurea or UV irradiation [30]. We speculated therefore that nitrogen starvation may be recognized as a stress, which could then affect the phosphorylation status of the CTD in fission yeast. We hence examined the possible involvement of Sty1 (also called Spc1/Phh1), a MAP kinase known to be crucial to the response to stress [31]–[33], in CTD phosphorylation. As shown in Figure 3A, the phosphorylation of CTD Ser-2 in response to nitrogen starvation was dramatically reduced in sty1Δ cells compared with wild-type cells. Deletion of the atf1 gene, which encodes a target of Sty1 MAPK, also significantly affected Ser-2 phosphorylation, whereas the ste11 and mei2 genes appeared to be dispensable for this phosphorylation event in response to nitrogen starvation (Figure 3A). Deletion of pcr1, which encodes a bZIP protein that forms a heterodimer with Atf1 [21], [23], did not affect Ser-2 phosphorylation significantly (Figure S5), and produced a much less severe phenotype compared with mutants lacking atf1, as observed previously for other features [23], [34]. The deletion of rst2, which encodes a transcription factor necessary to activate ste11 in response to glucose starvation and cAMP reduction [24], [25], also had no affect on Ser-2 phosphorylation (Figure S5).
We then examined the effects of a forced activation of the Sty1 MAPK pathway, by expressing a constitutively active form of Wis1 MAPKK in the yeast cells. Phosphorylation of Ser-2 was induced by expression of the active MAPKK from a plasmid, even in the presence of ample nitrogen (Figure 3B). However, this ectopic phosphorylation was not observed in lsk1Δ cells (Figure 3B), indicating that the observed phosphorylation was mediated by CTDK-I. These results suggest that the activation of Sty1 MAP kinase in response to nitrogen starvation is pivotal to the promotion of CTD Ser-2 phosphorylation by CTDK-I.
To examine if the stress-responsive MAPK Sty1 directly phosphorylates CTDK-I, we prepared an in vitro phosphorylation system as detailed in Materials and Methods. Each subunit of CTDK-I, namely Lsk1, Lsc1 or Lsg1, was fused with GST, and the fusion proteins were affinity-purified. Pk-tagged Sty1 MAPK (Sty1-Pk) and its kinase-dead form (Sty1-KD-Pk) were prepared respectively from S. pombe strains NJ761 and NJ767, provided kindly by N. Jones, as described previously [34]. The kinase preparation and each GST-fusion protein were mixed and incubated in the kinase reaction buffer supplemented with [γ-32P]-ATP. As shown in Figure 3C, GST-Lsk1 appeared to be phosphorylated by Sty1, although the full-length protein apparently underwent extensive proteolysis and a possible degradation product was the most heavily labeled. GST-Lsc1 and GST-Lsg1, as well as the control GST, did not appear to be a good substrate of Sty1 in this analysis (Figure 3C). To confirm that Sty1 could phosphorylate Lsk1, we divided Lsk1 into two parts, the N- and C-terminal halves, and fused each of them to GST (Figure 3D). These fusion proteins were relatively stable, and when mixed with active Sty1, the N-terminal half was significantly phosphorylated (Figure 3D). Moreover, our preliminary analysis has shown that at least serine 109 on Lsk1, which constitutes a MAPK substrate consensus sequence PGSP, is a preferred phosphorylation site for Sty1 (data not shown). Analysis of Lsg1 dissected into two parts confirmed that it was not likely to be a substrate of Sty1 (data not shown). These results indicate that Sty1 MAPK is likely to phosphorylate Lsk1 directly and thereby activate CTDK-I, which in turn phosphorylates CTD Ser-2 residues.
We made a surprising observation when we analyzed the status of CTD Ser-2 phosphorylation in cells undergoing ectopic meiosis induced by artificial expression of the activated form of Mei2, i.e., Mei2-SATA. As we reported previously [6], these cells underwent meiosis in the presence of ample nitrogen, a condition that does not stimulate the stress-responsive Sty1 MAP kinase cascade. However, the phosphorylation of CTD Ser-2 was observed in these meiotic cells (Figure 4A). Given this finding, we speculated as to whether the phosphorylation of CTD Ser-2 during Mei2-SATA-induced meiosis was dependent on CTDK-I and/or Sty1. We further tested relevant mutant strains and found that the Mei2-SATA-induced Ser-2 phosphorylation was abolished in lsk1Δ and reduced dramatically in sty1Δ, indicating its stringent dependency on both of these factors (Figure 4B). Sty1 has been positioned upstream of mei2 expression in the stress-responsive signal transduction pathway and in cooperation with a chromatin-remodeling factor Atf1, activates the transcription of ste11 [20]–[22]. The produced Ste11 in turn binds to the upstream region of mei2 and activates the transcription of this gene [16]. We thus hypothesized that activated Mei2 can affect its upstream factors through a feedback regulation.
To identify the component of the stress-responsive signaling pathway that is feedback-regulated by Mei2, we examined mutants that are defective in components of the pathway that function upstream of Sty1. Sty1 MAPK is activated by Wis1 MAPKK [31], [32], [35], [36], which in turn is activated by either Wis4/Wak1 MAPKKK or Win1 MAPKK [37]–[39]. A response regulator protein, Mcs4, associates with Wis4/Wak1, and probably also with Win1, to regulate the MAPKKK activity [38], [40]. We investigated the phosphorylation of Ser-2 during Mei2-SATA-induced meiosis in mcs4Δ, wis4Δ, win1Δ, and wis4Δ win1Δ mutant strains, together with control wild-type, lsk1Δ, sty1Δ, and ste11Δ strains. As summarized in Figure 4B, the phosphorylation of Ser-2 was observed in mcs4Δ and ste11Δ cells, indicating that Mcs4 and Ste11 are not directly involved in the feedback activation of Ser-2 phosphorylation. Ser-2 phosphorylation was observed also in the wis4Δ and win1Δ mutants but was found to be greatly reduced in the wis4Δ win1Δ double mutant. These results indicated that the feedback signals from activated Mei2 might ultimately merge with the stress-responsive MAPK cascade at the Wis4/Wak1 and Win1 MAPKKKs, although there could be a third target because Ser-2 phosphorylation was not completely abolished in wis4Δ win1Δ (Figure 4B). We observed that the level of Sty1 MAPK phosphorylation increased during Mei2-SATA-induced meiosis (Figure 4C), which reinforces the presence of a signaling pathway from Mei2 to the MAPK cascade.
To evaluate physiological significance of the feedback, we examined whether activation of Mei2 would result in enhancement of ste11 expression during meiosis. Firstly, we induced ectopic meiosis by shifting the mei2-L-SATA strain from 32°C to 25°C in the presence of rich nutrition. As shown in Figure 5A, expression of ste11 was evident in this strain but not in the wild-type, and this expression was dependent on lsk1. Secondly, we induced ectopic meiosis by shifting the temperature-sensitive pat1-114 mutant from 25°C to 34°C. Again, expression of ste11 was induced significantly in pat1-114 cells under rich nutrition, in an lsk1-dependent manner (Figure 5B). Deletion of mei2 blocked ste11 expression in these cells. The temperature-shift did not induce ste11 expression in wild-type (Figure 5B) or mei2Δ cells (not shown). These results indicate clearly that activation of Mei2 can stimulate expression of ste11 through phosphorylation of PolII CTD.
We finally evaluated the contribution of the feedback regulation to the expression of ste11 during meiosis under physiological conditions. To do so, we used the mei2-FA allele, which produces inactive Mei2 protein [5], [6]. We compared expression of ste11 and mei2 in wild-type and mei2-FA cells subjected to nitrogen starvation. As shown in Figure 5C, the level of ste11 mRNA, normalized by ribosomal RNA, and that of mei2 mRNA also, were higher in wild-type cells than in mei2-FA cells, and the difference became greater in later stages. This suggests that activated Mei2 protein in wild-type cells indeed enhances ste11 expression via feedback.
Taken together, we propose that fission yeast possess a regulatory circuit, as depicted in Figure 5D, which is likely to be crucial in ensuring an irreversible commitment to meiosis and a strict differentiation of the mitotic and meiotic cell cycle programs.
In our present study, we have demonstrated that a genetic interaction exists between the subunits of CTDK-I, a protein kinase complex that phosphorylates RNA polymerase II CTD, and the master meiotic regulator in fission yeast, Mei2. Furthermore, our analyses indicate that a loss of CTDK-I function impairs the transcription of the ste11 gene, which encodes a transcription activator essential for the expression of mei2 and other genes crucial for sexual development. However, this loss of function does not significantly affect the gene expression required for vegetative growth. In an independent study, Hermand and colleagues have performed genome-wide mapping of three kinds of CTD kinases and also of serine 2- and 5-phosphorylated Pol II in fission yeast to investigate the link between CTD phosphorylation and specific cellular events [11]. Consequently they have found that the CTDK-I catalytic subunit Lsk1 and Ser-2-phosphorylated Pol II associate with a rather limited number of transcription units and play only minor roles during vegetative growth, but become essential during sexual development. These authors have further reported that nitrogen starvation enhances recruitment of Lsk1 to the ste11 gene, and remarked that the phosphorylation of CTD Ser-2 plays a highly specialized role in gene regulation in fission yeast, unlike in other organisms, and is virtually confined to the regulation of a single key gene controlling sexual differentiation. Our study fully supports this notion. While a subsequent study [26] suggests that the deleterious effects of loss of Ser-2 phosphorylation on ste11 transcrition can be compensated partially by loss of Ser-7 phosphorylation, the nature of such extreme specification and its evolution is an intriguing enigma.
Our present data have further shown that the stress-responsive MAP kinase pathway is crucial for the activation of CTDK-I under conditions of nitrogen starvation. The requirement for Sty1 MAPK and its target Atf1 for the expression of ste11 has been known for some time [20]–[22], but the details of the molecular mechanisms involved have remained unknown. It now appears that CTD Ser-2 phosphorylation is a key step in the activation of ste11 expression by the Sty1 MAPK cascade. It has been shown that when phosphorylated and activated by Wis1 MAPKK, the Sty1 protein migrates to the nucleus and resides on the promoter regions of stress-responsive genes [31], [34], [41]. This is also the case for the Sty1 ortholog in S. cerevisiae Hog1 [42], [43]. As shown above, Sty1 can directly phosphorylate Lsk1 in vitro. While the phosphorylation of Lsk1 in vivo remains to be confirmed, it appears to be conceivable that Sty1 may also be recruited to the ste11 promoter and phosphorylate CTDK-I staying there, which in turn phosphorylates CTD and licenses RNA polymerase II to transcribe the gene. In this regard, it is noteworthy that hsp9, which encodes a small heat-shock protein [44] and is one of the genes responsible for the “core environmental stress response” or CESR in fission yeast [45], was detected among our possible target genes upregulated by CTD Ser-2 phosphorylation (Table S1). Interestingly, Reiter et al. have shown previously that Sty1 MAPK is recruited to the promoter of hsp9 and other CSRE genes upon osmotic stress in an Atf1-dependent manner, but does not necessarily phosphorylate Atf1 as a substrate [34]. This suggests that ste11 and hsp9 may be similarly regulated by the Sty1 – CTDK-I – CTD phosphorylation system. However, conventional Chip analyses have not provided convincing evidence for the association of Sty1 with the ste11 promoter, and we are conducting further experiments to scrutinize this possible scheme.
The results of our present analyses demonstrate unambiguously that a feedback-regulatory system operates in fission yeast during the meiotic cell cycle. In this feedback loop, the active form of Mei2 can eventually stimulate the stress-responsive MAPKKKs and enhance the transcription of ste11 through the Sty1 – CTDK-I – CTD phosphorylation system. From our findings we can outline a framework of the molecular mechanisms that differentiate the mitotic and meiotic programs in fission yeast as in Figure 5D. However, it remains currently unknown how the RNA-binding protein Mei2 can fulfill such a never-anticipated task and how many steps may mediate between Mei2 and the MAPKKKs, raising another challenging scientific query as represented by the broken line in Figure 5D.
The S. pombe strains used in this study are listed in Table S2. The general genetic procedures used in the S. pombe experiments were as described previously [46]. Complete medium YE, minimal medium SD, minimal medium MM and its nitrogen-free derivative MM-N [47], synthetic sporulation medium SSA [48] were used to culture the cells. Transformation of S. pombe was performed using the lithium acetate method [49].
The ura4+ cassette used for insertion mutagenesis was amplified by PCR using the primers N18AGCTTAGCTACAAATCCCACTGGCT and N18TGTGATATTGACGAACTTTTTGAC (N18: 18b random DNA sequence). The PCR products were then introduced into JV312 (mei2-L-SATA ura4-D18) cells, and transformants were plated onto SD lacking uracil and incubated at 25°C. Colonies were selected, and the site of ura4+ integration was determined via the sequencing of inverse PCR products [50].
Samples were prepared for flow cytometry essentially as described previously [51] and then analyzed using a FACScan (Becton-Dickinson, San Jose, CA).
JY450 (wild-type) and JT659 (lsg1Δ) cells were grown to mid-log phase in MM medium and shifted to MM-N medium. The cells were collected 2.5 h after the shift, and total RNA was extracted as described previously [52]. Data acquisition and normalization were performed by Roche Applied Science, Japan. The microarray data was deposited to the GEO database under the accession number of GSE32516.
Northern blot analysis was performed as described [53]. DNA fragments used to probe for transcripts of ste11, rpb1 and mei2 were labeled with [α-32P] dCTP using random primers.
Cell extracts were prepared and separated essentially as described earlier [54]. Briefly, cells grown to the mid-log phase were shifted to nitrogen-free medium, and sampled at various intervals. Total lysates were extracted and resolved by SDS-PAGE. Immunoblotting was performed using primary antibodies specific to unphosphorylated CTD (8WG16, Covance, Princeton, NJ, used at 1∶2000), Ser-5 phosphorylated CTD (H14, Covance, used at 1∶2000), Ser-2 phosphorylated CTD (H5, Covance, used at 1∶1000), Mei2 (Our lab preparation, used at 1∶1000), the phosphorylated form of Sty1 MAPK (P-p38 MAPK, Cell Signaling Technology, Danvers, MA, used at 1∶500), or GFP (clones 7.1 and 13.1, Roche Applied Science, Indianapolis, IN, used at 1∶1000). As secondary antibodies, donkey anti-rabbit IgG conjugated with horseradish peroxidase (GE Healthcare, Waukesha, WI) was used for the Mei2 and P-p38 MAPK antibodies at a dilution of 1∶2000. Sheep anti-mouse IgG conjugated with horseradish peroxidase (GE Healthcare) was used to detect all other primary antibodies at a dilution of 1∶2000. Immunoblotting with a monoclonal antibodies specific for α-tubulin, either TAT-1 (a gift from Dr. Keith Gull, University of Birmingham) [55], or Clone B-5-1-2 (Sigma Aldrich, St. Louis, MO), was performed as a loading control.
Cells expressing chromosomally tagged Sty1-3Pk (NJ761), or Sty1KD-3Pk (NJ767) were subjected to nitrogen starvation for 1 h. Extracts were prepared, protein immunoprecipitated, and the immuno-complexes tested for kinase activity as described [34]. Affinity purified GST-fusion proteins were used as substrates.
|
10.1371/journal.pntd.0006073 | Health seeking behavior following snakebites in Sri Lanka: Results of an island wide community based survey | Sri Lanka has a population of 21 million and about 80,000 snakebites occur annually. However, there are limited data on health seeking behavior following bites. We investigated the effects of snakebite and envenoming on health seeking behavior in Sri Lanka.
In a community-based island-wide survey conducted in Sri Lanka 44,136 households were sampled using a multistage cluster sampling method. An individual who reported experiencing a snakebite within the preceding 12 months was considered a case. An interviewer-administered questionnaire was used to obtain details of the bite and health seeking behavior among cases.
Among 165,665 individuals surveyed, there were 695 snakebite victims. 682 (98.1%) had sought health care after the bite; 381 (54.8%) sought allopathic treatment and 301 (43.3%) sought traditional treatment. 323 (46.5%) had evidence of probable envenoming, among them 227 (70.3%) sought allopathic treatment, 94 (29.1%) sought traditional treatment and 2 did not seek treatment. There was wide geographic variation in the proportion of seeking allopathic treatment from <20% in the Western province to > 90% in the Northern province. Multiple logistic regression analysis showed that seeking allopathic treatment was independently associated with being systemically envenomed (Odds Ratio = 1.99, 95% CI: 1.36–2.90, P < 0.001), distance to the healthcare facility (OR = 1.13 per kilometer, 95% CI: 1.09 to 1.17, P < 0.001), time duration from the bite (OR = 0.49 per day, 95% CI: 0.29–0.74, P = 0.002), and the local incidence of envenoming (OR = 1.31 for each 50 per 100,000, 95% CI: 1.19–1.46, P < 0.001) and snakebite (OR = 0.90 for each 50 per 100,000, 95% CI: 0.85–0.94, P < 0.001) in the relevant geographic area.
In Sri Lanka, both allopathic and traditional treatments are sought following snakebite. The presence of probable envenoming was a major contribution to seeking allopathic treatment.
| Sri Lanka has a high incidence of snakebite with 80,000 bites occurring annually in a population of 21 million. However, there are very limited data on health seeking behaviour following bites. We undertook a community based island-wide survey in Sri Lanka during 2012–2013 and surveyed 165,665 individuals living in 44,136 households. There were 695 snakebite victims and 98% of them had sought health care after the bite; 55% sought allopathic (western medical) treatment and 43% sought traditional treatment. The decision on seeking allopathic treatment was influenced mainly by probable envenoming following the snakebite.
| Sri Lanka has a very rich snake fauna. There are 96 snake species in the country but only six of them are considered venomous (i.e. Naja naja, Bungarus ceylonicus, Bungarus caeruleus, Daboia russelii, Echis carinatus and Hypnale hypnale). The country has a high incidence of bites and envenoming, but a relatively low mortality associated with snakebite. A national survey conducted in 2012 estimated that there are an estimated 80,000 snakebites, 30,000 envenomings and 400 deaths annually (in a total population of about 20 million according to the National Census Data for Sri Lanka—2012) [1]. Both snakes and snakebites have a wide geographical variation [1]. The Russell’s viper is considered as the largest and most dangerous snake in Sri Lanka and is the most widely distributed venomous snake. Hump nosed vipers are commonly found in human habitats including rubber, tea, coconut and cocoa plantations and 35% to 45% of human bites result from this species. Saw scaled vipers are largely confined to the arid dry zones of the country including Northern and Eastern Provinces and are responsible for only 1% to 2% of the total bites. Cobras are the largest elapids found in Sri Lanka and are also widely distributed in the country. Kraits are mainly found in and around human habitats, and the majority of krait bites have been reported from within households in the dry zone of the country. Cobra and krait bites mainly cause neurotoxic effects while viper bites are principally associated with haematological abnormalities and soft tissue damage. Sri Lankan Russell’s viper bites could also cause neurotoxicity [2–4].
State healthcare is free at the point of delivery in Sri Lanka and includes both allopathic and indigenous medical systems. The state healthcare sector has a total of 17,000 allopathic practitioners (MBBS qualified doctors) and over 20,000 registered indigenous practitioners (84.6% Ayurveda, 12.7% Siddha system and 2.7% other). It is estimated that there are another nearly 8,000 who are descendants of traditional medical practitioners who also provide healthcare [5,6].
In many rural communities in low and middle income countries, including Sri Lanka, victims traditionally favour indigenous treatments for snakebites. In rural Bangladesh, Nigeria and Kenya, only 3%, 8.5% and 27% of snakebite victims respectively, sought hospital treatment [7–9]. This pattern of health seeking behaviour may be due to a combination of socio-cultural factors and poor access to often sub-optimal health care in these countries. Two qualitative studies investigating beliefs regarding snakebite and their influence on health seeking behaviour in four rural communities in Sri Lanka found that people firmly believed that Ayurvedic treatment was effective for snakebite [9,10]. Traditional healers were respected and although many victims finally sought allopathic treatment, they often initially consulted a traditional practitioner [10]. There were also common misconceptions about harms of hospital treatment, based mainly on misinformation regarding antivenom therapy [11]. However, in some middle income countries, such as Costa Rica, most of the population affected by snakebites seeks treatment in health posts where free allopathic treatment is provided [12].
Outcomes of snakebites could be associated with the health seeking behavior of victims following the bite, and the decision on health seeking behavior can be influenced by the victims’ social and natural environment [1,13]. Information on health seeking behavior from hospital based studies is bound to be biased and cannot be generalized to a community [14,15]. We report health seeking behavior following snakebite and factors influencing the decisions regarding health seeking among the individuals bitten by snakes as part of a community-based island-wide survey on snakebite in Sri Lanka.
An island wide community based “National Snakebite Survey” was conducted from 2012 to 2013 to collect data on snakebite [1]. The study was designed to sample about 1% of the Sri Lankan population. A Grama Niladhari division, which is the smallest administrative division in the country, was considered a cluster in the survey methodology. Sri Lanka has nine provinces and these nine provinces are further subdivided to 25 districts (i.e. the number of districts varies from 2 to 5 within provinces). The sample was equally distributed among the nine provinces where 125 Grama Niladhari divisions (clusters) were sampled from each province. These 125 clusters were proportionally allocated to the districts within the province based on the population living in the districts. Of the total of 14,022 GN divisions in Sri Lanka, 1125 GN divisions (125 from each of the 9 provinces) were allocated for the survey. In each cluster, 40 consecutive households were sampled based on a random starting point and all the permanent members of the households were included in the study. An interviewer administered questionnaire was used to collect data where direct interviews were done from an adult household member. Demographic data on household members and snakebite events which occurred within the previous year were recorded. Detailed data on snakebite and health seeking behavior following the snakebite were collected, this included place of bite, time of the bite, identification of snake, primary treatment modality, other treatment modalities if secondary healthcare sought, nature of treatment taken–whether inpatient or outpatient, state sector or private sector care, distance to health care facility, time taken to seek treatment, clinical features and outcome of bite. The presence of probable envenoming in the victims was determined based on the reports of presence of local tissue necrosis at the site of bite, presence of neurotoxicity or bleeding manifestations.
The proportion of victims who sought allopathic and traditional treatment within each province was evaluated. Health seeking behavior was compared between provinces using Western Province, which had the lowest proportion of people seeking allopathic treatment, as the comparator (i.e. the reference level). Linear logistic models were used to model the individual health seeking behavior (i.e. allopathic vs traditional) in relation to the corresponding province, considering the province as the independent variable. Odds ratios with 95% Confidence Intervals were generated to illustrate the differences in health seeking behavior each province.
Estimated snakebite and envenoming bite incidences at the Grama Niladhari (GN) divisions (cluster level for this study) were used for the analysis. Each individual snakebite recorded in the national survey and the health seeking behavior activity following that bite was presented graphically with respect to the relevant GN snakebite and envenoming incidences using scatter plots.
There were 695 snakebites in the survey sample and 13 had not sought treatment. Therefore, 682 victims’ data was used for the model. Health seeking behaviour (i.e. allopathic treatment seekers vs indigenous treatment seekers) was considered as the response variable, where those who sought allopathic treatment were coded as 1 and others were coded as 0. Separate binomial logistic models were used to evaluate the differences in health seeking behavior in different provinces, and to identify the variables associated with individual decisions on selecting the treatment type. In the first binomial logistic model, province was considered as an independent variable, where allopathic treatment seeking behavior was compared between provinces. Western Province, which had the lowest proportion of people seeking allopathic treatment was considered as the comparator (i.e. the reference level). Odds ratios with 95% Confidence Intervals were generated to illustrate the differences in health seeking behavior in each province with reference to Western Province. A second binomial logistic model was used to investigate the association between health seeking behavior (i.e. allopathic vs traditional) and socio-demographic variables of victim, envenoming status of victim, time of bite, month of bite, place of bite, identification of snake, distance from the bite location to the healthcare facility where the initial treatment was obtained, outcome of the bite and incidence of snakebite or envenoming at the geographical cluster level. Adjusted Odds ratios with 95% Confidence Intervals for the significant variables were obtained and a p value of 0.05 was considered as significant. Data analysis was done with R programming language version 3.2.3 and geographical mapping was done with ArcMap 10.3.1.
Ethical approval for the study was obtained from the Ethics Review Committee of the Faculty of Medicine, University of Kelaniya. All interviews were conducted after obtaining informed written consent. Approval from District and Divisional level public administrators were obtained for conducting the community-based survey. No animals were used in the study.
The “National Snakebite Survey” included 165,665 individuals in Sri Lanka. 695 individuals had experienced a snakebite event during the preceding year. Most snakebites (85.8%) had occurred outdoors and 75.4% patients had recognized the offending snake. 323 (46.5%) patients had clinical features of probable envenoming; 92.7% of patients sought treatment within the first day (24 hours) and another 5.0% during the second day.
Patterns of health seeking behavior following the bite are detailed in Table 1. 13 of 695 (1.9%) victims did not seek health care. 381 (54.8%) snakebite victims sought allopathic treatment initially and 301 (43.3%) snakebites sought traditional treatment initially. Secondary treatment was sought by 115 victims (16.7%) after primary treatment—some changed modality but others sought further advice within the same health care system. 24 (64.9%) out of 37 who previously sought indigenous treatment and 39 (50%) out of 78 who previously sought indigenous changed from their initial treatment to allopathic and indigenous healthcare respectively. However, in approximately 30% of these who accessed treatment from a different source within the first 24 hours, 24 (70.6%) of these were envenomed and approximately three quarters of these sought allopathic treatment.
The time of the bite did not influence whether allopathic or traditional treatment was sought. 138 (19.9%) victims received religious blessings following the snakebite; this proportion did not differ between those seeking allopathic or traditional treatment (19.2% vs 21.7% respectively).
Table 2 demonstrates the wide geographical differences in the pattern of health seeking behavior. More than 85% sought allopathic treatment in Northern and North central provinces in which high proportions of victims are envenomed (71.2% and 70.6% respectively) and which have the highest envenoming incidence among the provinces. Allopathic treatment seeking was lowest in the Western province which has the lowest proportion of bites that lead to envenoming and lowest incidence of envenoming in the country [1].
Allopathic health care seeking was least common in Western province and this province was therefore used as the comparator. There was no significant difference between patterns of health care seeking between Western province and Central and Sabaragamuwa provinces. All other provinces showed significantly higher odds ratios for seeking allopathic treatment which varied from 4.24 in Southern Province to 62.5 in Northern Province (Table 3).
The median snake bite incidence was 504 (minimum: 83, inter quartile range: 378–642, maximum: 1511) per 100,000 and the envenoming incidence was 215 (minimum: 13, IQR: 143–283, maximum: 535) per 100,000. Fig 1 shows graphically each bite related to the local incidence of snakebite and envenoming and whether allopathic or traditional health care was sought. Generally, there was greater allopathic healthcare seeking where the incidence of envenoming was high and traditional healthcare seeking was more common where the incidence of envenoming was low (Fig 2).
Multiple logistic regression model fitting showed that health seeking behavior was independently associated with the probable envenoming status of the patient, number of days taken to seek treatment after the bite, distance to the healthcare facility and the envenoming and snakebite incidence in the geographic area. Allopathic treatment seeking was higher in the geographical areas with higher envenoming incidence and among victims with probable envenoming; victims travelled further to the healthcare facility to seek allopathic treatment. On the other hand, allopathic treatment seeking was less common in geographical areas with higher snakebite incidence and in victims who had delayed presentation to healthcare (Table 4).
Most victims (98.1%) sought some form of treatment following snakebite. Overall, 55% of victims expressed an initial preference for allopathic treatment but this proportion varied considerably across the country. Allopathic treatment seeking behavior was mainly determined by the presence of probable envenoming; 70% of victims with probable envenoming sought allopathic treatment. The incidence of envenoming in a given geographical area was also a significant independent determinant of health seeking behaviour, with an association between increasing incidence of envenoming and allopathic treatment. In contrast, there was an inverse association between the overall incidence of bites in an area and seeking allopathic treatment. This may be related to the fact that geographical areas with the highest incidence of snakebite incidences were not the areas with the highest incidences of envenoming, and many victims with non-envenoming bites tended to seek traditional treatment; familiarity with snakebites may also reduce allopathic health seeking. We did not find any association between health seeking behavior and the victims’ age, gender, ethnicity, religion, employment, income level, education level, month of bite, time of bite, place of bite and type of snake.
More than 90% of snakebite victims sought treatment within the first day. Those who delayed treatment seeking were more likely to seek traditional treatment, probably because these victims experienced lesser degrees of envenoming or no envenoming. Distance from the site of bite to the healthcare facility was also a significant determinant of health seeking behavior; victims who sought allopathic treatment travelled longer distances as previously shown where distance to healthcare facility has impact on individual behaviour [16]. This is likely to simply reflect the easier availability of traditional treatment close to where patients are bitten, but may also be related to those who considered they needed allopathic treatment for more severe manifestations were prepared to travel further.
There are likely many differences in the manner in which people respond to a snakebite. A community based study in Sindh, Pakistan showed that less than 75% of snakebite victims sought any form of healthcare and managed the bites by themselves [17]. In contrast, we found that almost all victims (i.e. 98%) sought some form of treatment. This may be related to high rates of literacy and access to healthcare in Sri Lanka [17]. Another community based study from Kenya showed that 68% of snakebite victims sought treatment from traditional healers, and nearly all victims (98%) initially used traditional treatments [8]. In South Africa, 80% initially used traditional treatments with 63% subsequently accessing a traditional healer [18]. These figures are much higher than in our study [8,18]. The same Kenyan study reported that 25% of bite victims received some form of first aid, including traditional treatment, before reaching a health facility. We found that almost 65% of victims who initially sought traditional treatment subsequently went on to seek allopathic treatment, in line with findings from a previous Sri Lankan hospital based study [15]. Reports from Costa Rica show that most victims seek allopathic treatment soon after a snakebite [19].
Seeking traditional treatment for snakebites has been previously associated with poor access to hospitals [18]. We however found that traditional treatment was often preferred even in the relatively developed areas of the country with good access to hospital care, for example, the Western Province and Central Province are well developed regions in the country and have good access to hospital care with a high concentration of hospitals including teaching hospitals. The Sabaragamuwa Province has both developed and underdeveloped areas and has a Provincial General Hospital and several District Hospitals [18], but the majority of snakebite victims in areas with a high incidence of envenoming sought allopathic treatment, even though these were underserved areas with relatively poor access to hospital care. There were also interesting findings about the time to seek care; in general, allopathic treatment was sought more rapidly than traditional treatment, although people had to travel further to access it. This may reflect the recognition of the urgency when severe snakebite occurs.
Overall, our study reflects the complex decision-making process about health care seeking following snakebite with decisions appearing to be influenced by the community experience of snakebite, as well as the individual circumstances of a bite. The influence of the strength of cultural beliefs is further emphasized by our finding that about one fifth of victims engaged in religious rituals following snakebites and this proportion was similar in those seeking allopathic or traditional treatments.
There are several limitations in our study. We obtained data on snakebite for the preceding one year and this could be associated with recall bias, although an event such as a snakebite is likely to be memorable. The presence or absence of envenoming following snakebite was based on the clinical features reported by the victims, and there is a possibility of misclassification. Distance to the healthcare facility was estimated by patients and so may not be completely accurate. Our analysis was mainly focused on comparing allopathic versus traditional treatment seeking behavior on a national level, and we therefore considered all the non-allopathic treatments as traditional. Individual and community qualitative assessment would have been helpful to generate a more detailed in depth understanding of the personal and household health seeking behaviour following snakebite. Future studies could also be done to assess the effect of macro-ecological variables on health seeking behaviour of snakebite victims in addition to the individual level socio-demographic variables [20].
Nevertheless, we have been able to generate an in-depth analysis of health seeking following snake bite in Sri Lanka, demonstrating that allopathic and traditional treatments are more or less equally sought following snakebite and that although the presence or absence of envenoming are important, many other factors also influence health seeking decisions.
|
10.1371/journal.pntd.0007326 | Prevalence, intensity and risk factors of tungiasis in Kilifi County, Kenya II: Results from a school-based observational study | Awareness of the public health importance of tungiasis has been growing in East Africa in recent years, but data on epidemiological characteristics necessary for the planning and implementation of control measures do not exist. The work presented here was part of a larger cross-sectional study on the epidemiology of tungiasis in coastal Kenya and aims at identifying risk factors of tungiasis and severe disease in school children.
A total of 1,829 students of all age groups from five schools and 56 classes were clinically examined for tungiasis on their feet based on standardized procedures and observations made about the school infrastructure. To investigate the impact of school holidays, observations were repeated after school holidays in a subset of children in one school. In an embedded case-control study, structured interviews were conducted with 707 students in the five schools to investigate associations between tungiasis and household infrastructure, behaviour and socio-economic status.
The overall prevalence of tungiasis was 48%; children below the age of 15 years were the most affected, and boys were twice as likely as girls to be infected. The highest risk of disease was associated with the socio-economic circumstances of the individual student at home. The study indicated that mild to moderate tungiasis could be reduced by a third, and severe tungiasis by over half, if sleeping places of children had hardened floors, whilst approximately a seventh of the cases could be prevented by sealing classroom floors in schools, and another fifth by using soap for daily feet washing.
There is a clear role for public health workers to expand the WASH policy to include washing of feet with soap in school-aged children to fight tungiasis and to raise awareness of the importance of sealed floors.
| Tungiasis is a neglected tropical skin disease caused by penetrated sand fleas, the adult female of which burrows into the skin of the feet. The parasite rapidly expands its body size by a factor of 2000. The growth causes immense itching, inflammation, pain and debilitation. The current lack of good treatment methods means people attempt to remove the fleas themselves with non-sterile instruments causing more damage. Control efforts focus on prevention but there is little data to guide this in East Africa. The current study reinforces our previously published results on the household level from the same communities, indicating that prevention needs to focus on hardening the floors of resource-poor families and integrating daily foot washing with soap into water, hygiene and sanitation programs.
| Sand flea disease (tungiasis) is a highly neglected parasitic skin disease which inflicts pain and suffering on millions of impoverished people in South America, the Caribbean and sub-Saharan Africa. Research on tungiasis is scant and most of the publications originate from South America [1] [2],very little work has been done in sub-Saharan Africa. This is partly because even though the disease bears all the hallmarks of a Neglected Tropical Disease [1] it was until very recently [3] not included in the World Health Organisation’s list of Neglected Tropical Diseases. This made it difficult for funding organisations to invest research funds into determining the disease ecology and consequently the development of urgently needed treatment and prevention tools. Most research currently undertaken on tungiasis is small-scale, supported by national and international non-governmental organizations and well-wishers, reinforced by community-based self-help groups and is hence highly resource-constrained [2].
The development of the juvenile stages of the sand flea Tunga penetrans, is similar to that of other Siphonaptera; they live off the host but depend during their development on loose, sandy soil [4, 5]. In contrast to other flea species, the adult female sand flea becomes permanently parasitic on its host, where it burrows into the skin and undergoes a dramatic growth, increasing its volume about 2000-fold within eight days [6]. Tungiasis affects mostly the feet and is associated with a pattern of debilitating morbidity [6]. Itching, pain, swelling, deep fissures, ulcers and abscess formation are symptoms of an acute inflammatory response to embedded fleas and bacterial superinfections of the lesions. Chronic infections result in chronic pain, disability, disfigurement and mutilation of the feet [6, 7, 8]. Children with tungiasis are often ridiculed by their peers and it has been shown that physical incapacity, mental strain and distress reduce quality of life [9].
Awareness of the public health importance of tungiasis has been growing in East Africa in recent years [10], but data on epidemiological characteristics, necessary for the planning and implementation of a control program, do not exist. The work presented here was part of a larger observational study performed in disease endemic areas in coastal Kenya and included simultaneously implemented cross-sectional household-based risk factor surveys recently published [11] and school-based surveys presented here. To the best of our knowledge, combined, the two studies provide the most comprehensive risk factor study to date on tungiasis for Sub-Saharan Africa.
The aim of the school-based study was to complement the household-based study to investigate prevalence and risk factors for tungiasis and whether targeting school-going children for treatment and prevention might be a viable option. The specific objectives included: 1) determine the prevalence of tungiasis in selected rural schools in Malindi sub-county, 2) whether there are specific school factors associated with disease, 3) whether school or home present the greater risk and 4) if similar or different risk factors associated with the disease might be identified through a school-based survey.
The study was performed in Malindi Sub-county, Kilifi County, eastern Kenya, where tungiasis is endemic and previous reports indicated disease prevalence in villages to range from 8% to 65% [11]. Malindi Sub-county is divided into two ecological zones: Kakuyuni Sub-location, a densely populated area in the coastal strip with tropical climate, and Malanga Sub-location, a less densely populated area inland with a much drier climate. Most rural homesteads in these areas consist of several mud-walled houses with a palm thatch roof and sandy floor. Domestic animals such as goats, cats, dogs and chickens walk freely on the compounds. Subsistence farming is the only activity practiced by most of the population to sustain their livelihood.
The cross-sectional study was implemented between August 5 and October 3, 2014 in five primary schools; two schools in Kakuyuni Sub-location (labelled as KS1 and KS2), and three schools in Malanga Sub-location (labelled as MS1, MS2, MS3). Schools were a minimum of 2 km apart with distinct catchment areas from where children originated. All except MS3 were public schools. The schools were purposely selected based on recommendation from the county Public Health Department as schools that were affected by tungiasis and no interventions had previously taken place in the catchment areas of these schools. The schools were located in the same communities as the household study. For the risk factor study a case: control design was used with a 1:1 ratio. The sample size calculation to yield results with 95% confidence limits, 80% power, an assumed prevalence of exposure of 40% among controls and least extreme odds ratio to be detected of 1.5, indicated the need for a sample of 776 students (388 cases, 388 controls). To allow for incomplete data sets and be able to adjust for confounders we increased this by 20%, aiming to interview 930 students (465 cases, 465 controls).
All schools were single-storey buildings and all were divided into several classrooms. MS2 had some classrooms of concrete and some of mud while all in MS3 were mud walls. The floor quality of the classrooms ranged from smooth cement surfaces to loose sand/soil (Fig 1). All schools had access to water and according to the information from the class teachers, classroom floors were swept daily by the students. For all classes, the classroom size, the number of students per class, and the type of classroom floor, walls and roof were recorded.
All female and male students of all age groups, for whom informed consent and assent was given, were clinically examined for tungiasis, class by class. Prior to the clinical examination, the feet of the students were carefully washed with soap in a basin. Each individual was then examined for tungiasis based on a standardized procedure [11], focussing on their feet and hands since a high number of lesions at the feet frequently coincide with the presence of ectopic lesions at the hands [12]. Patients were also asked whether they had tungiasis lesions in other regions of the body. Lesions were counted and staged according to the Fortaleza classification [6] as stage I: penetrating sand flea; stage II: brownish/black dot with a diameter of 1–2 mm surrounded or not by an erythema; stage III: circular yellow-white watch glass-like patch with a diameter of 3–10 mm and with a central black dot; stage IV: brownish-black crust with or without surrounding necrosis. Stage I to III are viable sand fleas. In stage IV the parasite is dying or already dead (non-viable). Lesions manipulated with a sharp instrument (by the patient or their caregiver) with the intention to remove the embedded parasite were documented as manipulated lesions. Based on the number of lesions present, the intensity of tungiasis was classified as light (1–5 lesions), moderate (6–30 lesions) or severe (>30 lesions) [13]. For every patient, the age, sex, and class were recorded.
To investigate potential changes in the presentation of tungiasis in students after school holidays we aimed to repeat the clinical examinations in as many students as possible in KS1 who had been originally examined in the week immediately before the 4-week August holiday. Only 248 of the original group of students examined could be traced in the first week after the holiday.
Within each school a subset of students was selected for interviews. In KS1 and MS1, all tungiasis cases over the age of 4 years and the following uninfected student (as an age-matched control) were interviewed. In the remaining schools, only the first half of the cases (in chronological order of their identification) and the following uninfected student, were interviewed due to time constraints. It was not possible to recruit equal numbers of controls for interviews in MS2 and MS3 because the majority of students had tungiasis. Structured interviews were conducted using a pre-tested questionnaire in Giriama or Swahili language, including questions about the physical structure of the house in which children slept (house walls, roof and floor), water sources and access at home, hygiene habits (washing frequency, soap use, toilet facility at home), livestock and companion animals kept in homestead and walking time to school. Furthermore, observations were recorded about the condition of the students’ school uniforms and the type of shoes worn if any.
Generalized Estimating Equations were used to analyse potential associations between the prevalence of tungiasis or the number of lesions (of different stages) and multiple variables recorded during interviews and observations. Prevalence data were modelled using binomial probability distributions with logit link functions fitted, count data were modelled using negative binomial probability distributions with log link functions fitted. Depending on the analysis the unique school ID, the unique class ID or the unique student ID were included as repeated measure and an exchangeable correlation matrix assumed. In the final multivariable risk factor analyses only factors found significant when tested individually in a univariate analysis were included as predictors. Interactions were explored for variable combinations that were plausible to be potentially interacting. In the final model, only significant interactions were included. All reported mean proportions or mean counts and their 95% confidence intervals (CIs) were estimated as the exponentials of the parameter estimates for models with no intercept included. Frequency counts were compared using the Pearson Chi-Square test. The analyses were done with R statistical software version 2.14.2 [14]. Population Attributable Fractions (PAF) were calculated, representing the fraction of cases which would not have occurred if an exposure had been avoided, assuming the exposure is casual and the other risk factors in the population remain unchanged [15]. PAFs are the percent exposed among cases multiplied by the attributable risk (AR). The AR is the risk of tungiasis in the exposed due to the exposure and is calculated as (odds ratio (OR) −1)/OR.
The study was approved by the Ethics Review Committee at Pwani University, Kilifi County, Kenya; approval number ERC/PhD/010/2014. During the study preparation phase contact was made with the County and District leadership in the Ministry of Health and the Ministry of Education, the Zonal Education Officer and the Community Health Officers to obtain their approvals and support for the study. Meetings were held with Community Health Workers (CHWs) in each sub-location, training on tungiasis provided and the aims and procedures of the study explained, emphasizing that participation was completely voluntary, and subjects had the opportunity to withdraw from the study at any point in the study. Head Teachers were visited and provided with an information sheet. The information was read out at a parents’ meeting prior to the study, explaining the procedure and voluntary nature of participation, and asking for consent. The Head Teacher signed the consent form (S1 Annex) on behalf of the parents and school. Data were collected with the help of Community Health Volunteers of the respective Community Health Unit. All data analyses were conducted anonymously.
All students with tungiasis were treated after the survey by the Community Health Workers according to standard practice in Kilifi County [10]. For those with secondary bacterial infection and other illnesses requiring treatment, a referral form was prepared by a Community Health Worker, and patients were referred to the nearest Health Facility.
A total of 1,829 students from 56 classes in the 5 schools were examined, about 70% of the students enrolled in the schools (total 2,622) based on school records. Of these, 923 students were interviewed, just short of the calculated sample size of 930, but only 707 had fully completed the interview and were included in the data analysis; 398 were cases and 309 were controls (Fig 2).
Of the examined 1,829 primary school students 48% were boys and 52% were girls; 31% of the students were 2–9 years old, 52% 10–14 years old and 17% were 15–21 years old. Of the 870 (48% of 1,829) students with tungiasis, 58% had mild infections (1–5 lesions), 31% moderate (6–30 lesions) and 11% severe infections (>30 lesions;). The majority of all cases were males aged 10–14 years (28.4% of cases, Fig 3) and also had the highest percent of moderate and severe cases. The student population size examined ranged from 140 to 582 between schools and the prevalence of tungiasis varied significantly between the five schools ranging from 31% to 83% (Table 1).
The physical school environment was considered a potential risk factor for tungiasis either directly providing a conducive environment for the off-host host stages to develop and adult sand fleas to find a host or indirectly as a proxy measure for the socio-economic circumstances affecting the community from which the children were drawn to the respective schools (children from poor background might only be able to afford sending children to a school with very basic school environment). Size, materials of the floors, walls and roofs were recorded for every classroom from which children were examined during the surveys. The majority (70%) of the students were taught in classrooms between 40–70 m2 while 22% were taught in classrooms larger than 70 m2 and only 8% in classrooms smaller than 40 m2. Classroom size was not significantly associated with tungiasis prevalence in a univariate analysis, and neither was the number of children per m2 in a classroom, which ranged from 0.4 to 2.0.
Most classroom floors were made of concrete; 46% of all screened students were taught in a room with a good quality concrete floor and 40% in a classroom with cracked concrete floor. Concrete floors were always associated with concrete walls and iron sheet roofs. Only 14% of the surveyed students studied in classrooms with a natural sand/soil floor. These classrooms also had mud walls and thatched roofs (Fig 1).
The differences in tungiasis prevalence between schools were confounded by the physical classroom environment as revealed by the multivariable analysis (Table 1). The floor type of the classroom was an important risk factor for finding a tungiasis case, the degree of risk per floor type, however, was dependent on the school, as shown by the significant interaction between floor type and school. Natural sand or soil floor of a classroom was an independent risk factor, increasing the probability of finding a tungiasis case 3-fold as compared to finding a case among students that were taught in a classroom with a well-kept, smooth concrete floor. Classrooms with natural sand/soil floors were only present in MS2 and MS3, the two schools with the highest tungiasis prevalence. Whether a cracked concrete floor represented a risk for finding a tungiasis case depended significantly on the school. The impact of the interactions can be calculated by multiplication of the odds ratios [16]. This means a cracked concrete floor was only a factor significantly associated with increased disease risk in MS2 (OR 11.54 x OR 0.51 = OR 5.89), where classrooms with sand/soil and concrete floors coexisted but not in MS1 (OR 2.16 x OR 0.51 = OR 1.10; Table 1). Consequently, floor type was not an explanatory variable for tungiasis prevalence in KS1, KS2 and MS1, where tungiasis prevalence ranged between 31% and 51% (Table 1).
Both age and sex were significantly, and independently associated with tungiasis. Children below the age of 15 years were 1.4–1.6 times more likely to be diagnosed with tungiasis than older students and boys more than 2 times more likely than girls (Table 1).
No specific school-based risk factors were significantly associated with severe tungiasis, when the multivariable analysis was repeated with severe manifestation as the dependent variable.
Anecdotal information provided by the teachers suggested that children usually return to school with a higher tungiasis burden after school holidays. To investigate this, we compared the tungiasis prevalence and infection status immediately before and after the one-month August school holiday in the 248 students who were able to be traced in the first week after the holiday in KS1 (Tables 2 and 3). This sub-group comprised 41% boys, 59% girls, with 23% <10 years old, 55% 10–14 years old and 22% 15–20 years old.
Sex and age were identified, similar to the analysis on the larger data set, as independent risk factors for tungiasis before and after the school holidays, with boys and younger age groups more likely to be found with the disease than girls and older age groups (Table 2). There was no significant interaction between sex, age and survey time. The probability of finding a tungiasis case after the school holidays was 1.7 times higher than before the holidays, with a mean prevalence in students of 31% (95% CI 25–38%) before and 44% (95% CI 36–51) after the holidays (Table 2).
Taking a closer look at the number and developmental stage of the embedded sand fleas (Table 3), we observed that the number of viable lesions in students with tungiasis had significantly decreased (RR 0.30 (95% CI 0.10–0.84), p = 0.023) after the holidays from a mean number of 2.52 (95% CI 1.61–3.93) viable lesions before to 0.80 (95% CI 0.50–1.29) after the holidays, irrespective of sex and age. The number of non-viable lesions was generally highest in students with tungiasis between 10–14 year of age irrespective of sex and survey round, however significantly decreased (RR 0.17 (95% CI 0.04–0.66), p = 0.010) after the holidays from a mean of 8.87 to a mean of 3.21 (Table 3). A proportionally similar decrease in non-viable lesions was also seen in the younger age group of 2–9 years old students but not in the older students as shown by the significant interactions between age and survey round (Table 3). On the contrary, the number of manipulated lesions increased significantly after the school holidays. Significant interactions in the analysis (Table 3) highlighted a proportionally higher increase in manipulated lesions in girls from 3.06 (95% CI 1.84–4.94) before to 9.52 (95% CI 7.24–12.53) after holidays, than in boys, even though boys had overall a larger number of manipulated lesions (mean before holidays 7.98 (95% CI 5.35–11.91); after holidays 10.55 (95% CI 7.86–14.16)). There was also a proportionally higher increase in the number of manipulated lesions in the 10-14-year-old students than in the other age groups (Table 3).
Most of the characteristics assessed by observation and interview, for each of the 707 students interviewed, showed considerable heterogeneity between schools, except for sex and ownership of dogs and chickens (Table 4).
The significant between-group variations for schools was taken into consideration by including the school as a random factor in the subsequent multivariable analyses of the interview data (Table 5). Since tungiasis cases and healthy controls were matched by age at the enrolment stage, age was not a factor significantly associated with tungiasis as an outcome in the analysis of the interview data and hence not included in the multivariable analysis. Expectedly, sex was similarly associated with tungiasis risk in the interview data, with boys being >2 times more likely to be found with tungiasis than girls. The condition of the school uniform was not independently associated with tungiasis, but an interaction existed with sex. The chance of finding tungiasis was significantly higher (OR 4.30 (95% CI 1.47–12.60), p = 0.008) in boys with torn school uniforms than in boys with better uniforms or in girls with torn uniforms (Table 5).
Similarly, the absence or presence of open or closed shoes was not by itself a risk factor for tungiasis. When a child was however found not wearing shoes and had a badly torn school uniform it was highly likely (OR 5.32 (95% CI 3.22–8.79), p<0.001) to find tungiasis (Table 5). The time a child took to walk to school was not significantly associated with presence of tungiasis in the multivariable analysis. Whilst the building material of the students’ homes floors, walls, and roofs were all associated with tungiasis in a univariate analysis only the home’s floor was an independent risk factor in the multivariable analysis. A student from a home with a natural sand or mud floor indoors was nearly twice as likely to be diagnosed with tungiasis, than a student from a home with a concrete floor in the house (Table 5). Most of the students had access to piped water or a well either on their compound or shared in the village. Nevertheless, students coming from a home where water was fetched from a community tap or well were 1.6 times more likely to have tungiasis than those students that had tap water on their compound at home. The time it takes for the family to fetch water was not associated with the disease outcome. Whilst the frequency of washing feet was not associated with the presence of tungiasis, the use of soap strongly was. Students that responded never to wash their feet with soap were over 6 times more likely (95% CI = 3.2–12.6, p<0.001) to have tungiasis and students that responded to only sometimes wash they feet with soap were 1.6 times more likely (95% CI 1.50–1.76, p<0.001) to have tungiasis than those students always washing with soap (Table 5). Although washing frequency responses were not significantly associated with tungiasis, not answering this question was (OR 6.01 (95% CI = 3.64–9.92), p<0.001). Neither the type of toilet at home, nor the presence of a dog on the compound, were significantly associated with disease outcome when the analysis was adjusted for all other variables.
In an attempt to better understand what drives severe infection (>30 embedded lesions) we performed a multivariable analysis for severe disease as the outcome (n = 71, 18% of all cases N = 398), comparing the characteristics of these severe cases with mild to moderate cases (1–30 lesions, n = 327). Sex, condition of clothing, shoe-wearing and frequency of soap use were not significantly associated with disease severity. Of all tungiasis cases, severe infections were more likely to be found in the younger age groups (Table 6), in children having a natural sand/mud floor indoors at home than those that have a stone/cement floor at home, using a water source other than a private or community tap or well, and washing their feet less than once a day. Of all children with tungiasis, those that reported their family did not own at least one chicken had a significantly higher risk of heavy infection than children who reported they kept chicken (Table 6). Amongst all children with tungiasis, school floor characteristics were not a predictor for heavy infection.
For those factors which were significant risks for tungiasis or a high intensity of infection, and are amenable to being changed, the PAF were calculated. The PAF is the percent reduction in prevalence that would occur if exposure to the risk factor were removed. The highest PAF for both, any infection and severe infection was found to be having a home floor of sand or smeared mud (30.7% and 54.4% respectively, Table 7). Only using soap sometimes when washing had a PAF of 21.5%, and a classroom floor of sand had a PAF of 14.3% for any type of infection.
Our study confirmed that the prevalence of tungiasis is extremely heterogeneous, varying from school to school and community to community even though they are only a few kilometres apart [17]. The disease burden was highly aggregated even within an individual school, with more than half of the cases having only a few embedded sand fleas but a minority being severely affected. The overall prevalence of 48% of all screened school-aged children was twice as high as the prevalence in the simultaneously implemented household study [11], reflecting the high proportion of the most affected age group in the school-based study. As has been shown before in Brazil [17], Uganda [18], Nigeria [19], and Kenya [20, 21], school-aged children below the age of 15 years were the most affected by tungiasis and boys were twice as likely as girls to have the disease.
The risk factor interviews as well as the follow up examinations after the long school holidays suggest strongly that the highest risk of disease is associated with the socio-economic circumstances of the individual student at home. Whilst an unsealed, natural sand or soil floor of a classroom came out as an independent risk factor in the analysis it is important to note that such classroom floors were only present in two schools where the majority of pupils came from homes that had unsealed floors. The calculation of the PAF indicates that mild to moderate tungiasis could be reduced by a third, and severe tungiasis by over a half, if homes (sleeping places of children) had sealed floors, whilst approximately a seventh of the cases could be prevented by sealing classroom floors in schools. The presence of unsealed floors at home, has been previously indicated as an important risk factor for the disease [13, 21, 22], and can only be a consequence of the biology of the sand flea, with egg, larval and pupal (off-host stages) development taking approximately three weeks and requiring shaded, dry, loose soil or sand [4, 5]. Such unsealed floors provide a constant supply of the sand fleas searching for hosts as soon as they emerge as adults. This finding also corroborates the assumption that in settings where the prevalence of tungiasis is stable the whole year round, the transmission mainly takes place inside the house, particularly in the room where children sleep [1].
Neither the condition of the school uniform, nor wearing shoes potentially protecting against host-seeking sand fleas [23] was independently associated with tungiasis, however, the combination of a torn uniform and the absence of shoes can be considered an indicator of the poverty level or care given to the child at home. Complex interactions between risk factors suggest underlying behavioural differences in the care given by parents and guardians and/or hygiene behaviour expressed in boys and girls. Boys with torn uniforms were four times more likely to be affected by the disease than girls wearing uniforms in equally poor condition.
Whilst the frequency of washing was not associated with tungiasis, the availability of piped or well water within the homestead and the use of soap when washing was strongly associated with reduced risk. Both factors might be linked to the socio-economic status of the family to afford piped water and soap, but also to behavioural characteristics of the care givers and children, as already noted in the household study [18].
The higher risk of infection and severe disease observed for boys between the ages of 10 and 14 years may be a reflection of their hygiene practices, being less likely to wash daily, particularly with soap than girls in the same age group. Whilst such a sex-specific association was not detected in the analysis in the current study, previous studies using interviews and observation in other countries have found boys to have poorer skin hygiene than girls [24]. A recent study examining the relationship between socioeconomic status and WASH practices in India also highlighted the fact that over 80% of mothers did use soap to wash themselves but only 20% used soap to wash their children [25].
The data from before and after school holidays, whilst a small dataset, highlighted a number of findings that are significant and warrant replication in future. Not only did the overall prevalence of tungiasis increase after the holiday, there was also a significant increase in the number of manipulated lesions. These are the sores and cicatrices that remain after an embedded sand flea has been purposively extracted with a sharp instrument and are a clear indication that the person recently had a viable embedded sand flea. The likely explanation is that the children have acquired more sand fleas whilst spending more time at home during the holiday, but they or a caregiver have extracted them. There was a significant interaction between the number of manipulated lesions and girls, again suggesting differential hygiene and caring behaviours. Teasing apart these complex linkages of economic status and behavioural traits will be important in future studies and might indicate school-aged boys to be an important target for prevention programs.
Surveys in other countries have identified tungiasis as a zoonosis with the involvement of dogs [13] and pigs [26, 27] in disease transmission. Pigs are not frequently kept in communities in coastal Kenya (only four students in the survey reported owning pigs), whilst goats and dogs were relatively common with 75% and 30% of students reporting household ownership, respectively. However, neither the previously published household study [11] nor the here presented school survey identified the possession of any animal species to be a risk factor. Whether this is an indication that transmission in these coastal communities is purely intra-domiciliary and does not involve an animal reservoir needs further investigation by examining livestock and companion animals for tungiasis. The current school survey did identify the absence of chickens in a household as a risk factor for severe disease, which may simply be another reflection of extreme poverty as a risk factor, which needs to be studied however more systematically.
The fact that the same household risk factors were identified in this study by asking the children about their homes, as in the corresponding household study where adults were interviewed, and observations made, suggests that school-based surveys are a reasonable alternative to the more expensive and time-consuming household surveys and can be used for nation-wide evaluation of tungiasis prevalence. Modelling based on past household surveys with full age profiles will enable extrapolation to the whole population. School-based surveys have the advantage of a high concentration of at-risk subjects to survey during day time when there is good light for examinations. To be able to examine all house occupants for tungiasis a team must visit during evening hours and at weekends, and still many family members may be absent. Houses may be far apart, and therefore surveying costly to achieve suitable sample sizes.
Targeting school-aged children in school for diagnosis and treatment of tungiasis, using recently evaluated safe and effective treatment options, namely dimeticone or neem oil [2, 28, 29], might be the most cost-effective way to reduce the disease burden given that the affected resource-poor communities do not have access to optimal medical care and limited ability to pay for expensive medications. However, the treatment must be provided every time a new infection in a child is detected by the teachers to prevent the life cycle being introduced into classrooms with a cracked or natural sand/mud floor, and to break the cycle at home. The disadvantage of conducting surveys and treatment programs only in schools means the most severely affected children who cannot walk to attend school, the elderly and disabled, who also tend to have severe infections, will be missed.
This factor was a possible limitation of the study, possibly causing bias in the study findings. However, the study included a similar proportion of severe cases to that seen in the household survey (11% and 15% respectively), so any effect on outcomes is likely to be minimal. The higher proportion of severe cases in the household survey was more likely to be due to the inclusion of the elderly who tend to have more severe infections [11].
Another limitation of the study was the low number of schools with non-cemented floors that were able to be included in the study, and that the one school that was entirely dirt floors, was the only private school, with the majorty of children in the lower age groups. However, any potential confounding was adjusted for in the statistical modeling.
Observations from our study suggest that up to 70% of tungiasis cases may be prevented through simple prevention methods, namely washing feet at least once a day with soap and installing hard floors in homes and schools. Hence, foot washing needs to be incorporated into hygiene and sanitation education campaigns of the current global efforts to achieve Sustainable Development Goal 6; “by 2030, achieve access to adequate and equitable sanitation and hygiene for all”. Tungiasis has been implicated to impact children’s learning capacity [9], consequently, there is a clear role and need for schools (head teachers and class teachers), public health officials, community health workers and NGOs to educate in and enforce good hygiene practices, particularly the use of soap for daily washing of feet. Furthermore, acknowledging that sealed classroom floors can contribute to tungiasis reduction, governments and education officials need to make the cementing of all classroom floors a priority, along with adequate water supplies and provision of soap for washing.
The installation of hardened floors in family homes is not as simple as it sounds, and requires research and potentially government investment. Those resource-poor, marginalized families affected by tungiasis cannot, under most circumstances, afford the cost of a cement floor, which in Kenya costs a minimum of $200 for a typical rural house of 6 x 4 m. In discussions with community members, it was highlighted that in the past communities used traditional methods for hardening floors such as regular smearing with a mix of soil and cow dung and termite mound soil, but these methods have ceased, and floors are no longer hardened (Elson, personal communication). Clearly there is a need for research into understanding why house floors are not hardened with the simple, cheap methods currently available, as well as developing alternative, locally available and affordable floor technologies that the most resource-poor families can install themselves.
|
10.1371/journal.ppat.1006516 | Tpc1 is an important Zn(II)2Cys6 transcriptional regulator required for polarized growth and virulence in the rice blast fungus | The establishment of polarity is a critical process in pathogenic fungi, mediating infection-related morphogenesis and host tissue invasion. Here, we report the identification of TPC1 (Transcription factor for Polarity Control 1), which regulates invasive polarized growth in the rice blast fungus Magnaporthe oryzae. TPC1 encodes a putative transcription factor of the fungal Zn(II)2Cys6 family, exclusive to filamentous fungi. Tpc1-deficient mutants show severe defects in conidiogenesis, infection-associated autophagy, glycogen and lipid metabolism, and plant tissue colonisation. By tracking actin-binding proteins, septin-5 and autophagosome components, we show that Tpc1 regulates cytoskeletal dynamics and infection-associated autophagy during appressorium-mediated plant penetration. We found that Tpc1 interacts with Mst12 and modulates its DNA-binding activity, while Tpc1 nuclear localisation also depends on the MAP kinase Pmk1, consistent with the involvement of Tpc1 in this signalling pathway, which is critical for appressorium development. Importantly, Tpc1 directly regulates NOXD expression, the p22phox subunit of the fungal NADPH oxidase complex via an interaction with Mst12. Tpc1 therefore controls spatial and temporal regulation of cortical F-actin through regulation of the NADPH oxidase complex during appressorium re-polarisation. Consequently, Tpc1 is a core developmental regulator in filamentous fungi, linking the regulated synthesis of reactive oxygen species and the Pmk1 pathway, with polarity control during host invasion.
| Cellular polarity is an intrinsic feature of filamentous fungal growth and pathogenesis. In this study, we identified a gene required for fungal polar growth and virulence in the rice blast fungus Magnaporthe oryzae. This gene has been named TPC1 (Transcription factor for Polarity Control 1). The Tpc1 protein belongs to the fungal Zn(II)2Cys6 binuclear cluster family. This DNA-binding motif is present exclusively in the fungal kingdom. We have characterised defects associated with lack of Tpc1 in M. oryzae. We show that Tpc1 is involved in polarised growth and virulence. The M. oryzae Δtpc1 mutant shows a delay in glycogen and lipid metabolism, and infection-associated autophagy–processes that regulate appressorium-mediated M. oryzae plant infection. The saprophytic fungus Neurospora crassa contains a Tpc1 homolog (NcTpc1) involved in vegetative growth and sustained tip elongation, suggesting that Tpc1-like proteins act as core regulators of polarised growth and development in filamentous fungi. A comparative transcriptome analysis has allowed us to identify genes regulated by Tpc1 in M. oryzae including NoxD, an important component of the fungal NADPH complex. Significantly, Tpc1 interacts with Mst12, a component of the Pmk1 signalling pathway essential for appressorium development, and modulates Mst12 binding affinity to NOXD promoter region. We conclude that Tpc1 is a key regulator of polarity in M. oryzae that regulates growth, autophagy and septin-mediated reorientation of the F-actin cytoskeleton to facilitate plant cell invasion.
| Rice blast disease is one of the most serious diseases of cultivated rice worldwide and is caused by the filamentous, ascomycete fungus Magnaporthe oryzae[1,2]. The disease is initiated when a conidium lands on the rice leaf surface. Here it germinates to produce a single germ tube that differentiates at its tip to develop a specialised infection structure called an appressorium[3]. During the initial stages of appressorium formation, a septum defines the developing appressorium from the rest of the germ tube following a single mitotic division in the germ tube[4]. When the appressorium matures, the three conidial cells and germ tube collapse due to infection-associated autophagy and are no longer viable after 24h[4]. Subsequently, a penetration peg emerges from the base of the appressorium and ruptures the leaf cuticle. A toroidal filamentous actin network forms at the base of the appressorium pore, scaffolded by septin GTPases[5]. Assembly of the four core septin GTPases is regulated by the Nox2 NADPH oxidase complex, which is required for re-modelling of the F-actin cytoskeleton and assembling the exocyst at the appressorium pore [6,7,8]. F-actin ring formation is necessary for penetration peg emergence and re-establishment of polarized growth at the point of plant penetration. After penetration, the fungal peg grows as a narrow, short primary invasive hypha[9], before differentiating into bulbous invasive hyphae during colonisation of the first invaded host cell[10]. Disease symptoms appear between 72h and 96h after initial infection and coalesce into large spreading necrotic lesions from which the fungus sporulates. M. oryzae has also the capacity to penetrate roots by means of hyphopodia and can colonize root tissue and spread systemically throughout the plant under laboratory conditions [11,12].
In this study, we report the identification of a novel Zn(II)2Cys6 transcriptional regulator involved in the early stages of plant infection by M. oryzae. The Zn(II)2Cys6 binuclear cluster domain (IPR001138, PF00172) is exclusively found in the fungal kingdom[13,14]. The six cysteine residues bind two zinc atoms, which coordinate folding of the domain involved in DNA-binding. Most Zn(II)2Cys6 proteins have been studied in Saccharomyces cerevisiae and Aspergillus species[13,15,16]. Typically, the Zn(II)2Cys6 proteins are pathway-specific activators under the control of major regulators[15,16,17,18]. The regulator of galactose catabolism in yeast, Gal4p[19], and the regulators of acetate assimilation FacB[20] and the aflatoxin cluster AflR[21] in A. nidulans, are among the best studied examples. Several Zn(II)2Cys6 transcriptional regulators have been studied in the rice blast fungus (S1 Table). Of the 175 members of the Zn(II)2Cys6 binuclear cluster family present in M. oryzae (S2 Table), only nine of them (MoCod1, MoCod2, Pig1, Tra1, Tdg3, Xlr1, Ara1, Far1 and Far2) have been examined in any detail[22,23,24,25,26,27] (S1 Table). A high-throughput gene knockout approach of 104 Zn(II)2Cys6 proteins in M. oryzae revealed large variation in their biological functions, and reported seven additional Zn(II)2Cys6 proteins to be required for plant infection, including Gpf1 and Cln2[28]. However, despite this information, the mechanistic insights into how the Zn(II)2Cys6 proteins govern M. oryzae cellular processes are largely unknown.
In this study, we characterize a novel mutant of M. oryzae that shows defects in pathogenicity and vegetative growth following its selection from a M. oryzae T-DNA insertional library. The T-DNA insertion is located within a gene (MGG_01285) encoding a Zn(II)2Cys6 binuclear cluster protein, which we name TPC1. This gene was not included in the large-scale gene knockout analysis of 104 Zn(II)2Cys6 proteins [28], although a global gene expression analysis of transcription factors revealed that TPC1 is overexpressed during development (conidiation, germination and appressorium formation), oxidative stress (methyl viologen treatment) and carbon starvation [29]. Here, we reveal the involvement of this transcriptional regulator in polarized growth, cell patterning and virulence in M. oryzae. Among the genes regulated by Tpc1 we found NOXD, an important component of the fungal NADPH complex. Significantly, Tpc1 interacts with Mst12 and mis-localises in the Δpmk1 background, linking Tpc1 to this pathogenicity-associated MAPK signalling pathway. We provide mechanistic insight into the role of Tpc1, a key regulator of polarity in M. oryzae that controls growth, autophagy and septin-mediated reorientation of the F-actin cytoskeleton to facilitate plant colonisation.
In order to identify novel infection-related genes we screened a total of 300 T-DNA transformants for their ability to infect rice roots using a M. oryzae insertional library[30]. The M1422 mutant developed very restricted disease lesions on roots and was selected for further characterization (Fig 1A). On leaves, M1422 produced only a small number of resistant-type lesions (Fig 1B and S1A Fig). Colonies of M1422 were also compact and reduced in size, when compared with the wild-type (Fig 1C).
The insertion site of the T-DNA within M1422 was located 0.9 kb downstream of the start codon of locus MGG_01285 in the M. oryzae genome (S1B Fig). This gene encodes a putative transcription factor that belongs to a Zn(II)2Cys6 binuclear cluster family. We named this gene Transcription factor for Polarity Control 1 (TPC1). The predicted coding region of TPC1 is ~2.6 kb long and encodes 839 amino acids (http://fungi.ensembl.org/index.html; MG8). The TPC1 predicted amino acid sequence includes a putative nuclear localisation signal (NLS) and one Zn(II)2Cys6 binuclear cluster DNA binding domain (S1B Fig).
A single T-DNA insertion in M1422 genome was detected by Southern blot hybridisation using the hygromycin phosphotransferase gene as a probe (S1C Fig). We also generated a second mutant in TPC1 by targeted gene replacement (S2A and S2B Fig). We complemented both M1422 and Δtpc1 with a C-terminal TPC1:GFP gene fusion under control of its native promoter. The complemented mutants recovered normal mycelial growth, colonial morphology and full virulence on rice (S1D, S1E and S2C Figs). We conclude that mutants M1422 and Δtpc1 are impaired in TPC1 function.
Two striking characteristics of M1422 and Δtpc1 were their impaired hyphal growth and colony morphology (Fig 1C, S1E and S2C Figs). Vegetative growth of Tpc1-lacking strains was severely compromised in both complete (CM) and minimal (MM) medium (p < 0.01), and showed compact colonies and non-invasive colony morphology (S1E, S1F and S2D Figs). In Neurospora crassa, a class of mutants with polarity defects also exhibited this type of colony morphology[31].
To analyse integrity of the cell wall, development of M1422 was evaluated in the presence of the anionic dyes, Congo Red (CR) and Calcofluor White (CFW), which interfere with fungal cell wall assembly by binding to β-1,4-glucan and chitin, respectively [32]. Additive growth defects were observed on M1422 development in the presence of CR but not in CFW (Fig 1D). In addition, mycelial growth was affected by NaCl-induced hyperosmotic stress (Fig 1E). High concentrations of NaCl (0.6M - 1.0M) changed the growth ratio in colonial size between wild-type and M1422, leading to an increase in the relative growth rate of M1422 compared to wild-type. Therefore, the lack of Tpc1 affected plant virulence, vegetative growth, colony morphology and hyperosmotic stress adaptation.
We observed that M1422 and Δtpc1 mutants sporulated poorly compared to wild-type (Fig 2A and S2E Fig). In addition, M1422 asexual spores showed defects in septation (numbers of cell per conidium) and conidial morphology (Fig 2B). Wild-type conidia were uniformly pyriform, three-celled spores. By contrast, in M1422, although the majority of conidia were three-celled (80%), a significant percentage of two-celled conidia (17%), single-celled (2%) and four-celled conidia (1%) were observed. Up to 26% of spores showed abnormal morphology in contrast to wild-type where less than 4% were misshapen (n> 300). We also found that appressorium development was affected in M1422 (Fig 2C). On hydrophobic coverslips, wild-type conidia germinated to form one germ tube that emerged from the apical cell and formed an appressorium within 4h-8h (Fig 2C and 2D). In M1422, 40% of conidia germinated from two cells. This percentage increased to 50%-60% with extended incubation time (4h-8h). Formation of two appressoria was rarely observed in wild-type conidia (Fig 2E). We conclude that M1422 is impaired in the normal spatial patterning of appressorium development.
The impairment of appressorium-mediated plant infection by TPC1 mutants suggested that it might play a critical role in penetration peg development[33]. Appressorium function is known to depend on autophagic cell death of conidia, prior to appressorium maturation[4,34]. Therefore, we investigated whether infection-associated autophagy proceeded normally and if conidia underwent autophagic cell death. A GFP:MoATG8 construct was introduced into M1422 to determine the spatial and temporal dynamics of autophagy (Fig 3A). MoATG8 encodes an autophagic, ubiquitin-like protein involved in autophagosome function and has been shown to be a reliable marker for autophagy[4,34]. Compared to the wild-type Guy11 (33.5±4.4), GFP:MoATG8-labeled autophagosomes accumulated in M1422 conidia in significantly smaller numbers (21.6±5.5; p<0.01). In both strains, the number of conidial autophagosomes decreased during germination, appressorium maturation and at the onset of spore cell death and was significantly lower in M1422 conidia and germ tubes (Fig 3A). However, autophagosome numbers increased significantly during appressorium maturation (8h; 16.1±4.9) and dropped considerably after conidial death (24h; 5.0±1.8) in wild-type, whereas autophagosome number remained relatively constant in M1422 during appressorium maturation (8.4±4.1) and even after conidial cell death (7.5±3.3).
Appressorium development is accompanied by rapid degradation of glycogen from conidia during germination and from appressoria during turgor generation[35,36]. We therefore determined glycogen levels during appressorium development using potassium iodide (KI). Comparative analysis of KI staining between wild-type Guy11 and M1422 showed differences during the onset and later stages of conidial cell death (8h and 24h; Fig 3B). In Guy11 glycogen depletion was observed (no staining) within both conidial cells and appressoria during development. In the M1422 mutant, conidial cells were also depleted of glycogen although the appressorium still contained high levels of glycogen (95%) during maturation. We also looked at lipid metabolism, which is an additional driver of turgor generation in M. oryzae. The triacylglycerol lipase degrades lipid bodies that move to the appressorium during development[24,37]. Accordingly, we followed lipid body distribution during appressorium maturation in Δtpc1 using Nile red (Fig 3C). We consistently visualised delayed degradation of lipid bodies in conidia and germ tubes in Δtpc1, which were evident at 9h and 12h after germination on coverslips. Using a cytorrhysis assay, in which hyperosmotic concentrations of a solute are applied to collapse appressoria, we estimated the internal solute concentration and turgor of appressoria of the two strains. We observed that Δtpc1 appressoria clearly collapsed at higher rates than the WT at 1M concentration of glycerol (Fig 3D). This suggests decreased in turgor within Δtpc1 appressoria, consistent with the observed delayed degradation of glycogen and lipid bodies. When considered together, these observations point that autophagy, and glycogen/lipid metabolisms are delayed during appressorium development in Tpc1-lacking strains.
Following maturation of the appressorium, a penetration peg emerges from the appressorial pore to penetrate the plant cuticle and successfully colonise the plant host. To assess whether repolarization was impaired in the M1422 mutant, a penetration assay was performed on onion epidermis and rice leaf sheath (Fig 4A). After 24h, 91% of wild-type conidia formed an appressorium effectively, penetrated and invaded onion epidermal cells. The majority of M1422 conidia (60%) germinated and produced an appressorium, but failed to penetrate and invade onion cells. Only 40% of M1422 appressoria formed a penetration peg, but were not able to invade the onion epidermal cells and spread away from the point of penetration. Similarly, on rice leaf sheath preparations 81% of wild-type conidia penetrated successfully compared to 21% of Δtpc1 mutant spores, which managed to develop a penetration peg but hardly ever spread to adjacent cells.
To examine how formation of the germ tube and penetration peg was compromised, we investigated cellular organization of the F-actin cytoskeleton[38], using the actin-binding protein fimbrin tagged with GFP (S3 Fig). Once wild-type conidia attached to the surface, fimbrin:GFP spots were observed at the periphery of the germinating cells (0h). However, conidia harvested from M1422 instead localised F-actin randomly at the periphery of the three cells of conidia and not preferentially in the germinating cell (white arrowheads, S3 Fig). The most clear mis-localisation defects were observed in mutant appressoria. Fimbrin was localised in discrete puncta at the periphery of Guy11 appressoria, but in contrast was dispersed within appressoria of the mutant (6h). Furthermore, the F-actin network was more diffuse and several pores were observed in M1422 mature appressoria (white arrowheads, 24h). These results suggest that re-polarization of the appressorium is adversely affected in the M1422 mutant. To confirm this, we also tracked gelsolin:GFP and Sep5:GFP in Δtpc1 mutant. The use of gelsolin:GFP and Sep5:GFP to follow actin reorganization has helped to understand cytoskeleton dynamics during infection-related development[6]. The disorganisation of the appressorial cytoskeleton and actin ring was evident in Δtpc1. Sep5 was mis-localised in all mutant appressoria and only 26% of mutant appressoria formed an intact actin ring with a central pore (Fig 4B). Consequently, TPC1 is required for the correct penetration peg emergence in M. oryzae.
The Tpc1:GFP fusion protein co-localised with histone H1:RFP in nuclei of vegetative hyphae, attached conidia (30 min), and germinated conidia (Fig 5A). Moreover, whenever Tpc1:GFP was observed in nuclei, GFP fluorescence was never observed in the cytoplasm or other organelles within conidia. The results are consistent with TPC1 encoding a transcription factor that acts within the nucleus during the initial stages of spore germination and appressorium development, and correlate with the observed overexpression of TPC1 in these fungal structures [29].
To investigate whether TPC1 is associated with specific or multiple regulatory networks, TPC1:GFP localisation was observed in conidia of different mutant backgrounds (Fig 5B). In the Δpmk1 MAPK mutant[39], Tpc1:GFP was observed within the cytoplasm but not in nuclei. By contrast, strong GFP fluorescence was visualised in conidial nuclei of the autophagy-defective Δatg1 and Δatg8 mutants, compared to the fluorescence observed in M1422 complemented with TPC1:GFP or Guy11 expressing TPC1:GFP. These results suggest that Tpc1 activity is associated with the Pmk1 MAP kinase signalling pathway, which regulates appressorium formation[39], and the control of autophagy[34].
We further analysed the link with between Tpc1 and the Pmk1 pathway by looking at the ability of Tpc1 to interact with components of this pathway in a yeast two-hybrid system. Strikingly, we observed that Tpc1 interacted with Mst12, a transcription factor that functions downstream of Pmk1[40], although Tpc1 did not interact with Pmk1 itself (Fig 5C). The mis-localisation of Tpc1 in Δpmk1 and its interaction with Mst12 strongly support Tpc1 involvement in this pathogenicity-associated MAPK signalling cascade.
We investigated the phylogenetic relationship of Tpc1 to other putative Magnaporthe Zn(II)2Cys6 proteins and the closest orthologues of Tpc1 in other fungal species (S4 and S5 Figs). We observed that the six cysteine residues of the DNA-binding domain (DBD) in the Zn(II)2Cys6 proteins were ordered in a conserved pattern, CX2CX6CX5-12CX2CX6-8C (S4A and S4B Fig). In M. oryzae, the Zn(II)2Cys6 binuclear cluster family is diverse and composed of 175 members (S2 Table). The closest orthologues of Tpc1 (MGG_01285) were identified using BLASTP and used to construct a phylogenetic tree (S4C and S5 Figs). Tpc1 clustered in a group with sequences from other Sordariomycetes, such as Fusarium graminearum, N. crassa, Chaetomium globosum and Podospora anserina. Based on this tree, Tpc1 is a single copy gene and has not been subject to paralogous duplications. Our phylogenetic analysis reflected the diversification of the Zn(II)2Cys6-containing proteins in the fungal kingdom. Interestingly, we did not find putative homologues of Tpc1 in S. cerevisiae or Schizosaccharomyces pombe using a BLASTP search.
In F. graminearum, it is remarkable that only 16% (46/296) of the mutants lacking Zn(II)2Cys6 transcription factors showed a phenotype, compared to the 42% (30/72) of N. crassa mutants or the 59% (61/104) of M. oryzae mutants[28,41,42]. Among the F. graminearum mutants with clear phenotypes is found the orthologue of M. oryzae TPC1 (FgTPC1 = FGSG_08769; GzZC108), which is required for plant infection, perithecia formation, synthesis of mycotoxins (ZEA, zearalenone; and DON, deoxynivalenol) and growth[41]. Similar to M. oryzae (Fig 1E and S2F Fig), the Δfgtpc1 mutant is more resistant than wild-type to hyperosmotic and oxidative stresses.
We further investigated functional conservation of Tpc1 in the saprotrophic filamentous fungus N. crassa, and characterized a N. crassa NcTPC1 deletion mutant (Δnctpc1; NCU05996), obtained from the Fungal Genetic Stock Centre[43]. The analysis of the alignment of M. oryzae and N. crassa Tpc1 proteins showed that they share 67% amino acid identity (S6A Fig). Strikingly, the Δnctpc1 mutant was severely reduced in vegetative growth compared to an isogenic wild-type strain (p <0.01) (S6B Fig), and its vegetative hyphae also formed compacted colonies. In addition, we observed that the Δnctpc1 mutant of N. crassa was less severely affected when exposed to increasing osmotic stress using NaCl, compared with the N. crassa wild-type strain (p <0.01) (S6C Fig). Similar tolerance effect was also found in F. graminearum Tpc1[41] and in M. oryzae Tpc1 (Fig 1E). Consequently, N. crassa Tpc1 also plays a significant role in growth and development of the fungus and its responses to abiotic stress.
Tpc1 contains a Zn(II)2Cys6 binuclear cluster DNA binding domain, which is found only in fungal proteins considered bonafide transcription regulators[13,14]. We carried out a comparative transcriptome analysis using the wild-type strain and the TPC1 deletion mutant to identify the biological processes and genes regulated by Tpc1. For this experiment, RNA was extracted from fungal material grown on cellophane on top of CM agar plates (S2D Fig). We considered it to be an optimal condition since fungal hypha is able to penetrate the cellophane, i.e. a change in polar growth occurs under these conditions, and allow us to obtain enough amount of RNA for subsequent microarray analysis. We identified 215 down-regulated genes and 185 genes to be up-regulated with at least a two-fold change in expression level in the Δtpc1 mutant (S3 Table). We classified all the genes that were up- and down-regulated into four functional groups according to potential roles in signalling (13 genes), cell wall biosynthesis or modulation of plant response (secreted proteins; 140 genes), metabolism (127 genes) and other functions (54 genes). Sixty-six genes encoded proteins that lacked any known domain. Remarkably, two gene ontology (GO) terms were found significantly enriched among these differentially expressed genes, the oxidation-reduction process (GO:0055114; 57 genes; p<0.001) and the oxidoreductase activity (GO:0016491; 58 genes; p<0.001).
Within the signalling functional group, two down-regulated genes encoded phosphatidyl ethanolamine-binding proteins (PEBP) that have been shown to regulate protein kinase A (PKA) and mitogen-activated protein kinase (MAPK) pathways[44,45]. Amongst the up-regulated genes, eight of them encoded transcriptional regulators, which suggests a link between the gene networks controlled by these transcriptional regulators and Tpc1.
The largest group of mis-regulated genes comprised 140 genes coding for secreted or cell wall-related proteins. Within this group, more than half of the members (74 genes) had no matches in databases. However, twenty-four genes were potentially involved in cell wall remodeling, and encompassed different types of glycosyl hydrolases (GH10, GH18, GH32, GH43, GH61 and GH81), seventeen proteases and two secreted phospholipases A2. Three Mas3/Gas1 paralogues and several effector proteins such as a Bas2-like, Bas113 and avrPi54 were also found[46,47,48,49]. The up-down regulation of two CFEM G-protein coupled receptors[50], including PTH11[51], suggested an alteration in the ability of the Δtpc1 mutant to perceive external signals.
The second largest group of genes with altered expression levels encoded proteins related with primary and secondary metabolism (127 genes). We found a significant number of them participating in oxidation-reduction processes (46%) and transport (9%). Alteration in nitrogen and glycerol metabolism was evidenced by the expression changes of four NmrA-like regulatory proteins[52], enzymes involved in amino acid biosynthesis, a glycerol kinase and the glycerol dehydrogenase Gcy1, an enzyme also associated with redox regulation in yeast[53]. Down-regulation of an α-glucosidase supported the glycogen degradation delay of Tpc1-lacking strains.
The last functional group included 54 genes encoding proteins that carry a wide range of biochemical roles. The reduced expression of the autophagy gene ATG22 and the up-regulation of three small chaperones Hsp20-like suggested the unbalanced signals for survival and cell death existent in Δtpc1[54]. Microtubule-dependent vesicle trafficking and cell cycle were also affected in the mutant as inferred from the misregulation of two dynamins, one kinesin light chain, one Marvel protein, two cyclins and the Cdc26 subunit. Genes involved in silencing pathways, spliceosomal snRNP assembly, tRNA processing, RNA-mediated heterochromatin silencing and translational arrest were also misregulated in Δtpc1, highlighting alterations in other cellular processes that regulate gene expression.
The majority of the Zn(II)2Cys6 binuclear cluster proteins are transcriptional activators and only few of them have been shown to act as repressors[14]. To identify novel pathogenicity genes we focused on genes that could play a role in TPC1-associated defects. Five out of the 133 down-regulated genes were selected for gene replacement (S3 Table; S7 Fig), including the conidiation-related gene CON6[55], a glycosyl transferase 18 gene (GH18) that undergoes a 50-fold increased expression in planta[48], and the two signaling-associated PEBP genes (S7 Fig). The PRO41/HAM-6 gene, which is required for hyphal fusion in Neurospora crassa and sexual development in Sordaria macrospora was also selected for the analysis[56,57]. We confirmed by RT-PCR that the five genes were down-regulated in the Δtpc1 mutant (S8A Fig).
Among the six deletion mutants generated, only Δpro41/Δham-6 displayed a severe pathogenicity-deficient phenotype (S7 Fig). Despite the links found between conidiogenesis and pathogenicity in M. oryzae[58,59,60], the Δcon6 mutant behaved like wild-type in planta. Similarly, Δgh18, Δpebp1, Δpebp2, and the double mutant Δpebp1Δpebp2 did not show any pathogenicity-associated defects possibly due to redundancy in related gene functions. Consequently, we selected Δpro41/Δham-6 mutants for further characterization.
The open reading frame of M. oryzae PRO41 was initially annotated in the EnsemblFungi database as HAM-6, a N. crassa gene required for cell fusion[61]. However, the orthologue of this protein was first characterized in S. macrospora and named Pro41[57,62]. Pro41 is a novel ER membrane protein required for fruiting body maturation in S. macrospora. Later, Pro41 was found to be the functional orthologue of the p22phox subunit of the NADPH oxidase complex in both Podospora anserina and Botrytis cinerea[63,64]. Therefore, we renamed the Pro41/Ham-6 protein NoxD (S8B Fig).
We looked at the growth of the M. oryzae ΔnoxD mutant in different media and stress conditions (Fig 6A and S8C Fig). ΔnoxD grew slightly faster than the wild-type on CM and MM, under salt stress (0.2 M LiCl, 0.4 M NaCl) and in Congo Red (CR). However, we did not observe differences in growth under carbon starvation, calcofluor white (CFW) or basic conditions (pH 9.5). Increased resistance to CFW was previously observed for M. oryzae Δnox1 but not for Δnox2[65], suggesting NoxD and Nox2 fulfill similar roles during cell wall biogenesis. Growth of M. oryzae ΔnoxD and Δnox1Δnox2 mutants in 1mM methyl viologen, 1mM H2O2 and 5mM H2O2 was similar or improved when compared to wild-type (S8D Fig). Thus, the lack of NoxD did not affect fungal growth under oxidative stressors in contrast to the growth defects displayed by B. cinerea NADPH oxidase mutants[64].
The infection ability of ΔnoxD was severely affected on leaves and roots (Fig 6B and S7 Fig), in accordance with the strong penetration defects displayed by Δnox1 and Δnox2[65]. The penetration defect displayed by ΔnoxD was confirmed using rice leaf sheaths (81% in the wild-type versus 19% in the mutant) and onion epidermis penetration assays (74% in the wild-type versus 22% in the mutant; Fig 6C). Subsequently, we crossed the ΔnoxD mutant with the rice isolate TH3, a M. oryzae strain of opposite mating type (Fig 6D). The inability to produce perithecia indicated that NoxD is required for sexual reproduction in M. oryzae.
To define whether superoxide production was impaired in the ΔnoxD mutant, we used nitroblue tetrazolium (NBT), which forms a dark-blue water-insoluble formazan precipitate upon reduction by superoxide radicals[65,66]. In the ΔnoxD mutant, we observed an increase in superoxide production at hyphal tips and a significant reduction in appressoria based on mean pixel intensity measurements (p<0.01) (Fig 7A). This was previously described for Δnox1Δnox2 mutants[65], and supports the existence of alternative routes for cellular ROS generation in M. oryzae during hyphal development. Since Δtpc1 was affected in oxidation-reduction processes, we also included Δtpc1 in this analysis. Increased superoxide production was found in Δtpc1 hyphal tips but to a lesser extent than nox mutants, while in appressoria Δtpc1 showed the highest superoxide levels among the strains analyzed, indicating that the lack of Tpc1 affects superoxide production pathways in M. oryzae.
A yeast two-hybrid assay was used to identify putative NoxD interactors. We found that M. oryzae NoxD interacts with the Nox1 NADPH oxidase subunit (Fig 7B) but not with Nox2 or NoxR, supporting previous work in B. cinerea and P. anserina[63,64].
To localize NoxD we generated C-terminal mRFP (cherry variant) and GFP translational fusions under the control of strong or native promoters, respectively. Both constructs fully complemented ΔnoxD plant infection defects (Fig 6B), which indicated that the C-terminal tag does not affect NoxD function, although expression of NoxD:mRFP was clearly stronger. M. oryzae NoxD was mainly observed in subapical vesicles and the plasma membrane of appressoria and conidia (Fig 7C). Co-localization of NoxD:mRFP with GFP containing the ER retention signal KDEL showed that the vesicles are closely associated with the ER, overlapping with some of them (Fig 7, white arrowheads). The subapical vesicles observed near plasma membranes and septa in M. oryzae structures correlated with the localisation of NoxD in P. anserina [63]. In P.anserina, these vesicles co-localised with the GFP:Idi7 reporter protein, suggesting that they originate from the ER and travel towards the vacuolar system [63].
The Nox2-NoxR complex is essential for septin-mediated cytoskeletal reorientation, whereas Nox1 is dispensable although may have important roles to play in maintenance and elongation of the penetration peg[6]. To test if NoxD was also involved in this process, we expressed the acting-binding protein gelsolin:GFP and Sep5:GFP in ΔnoxD. In the wild-type, both a septin and gelsolin ring was present at the appressorium pore[6] (Fig 7D). In the ΔnoxD mutant, however, Sep5:GFP formed a disorganized mass in the infection cell as previously reported for Δnox2 and ΔnoxR expressing Sep5:GFP[6]. Gelsolin:GFP rings in ΔnoxD also possessed distorted pores. Considering that gelsolin colocalizes with F-actin at the appressorial pore[6], the altered fluorescence pattern of gelsolin:GFP revealed that the toroidal F-actin ring was disorganized (Fig 7D). Previous reports showed that Sep5:GFP and gelsolin:GFP patterns in the Δnox1 mutant displayed normal conformation[6]. NoxD and Nox1 therefore appear to play alternative roles in cytoskeletal re-modeling in appressoria of M. oryzae.
The down-regulation of NOXD in Δtpc1 suggested that this gene may be directly regulated by Tpc1. To investigate this idea, we carried out chromatin immunoprecipitation (ChIP) followed by qPCR (Fig 8A and 8B). We observed that the promoter region of NOXD comprising the NOXD1, NOXD2 and NOXD3 fragments immunoprecipitated with Tpc1:GFP, which indicated that NOXD expression is regulated in vivo by this transcription factor. In addition, we performed electrophoretic mobility shift assays (EMSA) with M. oryzae Tpc1 and Mst12 since both proteins can interact in yeast two-hybrid assays (Fig 5C). We found that Mst12 strongly recognised the probe 1 located between -1120 and -643 upstream of the start codon of the NOXD gene (Fig 8C). Mst12 also recognized probes 2 and 3, but less strongly. Mst12 bound to the probes produced multiple bands, possibly due to the presence of several protein molecules on the biotinylated DNA (Fig 8D). Intriguingly, Tpc1 itself was not capable of recognizing any of the three probes under the conditions tested (Fig 8C). However, the addition of Tpc1 to Mst12 increased its DNA-binding capacity (Fig 8E), which is consistent with both, the ability of these proteins to interact, and with Tpc1 as modulator of Mst12 DNA-binding affinity. Increasing amounts of Tpc1 did not alter significantly Mst12affinity. Importantly, the promoter regions tested using these in vitro DNA-binding assays correlated perfectly with the enriched fragments obtained in the ChIP analysis, which supports that Mst12 and Tpc1 are part of a complex that coordinately regulate NoxD expression. To further confirm these results, we checked NOXD expression levels in the Δmst12 mutant and corroborated that they were reduced (Fig 8F). We also observed that MST12 and TPC1 genes were overexpressed when the corresponding partner was not present in the fungal cell (Fig 8G). We conclude that Tpc1 regulates NoxD expression through its interaction with Mst12 and confirm the link between Tpc1 and the participation of the Pmk1 pathway in the regulation of NoxD expression.
To cause disease in rice, M. oryzae forms a specialised cell called an appressorium, the development of which involves transitions from polarised to isotropic cellular growth, followed by rapid turgor-driven polarisation to penetrate the leaf surface. Understanding how these cellular transitions occur is critical to controlling the disease at an early stage, prior to entering the plant. In this study, we have identified a transcription factor, Tpc1 that plays a key role in regulating plant infection, due to its role in polarity control. We have also identified one putative mechanism by which it acts, via the regulated synthesis of reactive oxygen species and control of the NADPH oxidase complex, which regulates septin assembly and F-actin re-modelling at the base of the appressorium. Furthermore, we have found that Tpc1 directly participates in the Pmk1 pathway and is required for infection-associated autophagy, which are both essential pre-requisites for appressorium formation and function.
We observed that the TPC1 mutants formed compact colonies, which resembled the colony morphology shared by a class of mutants with polarity defects in N. crassa[31]. Conidial germination, and growth of vegetative hyphae were severely impaired in the two mutants lacking functional Tpc1, supporting defects in sustained tip elongation and establishment of polarity in apically-growing hyphae. Autophagy plays a major role in supplying amino acids, fatty acids, and glucose to maintain cellular functions during stress and starvation[67]. The absence of Tpc1 function altered the onset of infection-associated autophagy which occurs during appressorium development[68]. Conidial cell death is necessary to initiate appressorium penetration and it is regulated by the Pmk1 pathway [33]. Although M1422 conidia appeared able to undergo conidial cell death, the cellular localization of autophagosomes and glycogen/lipid deposits suggested that the process was delayed. Consistent with this observation, Tpc1:GFP was also highly expressed in Δatg1 and Δatg8 mutants impaired in autophagy, suggesting that the expression of TPC1 is de-repressed as a consequence of the inability to carry out autophagy and may therefore be an upstream positive regulator of infection-associated autophagy during appressorium maturation (Fig 9).
Autophagic cell death is linked with appressorium function and penetration in M. oryzae[4], and mutants lacking Tpc1 are also penetration defective. The formation of a penetration peg at the base of the appressorial pore is a cellular process intrinsically linked to polar growth[69,70]. The F-actin cytoskeleton plays a crucial role during germ tube re-polarisation and penetration peg emergence. We therefore investigated cytoskeletal dynamics during appressorium maturation in the mutant background. The network of F-actin observed with fimbrin:GFP in mature wild-type appressoria (8h-24h) delineated the appressorial pore, which was strikingly absent in M1422. This result together with the mis-localization of Sep5:GFP and gelsolin:GFP in Δtpc1 indicated that the F-actin network is disturbed in Tpc1-lacking strains.
The microarray analysis helped us to identify gene expression changes due to the lack of Tpc1, which correlated with the observed involvement of this protein in glycogen metabolism, autophagy and polar growth. Oxidation-reduction processes were also significantly affected in Δtpc1 including superoxide production pathways, likely due in part to the down-regulation of the fungal homologue of the p22phox NADPH subunit, the NOXD gene[63,64]. The M. oryzae ΔnoxD mutant was unable to infect rice leaves and roots. We established an interaction of NoxD with Nox1, but not with NoxR or Nox2, and confirmed the requirement of NoxD for superoxide generation and sexual reproduction in M. oryzae, consistent with NoxD functions in B. cinerea and P. anserina[63,64]. We also identified ΔnoxD defects in repolarization of the F-actin cytoskeleton during infection-related development, supporting the previous role described for the M. oryzae NADPH oxidase complex[6]. Remarkably, the disorganization pattern of gelsolin:GFP and Sep5:GFP in ΔnoxD was similar to that observed in Δnox2 and ΔnoxR mutants, whereas Δnox1 formed nearly intact ring shapes[6]. This result suggests that Nox1 and NoxD participate differently in septin-mediated cytoskeleton organization despite their interaction, and strengthens the view of the fungal NADPH oxidase as a dynamic complex[71]. It seems likely that a Nox2-dependent process initiates septin ring formation, while Nox1 is necessary for maintenance of this conformation (Fig 9). NoxD may therefore be associated at a relatively early stage in recruiting Nox1 to the appressorium pore, perhaps explaining why its absence results in a more severe phenotype with respect to actin and septin assembly at the pore. The role of NoxD, however, highlights that the Nox1 and Nox2 complexes are both necessary for penetration peg elaboration and extensive polar growth. It is worth noting that the tetraspanin PLS1 deletion mutants exhibit the same phenotype as Δnox2 in M. oryzae and P. anserina[6,72], suggesting that Pls1 may act as the missing link between Nox2 and NoxD subunits of the fungal NADPH oxidase complex (Fig 9). The recent discovery in B. cinerea of the RasGAP protein homologue IQGAP and its interaction with NoxD also points to IQGAP as a scaffold protein of the fungal NADPH complex[73]. In mammals, Nox complexes can act upstream[74] or downstream [75] of MAPK signaling pathways. IQGAP also interacts with different modules of MAPK- and Ca-dependent signalling cascades[73], pointing the link between Nox complexes and signalling cascades. Interestingly, the B. cinerea ΔnoxD mutant showed growth defects in the presence of oxidative stressors in contrast to the wild-type growth exhibited by the M. oryzae ΔnoxD mutant, which suggests a diversification of the cellular functions of NoxD in fungi. This result also hints differences in the regulation of ROS-mediated signalling pathways in the fungal kingdom.
Importantly, two lines of evidence support the direct involvement of Tpc1 in NoxD expression regulation. The ChIP analysis demonstrates that Tpc1:GFP immunoprecipitates in vivo with the NOXD promoter region. Tpc1 also regulates Mst12 DNA-binding activity in vitro using the corresponding NOXD promoter region, and indicates a direct participation of Tpc1 in the MAPK Pmk1 signalling pathway. Despite the ability of Tpc1 and Mst12 to regulate NOXD expression together, and their participation in common cellular processes such as penetration peg formation and plant invasion[9], Δmst12 and Δtpc1 mutants have different colony morphology. In contrast to Tpc1, Mst12 is dispensable for growth and appressorial turgor generation[9]. Consequently, Tpc1 has the ability to modulate expression of genes that participate in additional cellular processes, either by interacting with other transcription factors, or activating directly the expression of different genes.
Here, we identified one potential mechanism by which the transcription factor Tpc1 regulates appressorium maturation and plant infection. The loss of pathogenicity associated with M. oryzae and F. graminearum TPC1 mutants and similar growth defects associated with the N. crassa Δnctpc1 mutant, suggest that Tpc1 plays a key role as a transcriptional regulator in the re-establishment of polarity and the response to numerous signalling pathways, such as the Pmk1 MAP kinase and Atg1 kinase cascades. The role of Tpc1 in appressorium-mediated plant infection appears to be associated with the NADPH oxidase-dependent re-polarisation process of the appressorium, and the associated physiological changes such as autophagy, glycogen/lipid mobilisation and asymmetric reorganization of the F-actin cytoskeleton. Future studies will allow further dissection of this role and precise definition of the biological processes regulated by Tpc1 in filamentous fungi.
M. oryzae was routinely incubated in a controlled temperature room at 25°C with a 12h light/dark cycle. Media composition of complete medium (CM), minimal medium (MM), minimal medium without carbon (MM-C) or nitrogen (MM-N), and DNA extraction and hybridisation were all as previously described[76]. Growth tests were carried out with 7 mm plugs of mycelium from Guy11 and the M1422 mutant strains as initial inoculum. The wild-type Neurospora crassa strain and isogenic deletion mutant NCU05996 were obtained from the Fungal Genetics Stock Centre (FGSC, Kansas City, Missouri, USA). Vogel’s minimal medium was used for cultivation of N. crassa strains at 25°C with a 12h light/ dark cycle and for stock-keeping at 4°C (http://www.fgsc.net/Neurospora/NeurosporaProtocolGuide.htm). Growth tests were carried out on Vogel plates with 5 mm plugs of mycelium from N. crassa wild-type (wt) and NcTPC1 KO strains. Plates were incubated at 25°C for 2 days. M. oryzae leaf and root infection assays were carried out, as previously described [30,77].
Conidia were harvested using 2 ml of sterile water per plate after fungal cultures were incubated at 25°C for a period of 10 days on CM. Calculations were then carried out to determine the number of conidia generated per cm2 of mycelium using a Neubauer counting chamber. Values are the mean ± SD from >300 conidia of each strain, which were measured using the ImageJ software [78]. Photographs were taken using the Zeiss Axioskop 2 microscope camera using differential interference contrast (DIC) microscopy or epifluorescence. Conidia were stained with 5μl calcofluor white (CFW) solution (Fluka) and incubated at 25°C for 30 minutes. Cell number per conidium was determined by visualizing septa with CFW. Appressorium-mediated penetration of onion epidermal strips was assessed using a procedure based on Chida and Sisler[79]. A conidial suspension at a concentration of 1 x 105 conidia mL-1 was prepared and dropped onto the adaxial surface of epidermal layers taken from onion. The strips were incubated in a moist chamber at 25°C and penetration events scored 24h to 48h later by recording images with an Olympus IX81 inverted microscope system. Leaf sheath assays were carried out as previously described [10]. Glycogen staining solution contained 60 mg of KI and 10 mg of I2 per milliliter of distilled water. Glycogen deposits are visible immediately. For cytorrhysis assays, 105 spores were allowed to form appressoria for 18h on coverslips prior the addition of external glycerol (1M or 3M). After 10 minutes in glycerol ~500 appressoria were analyzed in each biological replica; experiment was carried out by triplicate. To visualize lipid droplets, conidia were allowed to germinate in water on coverslips. After 0h, 2h, 9h and 12h water was removed and conidia directly stained with Nile red (Nile Red Oxazone (9-diethylamino-5Hbenzo[alpha]phenoxazine-5-one; Sigma). Nile red was used to 2.5 mg/ml diluted in 50mM Tris/Maleate, pH 7.5 and polyvinylpyrrolidone (PVP) (2–3% w/v). Lipid droplets begin to fluoresce within seconds. Samples were visualized under a confocal laser scanning microscope using a 561 nm excitation wave length and a long pass emission filter (592–700 nm). All images were taken using the same parameters.
Gene deletion constructs were generated using multisite gateway technology (Invitrogen) as previously described[77,80]. TPC1, CON6, GH18, PEBP2 and NOXD coding sequences were replaced by the hygromycin resistance cassette and PEBP1 by the sulfonylurea resistance cassette in the gene replacement constructs. Primers for constructing entry plasmids are described in S4 Table. Fungal transformants generated by Agrobacterium-mediated transformation [81] were selected growing in DCM solid media supplied with 5-fluoro-2’-deoxyuridine (50μM) and 200μg/ml Hygromycin or 150μg/ml Chlorymuronethyl in the case of Δpebp1. DCM is 1.7 g yeast N-base without amino acids, 1.0 g NH4NO3, 2.0 g of L-asparagine and 10 g of D-glucose. Knockout strains were confirmed by PCR or Southern blotting using radioactive probes (32P; primers listed in S4 Table). Sequence data and gene numbers used in this study were taken from EnsemblFungi (Magnaporthe oryzae MG8; http://fungi.ensembl.org/index.html).
To determine the localisation of Tpc1, live-cell imaging was performed using a M. oryzae Guy11 strain containing two constructs, histone H1 tagged with red fluorescent protein (H1:RFP; tdTomato) to visualize nuclei [82], and TPC1:GFP. For the construction of a functional TPC1:GFP gene fusion, primers were designed in order to amplify the TPC1 (MGG_01285) promoter region and ORF from genomic DNA of M. oryzae Guy11 (S4 Table). The TPC1_GFP_F forward primer was designed approximately 1.3 kb upstream from the TPC1 start codon to include a substantial component of the promoter sequence. The TPC1_GFP_R reverse primer spanned the stop codon and contained a complementary region to the GFP sequence. GFP primers were designed to amplify the 1.4 kb sGFP:TrpC construct cloned in pGEMT. Both fragments were joined together by fusion nested PCR. The amplicons were cloned into pGEMT-easy digested with EcoRI. The 4.3 kb TPC1:GFP fragment was gel purified and cloned into pCB1532 that had previously been digested with EcoRI. The pCB1532 vector contains the 2.8 kb ILV1 gene, which encodes the acetolactate synthase-encoding allele bestowing resistance to sulfonylurea[83]. The resulting plasmid pCB1532-TPC1:GFP was used to transform protoplasts of M1422 mutant. For all rounds of PCR amplification, Phusion High-Fidelity DNA polymerase (Finnzymes, Thermo Fischer Scientific Inc.) was used, following the manufacturers’ guidelines for PCR conditions.
The GFP:MoATG8[34] and the FIM:GFP constructs were used to transform protoplasts of M1422 mutant. Protoplast generation and transformation were carried out as previously described[76]. The GFP:MoATG8 and the FIM:GFP protein fusion vectors were generated using the native M. oryzae MoATG8 gene (MGG_01062) and the native M. oryzae fimbrin-encoding gene (MGG_04478), respectively. Both fragments were cloned into pCB1532 vector that contains the 2.8 kb ILV1 gene, which encodes the acetolactate synthase allele conferring sulfonylurea resistance. Transformants showing identical growth and colony morphology to the background strain were selected for further examination using epifluorescence or confocal microscopy. At least three different transformants of each were independently analysed.
The TPC1:GFP gene fusion was cloned into pCB1532 vector (SURR) and used to transform protoplasts of Guy11 expressing Histone H1 fused with red fluorescent protein (H1:RFP)[33], and also introduced into isogenic Δpmk1, Δatg1 and Δatg8 mutants. Transformants were selected for further examination using confocal microscopy and verified as containing a single copy of the gene fusion construct by Southern blot hybridisation. At least three different transformants of each were used in all experiments.
Using a modified protocol of LiCl method[77], RNA was extracted from 8-day old fungal mycelia grown on cellophane placed on top of CM agar plates (S2E Fig). Two to three additional washes with phenol:chloroform were implemented to avoid RNA degradation from cellophane samples. RNA quality control was carried out with Agilent RNA 6000 Nano kit (ref. 5067–1504). Four biological replicates were independently hybridized for each transcriptomic comparison. Each of these replicates derived from three technical repetitions. Slides were Agilent Magnaporthe II Oligo Microarrays 4x44K (ref. 015060). Background correction and normalization of expression data were performed as previously described[77]. Hybridizations and statistical analysis were conducted by the Genomics Facility at the National Biotechnology Centre (Madrid, Spain). The GO term analysis was carried out with gProfiler[84]. Enriched motifs were not found when using the promoter regions of the 185 up-regulated genes. Microarray data are available in the ArrayExpress database (EMBL_EBI) under accession number E-MTAB-4127.
In-Fusion Cloning based on in vitro homologous recombination was performed to generate vectors including NoxD and Tpc1 into the pGADT7 prey vector, and Nox1, Nox2 NoxR, Pmk1 and Mst12 into the pGBKT7 bait vector. Genes were amplified from M. oryzae cDNA derived from mycelia grown on liquid CM using primers with a 15bp overhang and restriction site complementary to the target vector (S4 Table). For NoxD, a 435bp fragment was amplified, for Nox1, a 1662bp fragment was amplified, for Nox2, a 1749bp fragment was amplified, and for NoxR, a 1578bp fragment was amplified. Respective fragments were cloned into pGBKT7 and pGADT7 plasmids linearized by digestion with EcoRI and SmaI. Yeast two-hybrid assays using pGADT7 or pGBKT7 (Clontech) based constructs were performed according to the manufacturer’s instructions (MATCHMAKER Gold Yeast Two-Hybrid System).
For the construction of NoxD:GFP, primers were designed to amplify the ORF including 2kb upstream of the start codon, GFP and TrpC terminator with 15bp overhangs complementary to adjacent fragments (S4 Table). Fragments were ligated into pCB1532[83], which carries the sulphonyl urea resistance cassette and had been digested with BamHI and HindIII and this construct transformed into of the wild-type strain Guy11 using protoplasts[6]. The NoxD:mRFP construct was generated using multi-site gateway technology (Life Technologies) with the entry mCherry-withSTOP and destination SULPH-R3R4 vectors[77], and PCR fragments amplified from M. oryzae genomic DNA using Phusion DNA polymerase (NEB) and primers detailed in S4 Table. Appressorium development assays were performed on hydrophobic borosilicate glass coverslips (Fisher Scientific), as described previously[6]. For epifluorescence microscopy, conidia were incubated on coverslips and observed at each time point using an IX-81 inverted microscope (Olympus) and a UPlanSApo X100/1.40 oil objective. All microscopic images were analyzed using MetaMorph (Molecular Devices). Confocal imaging was performed with a Leica SP8 microscope.
To confirm microarray results, the relative abundance of gene transcripts were analysed by qPCR (S4 Table). One μg of total RNA from 8-day old fungal mycelia grown on cellophane placed on CM agar was reverse transcribed using PrimeScript RT reagent Kit (Takara). The average threshold cycle (Ct) was normalized against actin transcript and relative quantification of gene expression was calculated by the 2ΔΔCt method[85]. Primer efficiency was tested using dilutions of cDNA samples. qPCR reactions were carried out with 1 μl of reverse transcribed products and fast-start DNA master SYBR green I kit (Roche Diagnostics) in a final reaction of 20 μl using the following program: one cycle of 95°C for 4 min and 40 cycles of 94°C for 30 s and 60°C for 30 s. The Ct (threshold cycle) provided a measure for the starting copy numbers of the target genes. Three technical repetitions from three independent biological experiments were used for each gene. For ROS detection in M. oryzae fungal structures, NBT staining[65] and quantification method of pixel intensities in hyphal tips[86] were carried out as previously described.
Two strains, the Δtpc1 mutant expressing TPC1:GFP and M. oryzae wild-type Guy11 strain as negative control were used for this experiment. Mycelia were grown in liquid CM at 25°C for 48 h in a shaker (120 rpm), and collected using two layers of Miracloth. Harvested mycelia were washed extensively with sterile water. To crosslink DNA and proteins, one gram of each washed mycelium was treated with 1% formaldehyde in 20 mM HEPES pH 7.4 buffer for 20 min with continuous shaking at 100 rpm. Then, 0.125 M glycine was added and incubated at room temperature for an additional 10 min to stop crosslinking. Mycelia were harvested with Miracloth, rinsed with water removing excessive water by squeezing and immediately frozen in liquid nitrogen, grinded into a fine powder and stored at -80°C until used. ChIP was conducted according to published procedures with some modifications [87]. 600 mg of each mycelium powder was used for chromatin extraction and sonication. The powder was added into 10 ml of Extraction buffer 1 (0.4 M sucrose, 10 mM Tris-HCl pH 8, 10 mM MgCl2, 5 mM β-mercaptoethanol/β-ME and Protease Inhibitors Complete-PIC/Roche) and mixed by vortexing. The solution was filtered through a double layer of Miracloth and centrifuged at 5000 g for 10 min at 4°C. The pellet was resuspended in 1 ml of Extraction buffer 2 (0.25 M sucrose, 10 mM Tris-HCl pH 8, 10 mM MgCl2, 1% Triton X-100, 5 mM β-ME and PIC) and centrifuged at 5000 g for 10 min at 4°C. The pellet was resuspended in 300 μl of Extraction buffer 3 (1.7 M sucrose, 10 mM Tris-HCl pH 8, 0.15% Triton X-100, 2 mM MgCl2, 5 mM β-ME and PIC) and, carefully layered on the top of additional 600 μl of extraction buffer 3. Then, samples were centrifuged at 16000 g for 60 min at 4°C. The chromatin pellet was resuspended in 300 μl of Nuclei Lysis Buffer (50 mM Tris-HCl ph 8, 10 mM EDTA, 1% SDS and PIC) and sonicated for 25 min at 4°C, operating a pattern of 30 sec ON and 30 sec OFF, at high power level in the Bioruptor Plus (Diagenode, Liege, Belgium) to obtain DNA fragments ranging from 500 to 1,000 bp. The chromatin solution was centrifuged at maximum speed for 5 min at 4°C to pellet cell debris. The supernatant was kept as chromatin solution and a small aliquot (10%) was stored as input DNA control. For each immunoprecipitation, 15 μl of Dynabeads Protein A magnetic beads (ref. 10001D, Life Technologies) was washed twice with 500 μl ChIP dilution buffer (1.1% Triton X-100, 1.2 mM EDTA, 16.7 mM Tris-HCl pH 8, 167 mM NaCl and PIC). Then, anti-GFP antibody (ref. A6455, Life Technologies) was added and incubated with gentle rotation for 1h at 4°C in 50 μl ChIP dilution buffer. Prepared anti-GFP coated beads were washed twice with 500 μl ChIP dilution buffer and resuspended in 100 μl of ChIP dilution buffer. For each immunoprecipitation, the latter and 100 μl of chromatin solution were gathered together and diluted up to 1 ml of ChIP dilution buffer. All immunoprecipitations were incubated overnight at 4°C with gentle rotation, then washed with a serie of wash buffers (2 washes with Low Salt Wash Buffer: 150 mM NaCl, 0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl pH 8; one wash with High Salt Wash Buffer: 500 mM NaCl, 0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl pH 8; one wash with LiCl Wash Buffer: 0.25 M LiCl, 1% NP-40, 1% sodium deoxycholate, 1 mM EDTA, 10 mM Tris-HCl pH 8, 2 washes with TE Buffer: 10 mM Tris-HCl pH 8, 1 mM EDTA). Immunoprecipitated DNAs and Input DNA control were reverse-crosslinked at 95°C for 10 min with 200 μl of 10% chelex 100 resin to remove any trace of metals. DNA samples were treated with proteinase K that was inactivated afterwards. After centrifugation, supernatants of DNA samples were stored at -20°C until used. Immunoprecipitated chromatin was diluted 10 times for qPCR analysis (primers listed in S4 Table). This was performed using a Roche LightCycler 480 machine. qPCR reactions were carried out using either 2 μl of input DNA or 2 μl of immunoprecipitated chromatin in a final reaction of 12 μl with the following program: one cycle of 95°C for 5 min and 58 cycles of 94°C for 10 s, 60°C for 10 s and 72°C for 10 s. The Ct (threshold cycle) provided a measure for the starting copy numbers of DNA. Three technical repetitions from 4 independent biological experiments were used. Ct values were used to calculate ratios evaluating the fold difference between experimental samples (GFP-tagged or untagged wild-type strains) and normalized the input. We normalized with “Fold Enrichment Method” using the untagged strain. The Wilcoxon Mann Whitney test was applied to analyze the difference between two independent groups. Statgraphics software was used to make pairwise comparisons between GFP-tagged strain and untagged wild-type strain.
M. oryzae MST12 and TPC1 cDNAs derived from mycelial RNA were cloned by PCR using a high fidelity Q5 DNA polymerase (NEB), primers (S4 Table) and the restriction enzymes BamHI-NotI and EcoRI- NotI for MST12 and TPC1 respectively, into a modified pET28 vector (5,667bp; Novagen). MST12- and TPC1-containing plasmids were transformed in E. coli Rosetta DE3 (Novagen) and colonies grown in LB medium containing chloramphenicol (34 μg/L) and kanamycin (50 μg/L) until reaching OD600nm = 0.8. Protein expression was induced 4 hours at 28°C with 1 mM IPTG (Sigma-Aldrich). Centrifuged cell pellets (30 min at 7000g) were resuspended in lysis buffer (20 mM sodium phosphate pH 8, 300 mM NaCl and one tablet of PIC/50 ml, 1 mM PMSF and 50 μg/ml Dnase I), lysed by sonication and pelleted at 4°C and high speed (20 min at 20,000g). Recombinant proteins were purified from clear lysate by metal affinity chromatography (HisTrap HP 1 ml, #17-5247-01 GE Healthcare) in denaturing conditions using 6 M Urea and eluted with 250 mM imidazole containing buffer. Samples were desalted on PD10 column (#17085101 GE Healthcare) to remove urea and imidazole using buffer (20 mM sodium phosphate pH 8, 10% glycerol and PIC). Protein samples purity was evaluated by SDS-PAGE.
EMSA probes were generated as follows. Amplified by PCR fragments using primers listed in S4 Table were prepared using modified Biotin 3’end DNA labeling procedure (#89818 Thermo-Scientific). Briefly, each ~500pb purified PCR products was KpnI-digested, purified and labelled (5 pmol of each probe) with Biotin-11-UTP and Terminal Deoxinucleotidyl Transferase at 37°C for 1 hour. Biotinylated probes were purified by Chloroform:IAA (24:1) extraction and stored at -20°C until use. EMSA reactions (20 μl) contained 10 mM Tris HCl pH 7.5, 50 mM KCl, 16 mM DTT, 1 mM ZnCl2, 1 mM MgCl2, 1% Glycerol, 50 ng/μl Poly dI-dC (#20148E Thermo-Scientific), 10 μg BSA, Protease inhibitor complete (Roche), and 80 fmol of biotinylated probe. Before probe addition proteins (0–12 μM) were incubated in binding buffer for 10 min, then probe was added and incubated during 30 min at room temperature before loading. The EMSA gel (0.2% agarose, 5% polyacrylamide, 1% glycerol in TBE 0.5x) was run for 2h 100V in TBE 0.5x and then transferred to a Hybond-XL nylon membrane (#RPN203S GE Healthcare) at 400 mA for 1 hour. The membrane was UV crosslinked at 120mJ/cm2. Detection was performed with stabilized streptavidin-horseradish peroxidase conjugate (#21134 Thermo-Scientific) and enhanced chemiluminescent substrates (#32106 Thermo-Scientific) following LightShift Chemiluminescent EMSA procedure (#20148 Thermo-Scientific).
First, 141 M. oryzae protein sequences containing a fungal Zn(II)2Cys6 binuclear cluster domain (PF00172) were identified from the Magnaporthe sequence database at the Broad Institute (http://www.broadinstitute.org/annotation/fungi/magnaporthe) and the Fungal Transcription Factor Database (http://ftfd.snu.ac.kr/intro.php). HMMsearch from HMMER3[88] was used to screen the genome assembly of M. oryzae proteins with the fungal Zn2Cys6 profile hidden Markov model pHMM zn_clus_ls.hmm (PF00172.13) from Pfam database[89] (http://pfam.xfam.org/). Subsequently, gene numbers were updated using the MG8 genome version of EnsemblFungi database (http://fungi.ensembl.org/index.html). Out of these 141 sequences, only 113 had a full length zinc cluster domain, and extra six closest sequences were included to build S5 Fig. Additional Zn(II)2Cys6 proteins found in Lu et colleagues[28] were included in S2 Table. Basic Local Alignment Search Tool (BLAST) was used to find orthologous proteins of TPC1/MGG_01285 (http://blast.ncbi.nlm.nih.gov/Blast.cgi). Protein sequences were pre-aligned using HMMalign and the pHMM zn_clus_ls.hmm (S4 Fig) from Pfam. The Zn(II)2Cys6 binuclear cluster domain region was extensively manually aligned in BioEdit (http://www.mbio.ncsu.edu/BioEdit/BioEdit.html). Unambiguous aligned positions were used for the subsequent phylogenetic analyses. The maximum likelihood (ML) analyses were performed with the program PhyML version 3.0.1[90]. All trees were visualised using the program Figtree (http://tree.bio.ed.ac.uk/software/figtree/).
M. oryzae sequence data from this article can be found in the GenBank/EMBL-EBI (EnsemblFungi) databases under the following accession numbers: TPC1 (MGG_01285), PMK1 (MGG_09565), MST12 (MGG_12958), ATG1 (MGG_06393), ATG8 (MGG_01062), CON6 (MGG_02246), GH18 MGG_04732, NOXD (MGG_09956), PEBP1 (MGG_06800), PEBP2 (MGG_14045), NOXR (MGG_05280), NOX1 (MGG_00750), NOX2 (MGG_06559), FIMBRIN (MGG_04478) GELSOLIN (MGG_10059), ACTIN (MGG_03982), YDIU (MGG_03159) and SEP5 (MGG_03087).
|
10.1371/journal.pgen.1006061 | Additive Expression of Consolidated Memory through Drosophila Mushroom Body Subsets | Associative olfactory memory in Drosophila has two components called labile anesthesia-sensitive memory and consolidated anesthesia-resistant memory (ARM). Mushroom body (MB) is a brain region critical for the olfactory memory and comprised of 2000 neurons that can be classified into αβ, α′β′, and γ neurons. Previously we demonstrated that two parallel pathways mediated ARM consolidation: the serotonergic dorsal paired medial (DPM)–αβ neurons and the octopaminergic anterior paired lateral (APL)–α′β′ neurons. This finding prompted us to ask how this composite ARM is retrieved. Here, we showed that blocking the output of αβ neurons and that of α′β′ neurons each impaired ARM retrieval, and blocking both simultaneously had an additive effect. Knockdown of radish and octβ2R in αβ and α′β′ neurons, respectively, impaired ARM. A combinatorial assay of radish mutant background rsh1 and neurotransmission blockade confirmed that ARM retrieved from α′β′ neuron output is independent of radish. We identified MBON-β2β′2a and MBON-β′2mp as the MB output neurons downstream of αβ and α′β′ neurons, respectively, whose glutamatergic transmissions also additively contribute to ARM retrieval. Finally, we showed that α′β′ neurons could be functionally subdivided into α′β′m neurons required for ARM retrieval, and α′β′ap neurons required for ARM consolidation. Our work demonstrated that two parallel neural pathways mediating ARM consolidation in Drosophila MB additively contribute to ARM expression during retrieval.
| One of tantalizing questions in neuroscience is how the brain processes memory. Studies in animal models such as fruit fly have brought innovations addressing the general principles underlying memory processing such as acquisition, consolidation, and retrieval. Here, we revealed an additive expression of aversive consolidated memory through fly mushroom body (MB) subsets. By thermogenetic blockade of neurotransmission, we identified the necessity of MB αβ and α′β′ neurons, and their respective downstream neurons, for consolidated memory retrieval. We also showed that MB αβ and α′β′ neurons harbor distinct signaling pathways for memory consolidation by genetic manipulation. Notably, the combinatorial assays of neurotransmission blockade and genetic manipulations confirmed the independency between the two sets of double-layered parallel circuits. Our work, together with previous finding of two respective modulatory neurons upstream of αβ and α′β′ neurons, favor the notion that memory is consolidated in different brain regions/circuits in parallel and later additively retrieved for behavioral outcome.
| Memory expression requires sequential processing such as acquisition, consolidation, and retrieval. The fruit fly (Drosophila melanogaster) is of great interest to neuroscientists studying memory because of its short lifespan, relatively simple brain, and powerful genetic tools. In fly aversive olfactory conditioning, the association between the electric shock and odor identity is first registered in odor-responsive MB γ neurons by dopamine signaling [1]. After acquisition, memory can be dichotomized to anesthesia-sensitive memory (ASM) and anesthesia-resistant memory (ARM) depending on the susceptibility to retrograde amnesia, and 3-h memory comprises equal extent of ASM and ARM [2–4]. ARM has been seen as a stable consolidated memory less costly than long-term memory and can be assessed by cold-induced anesthetization [4–6]. Many lines of evidence from gene mutation, RNA interference (RNAi)-mediated knockdown, and manipulation of neuronal activity support a model in which ASM and ARM constitute two independent types of memory that record the same episode [7, 8]. Here, we focus on deciphering the ARM processing.
The Drosophila MB, a paired neuropil structure that consists of ~2000 neurons, can be divided into subsets of neurons that comprise the αβ, α′β′, and γ lobes [9, 10]. The cell bodies of the MB neurons reside near the dorsal posterior surface of the protocerebrum. The dendrites of MB neurons form the calyx, where they receive olfactory input from the antennal lobe and transform it into a sparse neural code that benefits the establishment of associations between conditioned and unconditioned stimuli [11–14]. The axons of MB neurons project anteriorly to form a stalk-like pedunculus before branching into the vertical and horizontal lobes. The bifurcated axons of the αβ neurons form the α and β lobes of the vertical and horizontal lobes, respectively, and those of the α′β′ neurons form their corresponding α′ and β′ lobes. In contrast, the axons of the γ neurons constitute the horizontal γ lobe. According to the laminar zones within each MB lobes, α/β lobes have been further divided into core, surface, and posterior strata, while α′/β′ lobes have been divided into anterior, middle, and posterior strata [15]. Recently, Aso et al. used split-GAL4 screen and single-cell imaging to divide the MB neurons into seven cell types: αβp, αβs, αβc, α′β′ap, α′β′m, γmain, and γd [16, 17].
ARM requires bruchpilot expression in αβ neurons and acute radish expression right before training [2, 18, 19]. Two pairs of MB modulatory neurons, dorsal paired medial (DPM) and anterior paired lateral (APL) neurons, broadly innervating the MB also contribute to ARM formation by the serotonergic neurotransmission toward αβ neurons and the octopaminergic toward α′β′ neurons, respectively [15, 20–22], suggesting the presence of two parallel circuits for ARM consolidation in the MB. Here, we found that the outputs from both αβ and α′β′ neurons were required for complete ARM retrieval, suggesting two parallel ARM consolidation circuits additively contribute to retrieval. By RNAi-mediated knockdown experiments, the functional roles of radish and octopamine signaling in ARM consolidation were segregated in αβ and α′β′ neurons. Based on the segregation, we utilized the rsh1 mutant background to remove radish-dependent ARM that is consolidated in αβ neurons. In these rsh1 mutant flies, blocking output from αβ neurons during ARM retrieval failed to further impair the ARM expression, whereas blocking output from α′β′ neurons did, indicating the radish-independent ARM is mainly retrieved by the output of α′β′ neurons. We also showed that two MB output neurons downstream of αβ and α′β′ neurons, the MBON-β2β′2a and MBON-β′2mp neurons, respectively, mediated the ARM retrieval via glutamatergic transmission, supporting the model that two MB subsets, the αβ and α′β′ neurons, additively contribute to ARM consolidation and retrieval. Finally, we characterized the GAL4 lines expressing in all or subsets of MB α′β′ neurons [17], and functionally divided the α′β′ neurons into α′β′m neurons and α′β′ap neurons, whose outputs are required for ARM retrieval and ARM consolidation, respectively.
We first adopted a collection of GAL4 lines that represent specific subsets of MB neurons: 1471-GAL4 and VT44966-GAL4 for γ neurons (Fig 1A and 1C); C739-GAL4 and VT49246-GAL4 for αβ neurons (Fig 1E and 1G); and VT30604-GAL4 and VT57244-GAL4 for α′β′ neurons (Fig 1I and 1K). We expressed the temperature-sensitive dominant-negative dynamin mutant transgene, shibire (UAS-shits) [23], using each of these GAL4 lines. Flies were trained to associate the electric shock and odor using classical conditioning protocol at 23°C, and a 2-min cold shock 2 hours after training was applied to assess 3-h ARM [4, 24]. We tested which MB subset output is required for 3-h ARM retrieval by shifting the temperature to 31°C 15 min prior to and during the test. Following previous studies that both αβ and γ neurons were required for 3-h memory retrieval [25, 26], we found that the output from αβ neurons is also required for 3-h ARM, a component of 3-h memory (Fig 1F and 1H) whereas the output from γ neurons is not (Fig 1B and 1D), suggesting that γ neurons might exclusively mediate retrieval of the other component, the ASM. This was confirmed by the observation that blocking the output from γ neurons using VT44966-GAL4 impaired 3-h memory, instead of ARM, retrieval (S1 Fig). However, in contrast to the previous finding that α′β′ neurons are not required for 3-h memory retrieval [27], we found that blocking output from α′β′ neurons during retrieval using VT30604-GAL4 or VT57244-GAL4 impaired 3-h ARM (Fig 1J and 1L). This is a perplexing finding. Since 3-h ARM is a component of 3-h memory, any neural mechanism required for 3-h ARM should also be required for 3-h memory. This discrepancy prompted us to examine expression pattern of the GAL4 lines and eventually led to an explanation (see below).
We have used two independent GAL4 lines to reveal the functional role of αβ and α′β′ neurons in 3-h ARM retrieval. Next, we wonder whether there is an additive effect when blocking both MB subsets simultaneously. Although VT49246-GAL4 line has more restricted expression pattern than C739-GAL4 line, we only can use the double GAL4 line C739-GAL4; VT30604-GAL4 expressing in both αβ and α′β′ neurons (Fig 1M) for shibire manipulation due to the genetic feasibility (see the materials and methods). Indeed, we found an additive effect in which 3-h ARM retrieval was reduced further in C739-GAL4; VT30604-GAL4 > UAS-shits flies than in C739-GAL4 > UAS-shits or VT30604-GAL4 > UAS-shits flies (Fig 1N). This data indicated that outputs from αβ and α′β′ neurons additively contribute to 3-h ARM retrieval. All groups of flies showed no memory deficit at the permissive temperature and normal avoidance of odor or shock at the restrictive temperature (S2 Fig).
The radish gene encodes a protein that is required for ARM formation and is preferentially immunolabeled in the α, β, and γ lobes as well as the calyx and ellipsoid body [19]. A previous study showed that feeding rsh1 mutant flies with serotonin synthesis inhibitor had no effect on 3-h memory, suggesting that RADISH and serotonergic DPM–αβ neurons circuit are in the same pathway for ARM consolidation [21]. Also, our previous study showed an additive effect on ARM deficit when combining knockdown of octopamine synthesis gene in APL neurons with serotonin synthesis inhibitor or with rsh1 mutant background, suggesting that octopaminergic APL–α′β′ neurons and radish-dependent serotonergic DPM–αβ neurons circuits independently mediate ARM consolidation [22]. However, all the radish-related studies are mainly based on rsh1 mutant background, and the subset of MB neurons in which radish functions for ARM consolidation has not been identified. We used an inducible RNAi-mediated knockdown strategy to suppress radish expression in the adult stage. Flies with the tubulin promoter-driven temperature-sensitive GAL4 repressor GAL80 (tub-GAL80ts) were raised at 18°C and transferred to 30°C for 7 days after eclosion. Inducible RNAi-mediated knockdown of radish (see S3A and S3B Fig for validation by quantitative PCR) in αβ neurons using VT49246-GAL4, but not in α′β′ neurons using VT30604-GAL4, caused a significant deficit (Fig 2A), suggesting that RADISH mediates ARM consolidation in αβ neurons. Normal radish expression in γ neurons is not required for 3-h ARM (S3C Fig). We also used a combinatorial assay to confirm that the octopamine signaling for ARM consolidation in α′β′ neurons is independent of radish. In the rsh1 mutant background, RNAi-mediated knockdown of octβ2R in α′β′ neurons, but not in αβ neurons, further impaired ARM (Fig 2B). Taken together, these data showed that RADISH in αβ neurons and octopamine signaling in α′β′ neurons mediated ARM consolidation in parallel.
To further demonstrate the independency between two parallel neural pathways/circuits expressing ARM, we conducted an experiment in which gene for memory consolidation and neurotransmission for retrieval were manipulated in the same flies. We first confirmed that blocking neurotransmission during retrieval in either αβ neurons using VT49246-GAL4 or α′β′ neurons using VT30604-GAL4 impaired ARM expression (Fig 2C, left panel). After switching the genetic background to rsh1 to disrupt radish-dependent ARM consolidation, only neurotransmission blockade in α′β′ neurons, but not in αβ neurons, during retrieval caused further reduction of ARM expression (Fig 2C, right panel). All shibire-expressing flies in rsh1 background showed normal avoidance of odor or shock at the restrictive temperature (S3D Fig). These data indicate that the output from α′β′ neurons mediates radish-independent ARM retrieval while the output from αβ neurons, not surprisingly, mediates radish-dependent ARM retrieval.
MBON-β2β′2a and MBON-β′2mp neurons are two pairs of MB output neurons for αβ and α′β′ neurons, respectively [17, 28, 29] (see also Fig 3A and 3B). The dendrites of the MBON-β2β′2a neuron were marked by Dscam::GFP in the β lobe tips, whereas the dendrites of the MBON-β′2mp neuron were found only in the middle stratum of the β′ lobe (Fig 3C and 3D), albeit sparse DenMark-positive signals were additionally seen in the β′2a region for the MBON-β2β′2a neuron (S4A Fig)[17]. Blocking neurotransmission from MBON-β2β′2a or MBON-β′2mp neurons during the test, but not the first hour after training, impaired ARM expression (Fig 3E and 3F), indicating that the outputs from these neurons are required for ARM retrieval. All shibire-expressing flies showed normal avoidance of odor or shock at the restrictive temperature (S4B Fig). It has been shown that both MBON-β2β′2a and MBON-β′2mp neurons are vesicular glutamate transporter- (VGlut-) antibody immunopositive, which is indicative of glutamatergic neuron in Drosophila melanogaster [28,30]. To determine whether glutamatergic transmission mediates the retrieval, we knocked down VGlut expression in MBON-β2β′2a or MBON-β′2mp neurons by RNAi (see S5A Fig for validation of UAS-VGlutRNAi by quantitative PCR) and tested the flies for 3-h ARM. VT0765-GAL4 > UAS-VGlutRNAi and VT41043-GAL4 > UAS-VGlutRNAi flies showed 3-h ARM deficit (Fig 3G). Furthermore, knockdown of VGlut in both MBON-β2β′2a and MBON-β′2mp neurons showed an additive effect on ARM deficit (Fig 3H). Consistent with the additive contributions of αβ and α′β′ neurons to 3-h ARM retrieval, this additive effect suggests that glutamatergic MBON-β2β′2a and MBON-β′2mp neurons also additively contribute to ARM retrieval, although blocking output from the former alone was sufficient to abolish 3-h ARM. Since these experiments adopted chronic knockdown of VGlut that may provoke secondary effects, we introduced tub-GAL80ts for inducible VGlut knockdown in the adult stage (S5B Fig), confirming that losing VGlut itself in MBON-β2β′2a or MBON-β′2mp neurons impaired ARM. In summary, our data draw two sets of double-layered parallel circuits additively expressing 3-h ARM in the fruit fly.
Krashes et al. used C305a-GAL4 and C320-GAL4 lines to conclude that the output from α′β′ neurons was required for 3-h memory acquisition and consolidation but not for 3-h memory retrieval [27](see also S6D1–S6D2 and S6E1–S6E2 Fig). Since 3-h memory can be dissected into ASM and ARM, a neural mechanism required for 3-h ARM should intuitively be required for 3-h memory. However, the output from α′β′ neurons was shown to be required for the retrieval of 3-h ARM in our study (Fig 1J and 1L), arguing that the output from α′β′ neurons should be required for the retrieval of 3-h memory. In order to reconcile this conflict, we closely revisited the GAL4 expression pattern and found that although C305a-GAL4 expresses in both α′β′ap and α′β′m subsets, the marked GFP signal did not occupied the whole region of each stratum of α′/β′ lobes in magnified horizontal, sagittal, and frontal sections (S6B1–S6B4 Fig), especially the middle stratum. The other more restricted C320-GAL4 expresses mainly in α′β′ap neurons and has no noticeable GFP signal in the middle stratum, either (S6C1–S6C4 Fig). In contrast, VT30604-GAL4 and VT57244-GAL4 expression patterns occupied strongly and comprehensively all strata of the α′/β′ lobes (Fig 4B1–4B4 and 4C1–4C4), suggesting that these two VT lines express in most if not all α′β′ neurons, while C305a-GAL4 and C320-GAL4 express weakly in a subpopulation that has few α′β′m subset neurons. Consistently, cell-counting data showed that VT30604-GAL4 or VT57244-GAL4 expresses in about twice as many MB neurons as C305a-GAL4 or C320-GAL4 does (Table 1). This conclusion derived from imaging observation led to a speculation that using C305a-GAL4 or C320-GAL4 for shibire manipulation, which requires high enough expression level to perturb synaptic transmission [23], cannot reflect the full functional role of α′β′ neurons, instead a skewed role for α′β′ap subset. Hence, we sought specific GAL4 lines expressing in the subsets of α′β′ neurons to decisively address this issue. By visually screening the Vienna Tile (VT) library and Janelia collection, we identified the VT50658-GAL4 and VT37861-GAL4 lines for α′β′ap neurons (Fig 4D1–4D4 and 4E1–4E4) as well as the R42D07-GAL4 and R26E01-GAL4 lines for α′β′m neurons (Fig 4F1–4F4 and 4G1–4G4). Take counts of MB neurons labeled by these GAL4 lines into consideration (Table 1), we decided VT30604-GAL4, VT37861-GAL4, and R42D07-GAL4 as good single-transgene GAL4 lines to study α′β′, α′β′ap, and α′β′m neurons, respectively.
To clarify the functional roles of different subsets of MB α′β′ neurons in ARM, we used the GAL4 lines characterized above for shibire manipulation. Blocking the output from α′β′ap neurons using VT50658-GAL4 or VT37861-GAL4 during the first hour after training impaired subsequent ARM expression whereas the same manipulation during retrieval had no effect (Fig 5A and 5B), suggesting that output from the α′β′ap neurons is involved in ARM consolidation. This involvement in 3-h ARM consolidation, but not retrieval, is an analogy to the finding with the C305a-GAL4 and C320-GAL4 lines, in which output from GAL4-expressing neurons is required for 3-h memory consolidation but not retrieval [27] (see also S6 Fig). In contrast, blocking the output from α′β′m neurons using R42D07-GAL4 or R26E01-GAL4 during retrieval impaired ARM expression whereas the same manipulation during the first hour after training had no effect (Fig 5C and 5D), suggesting that the output from α′β′m neurons mediates 3-h ARM retrieval, despite the concern that the overall low scores make us hesitate to exclude the involvement in consolidation (left panels of Fig 5C and 5D). This role of α′β′m neurons is also supported by the dendritic distribution of the MBON-β′2mp neurons, which exclusively occupied the middle stratum of the β′ lobe (Fig 3D). Given that these flies showed normal avoidance of odor or shock at restrictive temperature (S7A1–S7A4 and S7B Fig), the data collectively indicate the functional heterogeneity of MB α′β′ neurons, where outputs from the α′β′ap and α′β′m neurons mediate the ARM consolidation and retrieval, respectively.
The key finding in our study is the identification of two parallel neural pathways that additively express 3-h aversive ARM through Drosophila MB αβ and α′β′ neurons. After training, RADISH in MB αβ neurons and octopamine signaling in α′β′ neurons independently consolidate ARM, which is additively retrieved by αβ–MBON-β2β′2a and α′β′m–MBON-β′2mp circuits for memory expression. Five lines of evidence support this scenario. First, the output from αβ or α′β′ neurons is required for ARM retrieval (Fig 1F, 1H, 1J and 1L), and the effect of blocking αβ output and that of blocking α′β′ output during retrieval are additive (Fig 1N). Second, knockdown of radish in αβ neurons, but not in α′β′ neurons, impaired ARM (Fig 2A), while knockdown of octβ2R in α′β′ neurons further impaired the residual ARM in rsh1 mutant flies (Fig 2B). Third, blocking output from α′β′ neurons, but not from αβ neurons, during retrieval further impaired the residual ARM in rsh1 mutant flies (Fig 2C). Forth, glutamatergic output from neurons downstream of the αβ or α′β′ neurons, i.e., MBON-β2β′2a or MBON-β′2mp neurons, is required for ARM retrieval, and the effects of knockdown of VGlut are additive (Fig 3). Finally, output from α′β′m neurons, but not α′β′ap neurons, is required for ARM retrieval, consistent with the dendritic distribution of MBON-β′2mp neurons (Figs 3D and 5).
The parallel pathways for 3-h ARM expression were spatially defined by the requirements of neurotransmission from two sets of circuits during retrieval, the αβ–MBON-β2β′2a neurons and the α′β′m–MBON-β′2mp neurons. In addition, blocking neurotransmission from αβ or α′β′ neurons during retrieval reduced ARM expression by about 50% (Fig 1H, 1J and 1L) whereas simultaneous blockade produced an additive effect that completely abolished ARM expression (Fig 1N). Similar additive effects were repeatedly observed in experiments that utilize manipulations in both pathways: an rsh1 mutant background plus octβ2R RNAi knockdown (Fig 2B) or plus retrieval blockade in α′β′ neurons (Fig 2C), and knockdown of VGlut in MBON-β2β′2a plus MBON-β′2mp neurons (Fig 3H). Thus, total four lines of evidence support the additive expression of 3-h ARM.
The parallel pathways for 3-h ARM expression shown here differ from the degenerate parallel pathways for the stomatogastric ganglion of the crab or CO2 avoidance in the fly [31, 32], as the latter enable mechanisms by which the network output can be switched between states. In our study, the two parallel neural pathways additively contribute to the expression of 3-h ARM. The nature of the ARM parallel pathways may be similar to that for cold avoidance behavior in the fly, where parallel pathways in the β′ and β circuits additively contribute but only the β circuit allows age-dependent alterations for potential benefits against aging [29]. Considering the robustness of ARM through the course of senescence [5], it’s unlikely to be age-dependent alterations in ARM system.
In studies of Drosophila neurobiology, C305a-GAL4 is a common GAL4 line for α′β′ neurons [16, 27]. Here, by examining three different zoom-in sections of the MB lobes and counting the cells (S6B1–S6B4 Fig and Table 1), we extensively characterized the following GAL4 lines expressing in α′β′ neurons: VT30604-GAL4 and VT57244-GAL4, which cover most α′β′ap and α′β′m neurons; VT37861-GAL4 and VT50658-GAL4, which cover α′β′ap neurons; and R42D07-GAL4 and R26E01-GAL4, which cover most α′β′m neurons. In contrast, C305a-GAL4 sporadically expresses in about half as many MB neurons as VT30604-GAL4 or VT57244-GAL4 does (Table 1). Although covering both subsets of α′β′ neurons, the expression pattern of C305a-GAL4 in α′β′m neurons is too few and/or weak to lead to a perturbation of synaptic transmission. This is shown by the data that retrieval of 3-h ARM was disrupted by shibire manipulation using all-α′β′ neurons driver (Fig 1J and 1L) or α′β′m-specific driver (right panels of Fig 5C and 5D), but neither α′β′ap-specific driver (right panels of Fig 5A and 5B) nor C305a-GAL4 for 3-h memory ([27], see also S6D2 Fig). Please note that our GFP signals were acquired from flies carrying two copies of 5XUAS-mCD8::GFP reporter and without any immunostaining-mediated amplification. With the assistance of immunostaining and/or advanced reporter such as increasing copy number of UAS or incorporating a small intron to boost expression [33], some studies have shown appreciable GFP signal in most α′β′ neurons [16, 27, 34]. Given that shibire-mediated neurotransmission blockade and RNAi-mediated knockdown require high enough expression level, the imaging method we adopted in this study can faithfully reflect the regions that were effectively manipulated in our, as well as Krashes et al’s [27], behavioral assays. Regarding the pervasive use of C305a-GAL4 for shibire or RNAi manipulation, some functional studies of α′β′ neurons might need to be carefully revisited. Here, we showed, by close examination and cell counting, VT30604-GAL4, VT37861-GAL4, and R42D07-GAL4 as useful GAL4 lines to study α′β′, α′β′ap, and α′β′m neurons, respectively, especially when split-GAL4 lines that span the second and third chromosomes are not genetically feasible [17].
ARM was thought to be diminished in radish mutant flies, in which a truncated RADISH is expressed [19]. It’s noteworthy that radish mutants still show a residual 3-h ARM with a PI of roughly 10, which is equal to the 3-h ARM score in wild-type flies fed with an inhibitor of serotonin synthesis to hinder the serotonergic DPM neurotransmission [19, 21, 22]. Interestingly, feeding radish mutant flies with the drug didn’t make the 3-h memory score worse [21], which has already implied that RADISH mediates the consolidation of ARM in the serotonergic DPM-αβ neurons circuit. Indeed, in this study we took advantage of RNAi-mediated knockdown to identify αβ neurons with RADISH-mediated ARM consolidation (Figs 2A and S3C). However, only the output from αβs neurons among three subsets of αβ neurons is required for aversive memory retrieval [35]. Whether the αβs neurons are the only aversive ARM substrate of RADISH remains to be identified.
APL and DPM neurons are two pairs of modulatory neurons broadly innervating the ipsilateral MB, although the DPM neuron’s fiber is lacking in the posterior part of pedunculus and the calyx [15, 20, 36, 37]. Broad, extensive fiber and non-spiking feature [38] allow these two pairs of neurons to have multiple functional roles through different types of neurotransmission [20–22, 36, 37, 39–41]. The APL neuron has been shown to receive odor information from the MB neurons and provide GABAergic feedback inhibition as the Drosophila equivalent of a group of the honeybee GABAergic feedback neurons [14, 42]. This feedback inhibition has been proposed to maintain sparse, decorrelated odor coding by suppressing the neuronal activity of MB neurons [14], which can be somewhat linked to the mutual suppression relation with conditioned odor and the facilitation of reversal learning [36, 43]. Interestingly, Pitman et al. proposed that the feedback inhibition from APL neurons sustains the labile appetitive ASM based on shibire manipulation [44]. Since shibire manipulation can impact small vesicle release, and APL neurons have been demonstrated to co-release at least GABA and octopamine [22, 36], it might worth conducting GABA-specific manipulation in APL neurons to confirm the role in appetitive ASM. For aversive olfactory memory, acute RNAi-mediated knockdown of Gad1 in APL neurons had no effect on 3-h memory [22]. Instead, the octopamine synthesis enzyme mutant, TβhnM18, knockdown of Tβh in APL neurons, the octopamine receptor mutant, PBac{WH}octβ2Rf05679, and knockdown of octβ2R in α′β′ neurons all phenocopied the 3-h ARM impairment caused by shibire-mediated neurotransmission blockade in APL neurons [22] (see also Fig 2B). Together with the serotonergic DPM–αβ neurons circuit [21], we favor a model that two sets of triple-layered parallel circuits, octopaminergic APL–α′β′–MBON-β′2mp and serotonergic DPM–αβ–MBON-β2β′2a, additively contribute to 3-h aversive ARM.
Although our data showed that 3-h ARM consolidation requires recurrent output from α′β′ap neurons but not from α′β′m neurons (Fig 5), RNAi-mediated knockdown of octβ2R in α′β′ap or α′β′m neurons impaired ARM (Fig 6), suggesting that Octβ2R functions for normal ARM expression in the entire population of α′β′ neurons. On the other hand, neuronal activity during memory consolidation is naturally more quiescent than that during memory retrieval, and the shibire-mediated neurotransmission blockade requires an exhaustion of already-docked vesicles. Together with the unfavorable performance for experiments blocking the output from α′β′m neurons during consolidation (left panels of Fig 5C and 5D), we cannot exclude the possibility that output from α′β′m neurons is also required for ARM during consolidation. Alternatively, octopamine signaling may also be involved in ARM retrieval.
Drosophila melanogaster were raised on standard cornmeal food at 25°C and 70% relative humidity on a 12 h:12 h light:dark cycle. The “Cantonized” w1118 w(CS10) was used as the wild-type control. The UAS-shits line used in this study has multiple insertions on the third chromosome. UAS-Dscam[17.1]::GFP flies were obtained from Tzumin Lee. elav-GAL4;+;tub-GAL80ts flies were obtained from Hsueh-Cheng Chiang. The UAS-DenMark flies have been described [45]. The C305a-GAL4 and C320-GAL4 have also been described [16, 27]. The 1471-GAL4, C739-GAL4, R42D07-GAL4, R26E01-GAL4, and tub-GAL80ts were obtained from the Bloomington Stock Center. VT49246-GAL4, VT37861-GAL4, VT50658-GAL4, VT41043-GAL4, and VT0765-GAL4 were obtained from the Vienna Drosophila Resource Center, Vienna Tile (VT). UAS-radishRNAi(v39931) and UAS-VGlutRNAi(v104324) were obtained from the Vienna Drosophila RNAi Center; VT30604-GAL4, VT57244-GAL4, and UAS-octβ2RRNAi(v104524) have been described [22].
Fly brains were counterstained with the mouse 4F3 anti-discs large (DLG) monoclonal antibody to label all neuronal synapses. The brains were dissected in isotonic PBS and immediately transferred to 4% paraformaldehyde in PBS on ice for a 20-min fixation period. Fixed brain samples were incubated in PBS containing 2% Triton X-100 and 10% normal goat serum (NGS) for 2 h. During the 2-h penetration and blocking period, the brain samples were also subjected to a degassing procedure. Thereafter the brain samples were incubated in a dilution buffer (PBS containing 0.25% Triton X-100, 1% NGS) containing 1:10 mouse anti-DLG monoclonal antibody (Developmental Studies Hybridoma Bank, University of Iowa) at 25°C for one day. After washing in PBS-T three times, the samples were incubated in 1:200 biotinylated goat anti-mouse IgG (Molecular Probes) at 25°C for one day. Next, the brain samples were washed and incubated in 1:500 Alexa Fluor 635 streptavidin or 1:500 Alexa Fluor 488 streptavidin (Molecular Probes) at 25°C overnight. After extensive washing, the brain samples were cleared and mounted in FocusClear (CelExplorer) for confocal imaging.
Sample brains were imaged under a Zeiss LSM 700 confocal microscope with either a 40× C-Apochromat water-immersion objective lens for whole-brain images (N.A. value, 1.2; working distance, 220 μm) or a 63× glycerine-immersion objective lens for horizontal, sagittal, and frontal cross sections (N.A. value, 1.4; working distance, 170 μm). To overcome the limited field of view, some samples were imaged twice, one for each hemisphere, with overlap in between. We then stitched the two parallel image stacks into a single dataset on-line with ZEN software, using the overlapping region to align the two stacks.
Groups of approximately 100 flies were exposed first to one odor (the conditioned stimulus, CS+; 3-octanol or 4-methyl-cyclohexanol) paired with 12 × 1.5-s pulses of 75-V DC electric shock presented at 5-s interpulse intervals. This was followed by the presentation of a second odor (CS–; 4-methyl-cyclohexanol or 3-octanol) without electric shock. In the testing phase, the flies were presented with a choice between CS+ and CS– odors in a T-maze for 2 min. At the end of this 2-min period, the flies were trapped in each T-maze arm were anesthetized and counted. From the distribution of flies between the 2 arms, the performance index (PI) was calculated as the number of flies avoiding the shocked odor (CS+) minus the number avoiding the non-shocked odor (CS–), divided by the total number of flies and multiplied by 100. If the flies did not learn, they were distributed equally between the 2 arms; hence, the calculated PI was 0. However, if all flies avoided the shock-paired odor and were distributed 0:100 between the CS+ and CS– in the T-maze, the PI was 100. To assess learning, performance was measured immediately after training. To evaluate intermediate-term memory, testing was performed 3 h after training. ARM was assayed as 3-h memory, after a 2-min cold shock was presented at 2 h post-training (1 h before testing) by placing a plastic vial containing trained flies in ice water. A brief cold shock, which completely erases short-term memory and labile ASM, leaves only ARM. For the shits experiments, flies were kept at 23°C throughout development. After eclosion, flies were kept at 23°C prior to shifting to 31°C, as indicated by the schematic diagrams above each behavioral graph in the figures. For the adult-stage-specific RNAi-mediated knockdown of radish with tub-GAL80ts, flies were kept at 18°C until eclosion and then shifted to 30°C for 7 days before training. The 3-h ARM assay was also performed at 30°C. Control flies were kept at 18°C throughout the experiment.
All raw data were analyzed parametrically with Prism 5.0 software (GraphPad). Because of the nature of their mathematical derivation, performance indices were distributed normally. Hence, the data with more than two groups were evaluated by one-way analysis of variance (ANOVA) and Tukey’s multiple comparisons tests. Data with only two groups were evaluated by paired t-test. Definition of statistical significant difference was set as P < 0.05. All data were presented as mean ± SEM.
Equal numbers of males and females for each GAL4 line were used for analysis. GAL4 expression patterns were reported by two copies of 5XUAS-mCD8::GFP reporter, and the MB structure was counterstained with DLG antibody. Each MB neuron labeled by GFP was manually marked with a landmark sphere in the Amira software, and the total number of landmark spheres in each hemisphere was calculated.
All flies were shifted to the restrictive temperature of 31°C for 15 min before the odor- and shock avoidance tests. For odor avoidance, groups of roughly 100 untrained flies received a 2-min test trial in the T-maze. Different groups were given a choice between either OCT or MCH versus “fresh” room air. The odor avoidance index was calculated as the number of flies in the fresh room air tube minus the number in the odor tube, divided by the total number of flies and multiplied by 100. For shock-avoidance, groups of approximately 100 untrained flies received a 2-min test trial in the T-maze. Each arm of the T-maze contained an electric shock grid, however, and different groups of flies were given a choice between shock and no shock. The shock avoidance index was calculated as the number of flies in the non-shocked grid minus the number in the shocked grid, divided by the total number of flies, and multiplied by 100.
The effectiveness of the UAS-radishRNAi(v39931) and UAS-VGlutRNAi(v104324) lines were verified with qPCR. Flies for qPCR were generated by crossing elav-GAL4 or elav-GAL4; +; tub-Gal80ts virgin flies with either wild type males or UAS-VGlutRNAi males or UAS-radishRNAi males. RNA in the isolated heads of adult flies or in whole 3rd instar larvae was extracted with TRIZOL Reagent (Invitrogen, Life Technologies, USA). The extracted RNA was used to synthesize first-strand cDNA with High-Capacity cDNA Reverse Transcription Kits (Applied Biosystems, USA). RNA expression levels were quantified by qPCR (StepOnePlus™ System, Applied Biosystems). qPCR and quantitative measurements were performed with the SYBR-Green PCR-Master Mix (Applied Biosystems).
|
10.1371/journal.pgen.1004980 | Antagonistic Roles for KNOX1 and KNOX2 Genes in Patterning the Land Plant Body Plan Following an Ancient Gene Duplication | Neofunctionalization following gene duplication is thought to be one of the key drivers in generating evolutionary novelty. A gene duplication in a common ancestor of land plants produced two classes of KNOTTED-like TALE homeobox genes, class I (KNOX1) and class II (KNOX2). KNOX1 genes are linked to tissue proliferation and maintenance of meristematic potentials of flowering plant and moss sporophytes, and modulation of KNOX1 activity is implicated in contributing to leaf shape diversity of flowering plants. While KNOX2 function has been shown to repress the gametophytic (haploid) developmental program during moss sporophyte (diploid) development, little is known about KNOX2 function in flowering plants, hindering syntheses regarding the relationship between two classes of KNOX genes in the context of land plant evolution. Arabidopsis plants harboring loss-of-function KNOX2 alleles exhibit impaired differentiation of all aerial organs and have highly complex leaves, phenocopying gain-of-function KNOX1 alleles. Conversely, gain-of-function KNOX2 alleles in conjunction with a presumptive heterodimeric BELL TALE homeobox partner suppressed SAM activity in Arabidopsis and reduced leaf complexity in the Arabidopsis relative Cardamine hirsuta, reminiscent of loss-of-function KNOX1 alleles. Little evidence was found indicative of epistasis or mutual repression between KNOX1 and KNOX2 genes. KNOX proteins heterodimerize with BELL TALE homeobox proteins to form functional complexes, and contrary to earlier reports based on in vitro and heterologous expression, we find high selectivity between KNOX and BELL partners in vivo. Thus, KNOX2 genes confer opposing activities rather than redundant roles with KNOX1 genes, and together they act to direct the development of all above-ground organs of the Arabidopsis sporophyte. We infer that following the KNOX1/KNOX2 gene duplication in an ancestor of land plants, neofunctionalization led to evolution of antagonistic biochemical activity thereby facilitating the evolution of more complex sporophyte transcriptional networks, providing plasticity for the morphological evolution of land plant body plans.
| Eukaryotes alternate between haploid (1n) and diploid (2n) stages during their life cycles, and often seen are remarkable differences in morphology and physiology between them. Land plants are multicellular in both generations, in contrast to their presumed ancestral green algae that develop multicellularity only in the haploid stage. TALE class homeodomain transcriptional factors play a key role in the activation of diploid development in diverse lineages of eukaryotes. A gene duplication event within this family in an ancestor of land plants had profound implications for land plant evolution. We show that the two subclasses resulting from the gene duplication event act to pattern, in a complementary manner, most above ground organs of the diploid stage of the flowering plant Arabidopsis. Their opposing activities sculpt the shape of leaves from entire to pinnate and control the architecture of the plant body, and thus providing plasticity for evolutionary tinkering. These results form a foundation for understanding how these genes have been co-opted from an ancestral role of regulating diploid gene expression in a zygote to directing sporophyte land plant body architecture and provide insight into the evolution of various forms of life cycles.
| Gene duplication is thought to be one of the key drivers in generating evolutionary novelty. Following gene duplication, paralogs can undergo a process of neofunctionalization, supplying a genetic basis for morphological novelty [1,2,3]. Transcription factors can undergo neofunctionalization via either a change in expression pattern or an alteration in functionality, e.g. the derivation of a repressor or inhibitor from an ancestral activator, or vice versa (e.g. [4]). Three amino acid loop extension (TALE) homeodomain transcriptional factors, characterized by having a homeodomain that has three extra amino acids between helices 1 and 2, are found in all eukaryotic lineages [5,6,7]. Plant TALE homeobox genes are classified into two subfamilies, KNOTTED-like homeobox (KNOX) and BELL-like (BELL) [8]. Whilst Chlorophyte algal KNOX genes are of a single class, a gene duplication in a common ancestor of land plants, produced two classes of KNOX genes, class I (KNOX1) and class II (KNOX2) [8,9] (Fig. 1A). KNOX genes of flowering plants have been studied for over two decades, however, the functional consequences of the KNOX gene duplication have been largely unexplored.
The first identified plant homeobox gene was Knotted1, a KNOX1 gene of maize [10]. Since, KNOX1 genes have been characterized in numerous flowering plants with a conspicuous loss-of-function phenotype being a failure in shoot apical meristem (SAM) maintenance. KNOX1 activity is also involved in maintenance of meristematic activity during leaf development, with prolonged activity in leaf margins observed in species with complex leaves and gain-of-function alleles result in more complex leaves. Thus, KNOX1 genes play a critical role in maintaining meristematic properties of cells in flowering plant sporophytes, the diploid generation of the land plant life cycle (reviewed in [11,12,13]). The KNOX1 genes of the moss Physcomitrella patens are only expressed in the sporophyte and mutants have decreased sporophyte growth, suggesting that KNOX1 genes have a conserved role in tissue proliferation during sporophyte development throughout land plants [12,13,14]. There is no evidence indicative of KNOX1 function in the gametophyte (haploid) generation in any characterized species, including the indeterminate meristems of the moss gametophyte, suggesting the role of KNOX1 is restricted to the diploid, sporophyte generation [14].
A functional distinction between KNOX1 and KNOX2 genes has been postulated from studies based on gene expression patterns in flowering plants. Northern blot analyses in maize demonstrated that KNOX1 gene expression is confined to less differentiated tissues whereas KNOX2 genes are broadly expressed in differentiating tissues and mature organs [9]. Similar broad expression profiles of KNOX2 genes have been reported in Arabidopsis [15,16] and tomato [17]. Characterization of spatial expression patterns in Arabidopsis revealed that KNOX2 genes have both overlapping and distinct expression patterns and that they are expressed in most tissues except for meristematic regions [15,18,19,20,21]. Despite several reports of expression patterns, comparatively little is known about KNOX2 gene function in flowering plants. One of the four Arabidopsis KNOX2 paralogs, KNAT7, is involved in secondary cell wall biosynthesis [18,19,21], and another, KNAT3, is reported to modulate ABA responses [22]. While these findings are consistent with the reported expression patterns, there exists a gap between broad expression patterns and known KNOX2 functions. For instance, unlike KNOX1 genes, which are important regulators of growth and development, it is not clear whether or not KNOX2 genes are involved in morphogenesis in flowering plants. These questions have gone unanswered owing to the paucity of functional studies on KNOX2 genes due to extensive genetic redundancy as noted by Truernit et al. [20].
From a wider perspective, a possible ancestral function of TALE homeodomain proteins is the regulation of diploid gene expression upon fusion of gametes, as is observed in the Chlorophyte alga Chlamydomonas reinhardtii and several fungi [23,24,25]. In C. reinhardtii the plus gamete expresses a BELL protein while the minus gamete expresses a KNOX protein; upon gamete fusion the KNOX and BELL proteins heterodimerize and regulate zygotic gene expression [25]. In the moss P. patens, KNOX2 genes are expressed in the egg cells and the sporophyte. Eliminating KNOX2 activity results in apospory, the development of a haploid body plan during the diploid generation, suggesting KNOX2 genes regulate the gametophyte-to-sporophyte morphological transition, a reflection of the hypothesized ancestral TALE homeodomain gene function [26]. Thus, both KNOX1 and KNOX2 mutant phenotypes in land plants are consistent with the hypothesis that ancestral function of KNOX genes was to regulate diploid gene expression. However, the seemingly different roles of KNOX1 and KNOX2 genes indicate functional diversification among land plant KNOX genes.
To gain insight into developmental roles for KNOX2 genes in flowering plants and the genetic relationship between KNOX1 and KNOX2 classes, we undertook a genetic study of KNOX2 genes in Arabidopsis thaliana, a species in which KNOX1 gene function is well characterized. We discuss the implication of our findings on the impact of the gene duplication producing KNOX1 and KNOX2 paralogs in the course of land plant evolution.
Four Arabidopsis genes encode KNOX2 proteins. As phylogenetic analyses place KNAT7 in a clade sister to the remaining Arabidopsis KNOX2 genes [9], we focused on KNAT3 (AT5G25220), KNAT4 (AT5G11060), and KNAT5 (AT4G32040) genes in this study (Fig. 1B, S1 and S2 Figs.). In contrast to KNOX1 genes, which are expressed primarily in meristematic tissues, KNOX2 gene expression occurs in differentiating organs suggesting distinct and perhaps complementary functions [15,16,20] (Fig. 1C, S3 and S4 Figs.).
KNOX2 mutant phenotypes were characterized using null alleles (S5 Fig.). As reported previously [20], single mutants lack conspicuous aberrant phenotypes. Amongst double mutants, knat3 knat5 seedlings are distinguishable from wild type by a longer petiole and narrower lamina of cotyledons, and more deeply serrated leaf margins (Fig. 2A-B). Venation pattern is also affected in knat3 knat5 cotyledons (S6 Fig.). knat3 knat4 plants also have serrated leaves (Fig. 2I) and are sporophytically female sterile with abnormal integument development. While knat3 knat4/+ knat5 plants are also female sterile, knat3/+ knat4 knat5 plants are phenotypically wild type and produce viable seeds, facilitating characterization of segregating triple mutant plants.
Selfed knat3/+ knat4 knat5 plants segregated small, dark-green plants with deeply lobed leaves (Fig. 2C-G and S14F Fig.), a phenotype reminiscent of gain-of-function KNOX1 alleles [12,13]. PCR-based genotyping indicated these plants were triple-mutant homozygotes (knat345). Since only a single mutant allele was available for each gene, we designed an artificial miRNA (amiRNA, [27]) targeting only KNAT4, amiR159-KNAT4 (S7A Fig.), and generated knat3 knat5 plants constitutively expressing amiR159-KNAT4 under the control of the Cauliflower Mosaic Virus 35S promoter (pro35S). pro35S:amiR159-KNAT4 knat3 knat5 lines closely resembled the identified knat345 plants (S8C-D Fig.). Another amiRNA, amiR159-KNAT345–1, was designed to target KNAT3, KNAT4, and KNAT5 (S7B Fig.). pro35S:amiR159-KNAT345–1 plants also show a deeply serrated leaf phenotype (S8E and S14H Figs.). We thus conclude that this is the triple mutant phenotype. Consistent with functional redundancy among these genes, dosage-dependent enhancement of the leaf serration phenotype was observed (Fig. 2H-L). Likewise, venation pattern in cotyledons is more severely affected in knat345 plants (S6 Fig.). Floral organs homologous with leaves are also affected. Sepals and petals are narrower and partially dissected in the knat345 mutant, and integument development is defective as seen in knat3 knat4 plants (Fig. 2M-N and S9 Fig.). Ectopic formation of tracheary elements is observed in knat345 embryo sacs (Fig. 2R). Although these genes are expressed in roots [15,16,20], the morphology of primary roots in knat345 plants appeared normal (S10 Fig.).
A proKNAT5:KNAT5-GUS translational fusion line was generated to monitor expression patterns (Fig. 2S-W and S11 Fig.). In line with the mutant phenotypes, GUS activity was observed in developing leaves but excluded from the shoot apical meristem (SAM) (Fig. 2S-V and S11C Fig.). During early stages of leaf development, GUS activity was not detected in youngest leaf primordia but was observed in older leaf primordia (Fig. 2U). Reduced signal levels were observed in older leaves (S11B Fig.). Prolonged incubation detected GUS signal along cotyledon and leaf veins and in ovules (Fig. 2T and S11H-I Fig.). proKNAT5:KNAT5-GUS expression is nuclear in trichomes, supporting a role for KNAT5 in transcriptional regulation (S11 Fig.). A transcriptional fusion line, proKNAT4:GUS, was generated to examine KNAT4 expression patterns. Ten independent T1 plants were examined, all of which exhibited KNAT4 promoter activity in leaves but not in the SAM (Fig. 2X). A similar expression pattern has been described for KNAT3 using either a GUS reporter line or RNA in situ hybridization [15]. Exclusion of KNOX2 expression from the SAM is also supported by cell-type specific expression analyses of the inflorescence SAM (S3 Fig.).
KNOX and BELL heterodimerization plays a pivotal role in regulating their activities as transcription factors [13]. We speculated that the lack of BELL partners may explain why no conspicuous phenotype has been described to date upon ectopic expression of KNOX2 genes [28] (S12 Fig.). The founding BELL gene, BELL1 (BEL1), and closely related paralogs, SAWTOOTH1 (SAW1) and SAW2, represent candidates for KNOX2 partners since loss-of-function phenotypes in ovules and leaf margins resemble those of KNOX2 mutants [29,30,31]. Physical interactions have been previously proposed between these BELL and KNOX2 proteins [22,29,32]. SAW1 and SAW2 are expressed in leaves but not in meristems [29] (S3 and S4 Figs.). Thus, we co-expressed SAW2 and KNAT3 throughout the SAM by trans-activating SAW2 under the control of SHOOT MERISTEMLESS (STM) regulatory sequences (proSTM>>SAW2; >> denotes the use of transactivation system hereafter) in pro35S:KNAT3 plants [33]. pro35S:KNAT3 proSTM>>SAW2 plants lack an embryonic SAM and resemble loss-of-function stm or stm knat6 mutant plants [34,35,36] (Fig. 3A-C, 3E). Combined expression of KNAT5 and SAW2 in the proSTM region resulted in a similar phenotype (Fig. 3D), confirming that the presence of both SAW2 and KNOX2 proteins simultaneously accounts for the phenotype. Collectively, these data indicate that concurrent expression, and by proxy, heterodimerization with BELL proteins, is important for KNOX2 function and that KNOX2 activity may thus be constrained by limited access to corresponding BELL partners.
A mutation in KNAT3 suppresses the gain-of-function phenotype caused by ectopic expression of another BELL gene, BLH1, suggesting that BLH1 is likely a functional partner for KNOX2 proteins [37]. This prompted us to examine genetic interactions between BLH1 and BEL1-related BELL genes, and we found that bel1 blh1 double mutants show color changes in unfertilized gynoecia as seen in knat3 knat4 and knat345 plants (Fig. 3G-J and S9 Fig.). Thus, BEL1 and BLH1 play a redundant role in gynoecium development and perhaps act in association with KNOX2 genes. More comprehensive genetic analyses as well as expression analyses are required to assign specific roles to functionally redundant BELL genes.
To further dissect BELL-KNOX interactions, plants expressing BELL and/or KNOX genes in the proSTM region were characterized. Among BELL proteins, PENNYWISE (PNY) and POUND-FOOLISH (PNF) are expressed in the SAM and act in conjunction with KNOX1 proteins to promote SAM activity [12,13]. As expected, plants expressing KNOX1 genes (STM or KNAT2) or PNY in the STM domain appeared wild type (S13B-D Fig.). In contrast, proSTM>>SAW2 plants displayed abnormal floral morphologies, such as fused sepals, reduced petals, and misshapen fruits (S13E, H Fig.), phenotypes often associated with reduced KNOX1 activity, e.g. weak stm mutants [38]. Flower development was not impacted in proSTM>>KNAT5 plants, but fused sepals are also observed in strong pro35S:KNAT3 lines (S12F Fig.). Concomitant expression of KNOX2 with PNY or KNOX1 with SAW2 did not enhance the KNOX2 or SAW2 overexpression phenotypes. We therefore conclude KNOX2 shows selectivity for BELL proteins in vivo.
Loss-of-function and gain-of-function KNOX2 phenotypes are reminiscent of gain-of-function and loss-of-function KNOX1 phenotypes, respectively [12,13]. To characterize the relationship between the two gene classes, loss-of-function alleles for KNOX1 and KNOX2 were combined. Plants constitutively expressing an amiRNA targeting KNAT3, KNAT4, and KNAT5, pro35S:amiR159-KNAT345–1, in KNOX1 loss-of-function (stm or bp knat2 knat6) backgrounds were examined. Neither the meristem failure of stm mutants nor the KNOX2 loss-of-function mutant leaf phenotype was suppressed in these plants (Fig. 3F and Fig. 4A-B). Similarly, neither knat2 knat3 knat5 knat6 nor bp knat345 showed significant suppression of the KNOX2 loss-of-function mutant leaf phenotype and the bp inflorescence phenotype (Fig. 4F-J). Thus, loss-of-function phenotypes of KNOX1 and KNOX2 mutants are not due to ectopic activation of KNOX2 and KNOX1, respectively. Furthermore, BP, STM, and KNAT2 expression was not altered in knat3 knat5 plants (Fig. 4K-P), arguing against mutual repression between KNOX1 and KNOX2 genes.
Deeply lobed leaves, a phenotype characteristic of gain-of-function KNOX1 alleles, occur in Arabidopsis plants where STM is driven by the leaf specific promoter, proBLS, proBLS:STM ([39]; S14B, C Fig.). These were crossed with loss-of-function KNOX2 plants (pro35S:amiR159-KNAT345–1) to generate plants with ectopic KNOX1 and reduced KNOX2 activities in the leaves. Compared to the parental lines, F1 plants harboring both transgenes displayed more extreme leaf margin elaboration (Fig. 4C-E). The additive effects, rather than epistatic interactions, suggest it is unlikely that the two subclasses negatively regulate one another.
An attractive hypothesis for the antagonism between KNOX1 and KNOX2 is that they regulate shared downstream events in an opposite manner. The complex leaf of gain-of-function KNOX1 alelles is suppressed by reduction in CUP SHAPED COTYLEDON (CUC) transcription factor activity [40] (S14C Fig.). Two CUC genes are targeted by the miR164 family of miRNAs, and expression of miR164b in young leaves using regulatory sequences of the FILAMENTOUS FLOWER (FIL) gene (designated as proFIL), proFIL:miR164b, flattens the leaf margin in wild-type plants (S14D-E Fig.). Thus, the miRNA-mediated CUC regulation plays a key role in leaf margin elaboration [41]. Introduction of proFIL:miR164b also suppressed the leaf dissection phenotype in a knat345 mutant background (S14F-G Fig.). Among miR164 targets, CUC2 plays a major role in leaf serration development [41]. We find leaf serration is largely suppressed in the cuc2 knat345 and pro35S:amiR159-KNAT345–1 cuc2 backgrounds (Fig. 5A-C and S14H-I Fig.). In addition, constitutive expression of KNOX2 (pro35S:KNAT3) can partially suppress the proBLS:STM leaf phenotype (Fig. 5D-F). Thus, a common developmental program mediates both gain-of-function KNOX1 and loss-of-function KNOX2 leaf phenotypes.
As observed in proBLS:STM plants, elevated levels of KNOX1 activity are often associated with increased leaf complexity (reviewed in [11]). In Cardamine hirsuta, a close relative of Arabidopsis, dissected leaf development requires KNOX1 expression in leaves, and additional KNOX1 expression leads to ectopic leaflet initiation [42]. We investigated the outcome of reduction in the level of KNOX2 activity in this species. In Cardamine, leaf shape exhibits heteroblasty with leaflet number increasing in later produced leaves. Although leaflet number can vary for a particular leaf position, the first and second leaves always consist of a single, undivided, lamina, and the third leaf typically consisting of three leaflets (S15A Fig.). An amiRNA, amiR159-KNAT345–2, was designed to target three Cardamine genes homologous to Arabidopsis KNAT3, KNAT4, and KNAT5. Constitutive amiR159-KNAT345–2 expression (pro35S:amiR159-KNAT345–2; S7C Fig.) results in plants with an extra lateral leaflet on the second leaf, observed in approximately 15% of individuals (27 of 188 plants derived from 6 independent lines), indicating KNOX2 activity influences complexity of dissected leaves in Cardamine (Fig. 5G-H). Furthermore, gain-of-function KNOX2 alleles (pro35S:KNAT3) simplify leaf shape, a phenotype particularly obvious in third leaves, which are undivided in strong lines (Fig. 5I-J and S15B Fig.). Thus, reduction or increase in KNOX2 activity leads to increase or decrease in leaf complexity, respectively, in Cardamine (Fig. 5 and S15 Fig.). This observation and the deduced KNOX1/KNOX2 antagonism are in consistent with the results in Arabidopsis.
Arabidopsis KNOX2 genes act redundantly to promote differentiation of all aerial organs in a manner broadly antagonistic to the action of KNOX1 genes. Loss-of-function KNOX2 alleles exhibit phenotypes with attributes of those of gain-of-function KNOX1 alleles, and vice versa, both in the maintenance of the shoot apical meristem and in the development of leaf complexity. In both contexts, KNOX2 functions to suppress meristematic capability, while KNOX1 promotes or maintains it. Our observations suggest that following the gene duplication giving rise to the KNOX1 and KNOX2 paralogs in an ancestor of land plants, neofunctionalization led to evolution of antagonistic biochemical activity thereby facilitating morphological evolution. Given the highly conserved nature of KNOX1 and KNOX2 genes in land plants, the antagonistic relationship may be a general phenomenon of diverse species.
Three Arabidopsis KNOX2 genes, KNAT3, KNAT4, and KNAT5, act redundantly in regulating plant development. Distinct phenotypes of double mutant combinations, however, indicate various degrees of contributions among three genes. For instance, knat3 knat4 and knat3 knat5 plants have more deeply serrated leaves, whereas knat4 knat5 plants appear phenotypically wild type. Distinctive expression patterns may explain different phenotypic consequences in mutants [20], or alternatively, the potency of three KNOX2 proteins may vary owing to structural differences, and the different relative contributions of the three genes to leaf development can be seen as a process of subfunctionalization. Although Arabidopsis KNOX2 genes are expressed in the root, no overt phenotype was recognized in triple mutant roots, perhaps due to genetic redundancy with the fourth KNOX2 gene, KNAT7. In addition to their developmental roles, KNOX2 genes may play an undetected physiological role as they are expressed in senescing leaves (S4 Fig.) and have been reported to have a role involved in seed germination and early seedling development through modulating ABA responses [22]. Characterization of the quadruple mutants and physiological experiments may illuminate additional cryptic mutant phenotypes of KNOX2 genes.
Expression of KNOX1 genes in leaves is correlated with increased leaf complexity and has been hypothesized to be influential in the evolution of leaf shape [42,43,44]. Given that seed plants leaves evolved from ancestral shoot systems, the ancestral seed plant leaf was likely complex, but fossil evidence and phylognetic analyses suggest that the ancestral angiosperm leaf may have been simple [45]. Regardless of the ancestral state, transitions from simple to more complex and vice versa have occurred repeatedly during angiosperm evolution [43,44,46]. In angiosperms, increase in leaf complexity is associated with increased KNOX1 activity while loss of KNOX1 activity in leaves results in decreasing complexity. While KNOX1 activity has been shown to play a pivotal role, other loci, such as REDUCED COMPLEXITY (RCO) in Cardamine and LEAFY (LFY) orthologues in legumes either contribute directly to modifying leaf shape or influence sensitivity to KNOX1 activity [11,47,48]. The lability of angiosperm leaf architecture may reflect that addition or loss of enhancer modules directing KNOX1 activity in leaves does not affect general plant viability.
The present study demonstrates that KNOX2 activities can also influence leaf shape—leaf dissection increases with decreasing KNOX2 activity (Fig. 2) in a dose dependent manner—raising the possibility of whether changes in KNOX2 activity could also have contributed to the evolution of leaf morphology. Just as KNOX1 gain-of-function alleles result in increases in leaf complexity, novel gain-of-function KNOX2 alleles that alter temporal or spatial expression patterns within developing leaves could contribute to the evolution from complex towards simple leaf morphology, as suggested by our experimental results in Cardamine, via acquisition of leaf specific enhancers. Alleles resulting in loss of KNOX2 activity could also contribute to increases in leaf complexity as suggested by the dose dependent changes to leaf shape in Arabidopsis, however, this may be less likely due to pleiotropic effects of loss-of-function KNOX2 alleles.
Intriguingly, in monilophytes KNOX1 gene expression is broadly similar to that of seed plants, with expression limited to less differentiated tissues including the shoot apical meristem, developing leaves, and procambial tissues [43,49,50]. KNOX2 gene expression has not been studied in detail, but similar to the situation in angiosperms, is reported to be throughout the sporophyte body [50]. In parallel with seed plants, simple leaves have evolved from more complex ancestral leaves within monilophytes [51]. Whether changes in KNOX1 or KNOX2 gene expression may be related to evolution of leaf form in monilophytes is presently unknown.
One plausible explanation for the opposing action of KNOX1 and KNOX2 genes is an epistatic relationship between the gene classes. While non-overlapping expression patterns have been observed between KNOX1 and KNOX2 genes, we found no evidence for mutual repression. Alternatively, KNOX1 and KNOX2 proteins may interfere one another’s activity. Such a mode of action was proposed for KNATM in Arabidopsis and PETROSELINUM (PTS)/TKD1 in tomato, both of which are KNOX-related proteins that lack a DNA-binding homeodomain [52,53]. It is suggested that these mini KNOX proteins act as passive repressors and interfere with formation of a functional complex composed of canonical KNOX and BELL proteins. That KNOX2 function depends on the availability of appropriate BELL partners to be active, argues against a similar mechanism for the KNOX1/KNOX2 antagonism. Instead, our data favor a model whereby the antagonistic roles of KNOX1 and KNOX2 are at the level of opposing modes of transcriptional regulation.
Since addition of a repressor domain causes a dominant negative phenotype, KNOX1 proteins can act as activators [39,54]. Conversely a KNOX2 protein, KNAT7, can repress transcription in a transient protoplast system [18,19], and a motif similar to known repression domains is found in the ELK domain of all KNOX2 proteins [55] (S16 Fig.). Comparison of KNOX1 and KNOX2 homeodomains reveals that the third helices, an important determinant of DNA binding specificity, are highly conserved, indicating similar DNA binding properties, at least in vitro (S16 Fig.). Concurrently expressed KNOX1 and KNOX2 proteins could thus conceivably compete with each other at some target genes. Indeed, a putative KNOX2-SAW2 complex can overcome endogenous KNOX1 activities in the meristem, as does a dominant-negative form of KNOX1 (e.g., TKN2-SRDX [39] and en298-STM [54]). However, as KNOX1 proteins have also been reported to act to repress gene expression, the activity of KNOX proteins may be modified by either BELL partners, or third parties, such as OVATE proteins that interact with KNOX/BELL heterodimers and influence both their cellular localization and transcriptional activity [32,37,56]. In a related scenario, KNOX1 and KNOX2 could act on different sets of paralogs of downstream targets. These hypotheses are not mutually exclusive, and depending on the cellular contexts, different modes of action could operate, as is the case for the yeast TALE protein, Matα2, which has different partners in different cell types (reviewed in [23]).
Phylogenetic analyses indicate land plant KNOX1 and KNOX2 genes are derived from a single, ancestral KNOX gene. We hypothesize that subsequent to the KNOX1/KNOX2 gene duplication, accumulating structural differences endowed a new mode of action to at least one paralog. Therefore a possible evolutionary scenario could have an ancestral KNOX protein acting primarily as a transcriptional activator, with the evolution of a transcriptional repressor following gene duplication and neofunctionalization. The evolution of a repressor from an ancestral activator may be a common event, with several instances documented in plant transcription factor families [52,53,57,58,59,60]. Thus, within the context of land plant KNOX genes two types of negative regulators, in which the modes of repressor action are mechanistically different, may have evolved. Mini KNOX proteins act to inhibit KNOX activity by interacting with and sequestering BELL proteins [52,53], as opposed to antagonistic action at the level of downstream gene expression as we propose for KNOX2. The latter provides more flexibility due to the potential to act independently. Accompanying divergence in protein functionality, our data provides additional evidence for nearly complementary expression patterns of KNOX1 and KNOX2 genes in Arabidopsis thaliana. In contrast, in P. patens KNOX1 and KNOX2 genes exhibit both overlapping and distinctive expression patterns [14,26]. Changes in cis-regulatory sequences must have contributed to the establishment of complementary expression patterns during land plant evolution. Flexibility in gene regulatory networks governing meristematic maintenance and differentition engendered by the combination of changes in protein functionality and expression pattern could provide plasticity enabling morphological evolution.
Heterodimerization between BELL and KNOX proteins is important for translocation of the complex into the nucleus [13]. BELL-KNOX2 heterodimerization may also be critical for providing specificity or increasing affinity of DNA binding (e.g. [61]). Although studies based on the yeast two-hybrid technique suggest physical interactions between BELL and KNOX proteins in a rather nonspecific manner [29,32], our genetic data suggest KNOX2 proteins interact in planta with a subset of BELL proteins, including those of the BEL1/SAW1/SAW2 clade. KNOX1 proteins rely on a distinct set of BELL proteins, e.g. PNY and PNF (reviewed in [12,13]). Due to an obligate heterodimerization requirement, the activity of a KNOX/BELL pair may be limited by the protein with the more restricted expression domain. In Arabidopsis KNOX2 functions appear to be regulated by restricted availability of corresponding BELL partners [29] (Fig. 3).
Similar to KNOX genes, land plant BELL genes evolved from a single gene in the algal ancestor [9]. However, the diversification of paralogs followed a different trajectory in the two families since BELL genes do not fall into discrete functional clades (S17 Fig.). For instance, KNOX1-interacting BELL genes (PNY and PNF) form a sister clade with KNOX2-interacting BELL genes (BEL1 and SAW1/2). Moreover, genetic interactions implicate BLH1, from a phylogenetically distinct clade, as a KNOX2 partner since knat3 alleles suppress the phenotype induced by ectopic BLH1 embryo sac expression [37]. These phylogenetic relationships might be expected if the genome of the land plant common ancestor encoded a single BELL protein that interacted with both KNOX1 and KNOX2 proteins. As the BELL gene family diversified, subfunctionalization would have restricted interactions of BELL paralogs to specific KNOX1 or KNOX2 partners.
The defining feature of land plants is the formation of an embryo—a multicellular diploid generation. One prominent feature within land plant evolution is the transition from a gametophyte-dominant life cycle to a sporophyte-dominant life cycle [62,63]. This process is regarded as progressive sterilization and elaboration of vegetative organs [62], and in flowering plants, the gametophyte is reduced to a ephemeral structure of only a few cells that is dependent on a sporophyte body that can live up to thousands of years. If the ancestral KNOX-BELL genetic program regulated gene expression in a single celled zygote [25], it follows that during the course of land plant evolution, the KNOX/BELL module has been recruited to control numerous aspects of sporophyte development, with KNOX1/BELL modules promoting meristematic maintenance and continued growth and KNOX2/BELL modules promoting differentiation. In some cases, there is resemblance to a presumed ancestral function, such as in P. patens where KNOX2 genes regulate the gametophyte-to-sporophyte morphological transition [14,26]. In other cases, however, KNOX/BELL modules direct the development of novel structures, such as sporophyte shoot meristems and leaves (Fig. 6), that evolved later in land plant evolution, suggesting the duplication and diversification of the KNOX/BELL genetic module is linked with the evolution of morphological diversity in the land plant sporophyte. Neofunctionalization, exemplified by opposing activities between KNOX1 and KNOX2 genes in Arabidopsis, may underlie the molecular mechanism of key innovations and modification of body plans in the land plant history, through elaboration of transcriptional networks.
The role of TALE genes in fungi and Chlamydomonas can be viewed as promotion of cellular specialization in the diploid zygote and progression towards a meiotic state. The life cycle of land plants arose by an interpolation of mitotic divisions between fertilization and meiosis. Thus there is cell proliferation and a delay in meiosis in the diploid generation. KNOX1 genes prevent differentiation and maintain an undifferentiated state of the cells, enabling the cells to proliferate and develop a multicellular body in the sporophyte generation. In organisms with two heteromorphic multicellular generations, such as land plants, the developmental programs for each must be tightly controlled—a role suggested for KNOX2 genes in preventing the haploid gametophyte genetic program to be active during the diploid sporophyte generation in Physcomitrella. We hypothesize the duplication and diversification of the KNOX/BELL genetic module was instrumental in the evolution of a diploid embryo such that multicellular bodies develop in both haploid gametophyte and diploid sporophyte generations known as alternations of generations [25,26]. Alternations of generations have evolved independently in phylogenetically diverse eukaryotic lineages [64,65], prompting the question of whether similar TALE class genetic diversification may be found in these lineages.
Arabidopsis thaliana accessions Columbia and Landsberg erecta (Ler) were used as wild type in most experiments. proKNAT2:GUS was generated in the C24 background and introgressed into Ler. Cardamine hirsuta ‘Oxford strain’ is a kind gift of A. Hay and M. Tsiantis. Plants were grown under long-day (18 hours light) or short-day (10 hours light) conditions at 20°C. knat3 and knat5 alleles are gift from V. Sundaresan and G. Pagnussat. bp-9 knat2–5 knat6–1 seeds are gift from V. Pautot. T-DNA insertion alleles for BELL and KNOX genes were obtained from the Arabidopsis Biological Resource Center (ABRC) or the Nottingham Arabidopsis Stock Center (NASC). Mutant and transgenic lines have been described previously: bp-9 knat2–5 knat6–1 [66]; stm-11 [67]; proBP:GUS [68]; proKNAT2:GUS [69]; Op:KNAT2 and Op:STM [39]; and proSTM:LhG4 [70]. The mutant and transgenic lines used in this study are listed in S1 Table. Homozygous mutant lines were identified by polymerase chain reaction (PCR)-based genotyping. Sequences of genotyping primers are available in S2 Table. The details of the transactivation system was previously described [33].
Multiple mutants combining knat3, knat4, and knat5 alleles were generated by crossing, and genotypes were confirmed by PCR-based genotyping. To generate bel1 blh1 double mutant, blh1 plants were crossed with bel1 plants, and the resulting F2 plants were examined. bel1 plants were identified based on self-sterility, and among them, plants with yellow gynoecia segregated and were confirmed to be bel1 blh1 double mutant plants by PCR-based genotyping. knat2 knat3 knat5 knat6 and bp knat345 plants were identified among F2 plants originating from a cross between bp knat2 knat6 and knat345 plants, and their genotypes were confirmed by PCR-based genotyping. cuc2 knat345 plants were identified in a F2 population derived from a cross between cuc2 and knat345 plants. To generate proFIL:miR164b lines in the knat345 mutant, self-fertile knat3 knat5 plants were transformed with the proFIL:miR164b construct, and tranformants were selected by resistance to herbicide Basta. Single insertion lines were selected and crossed with knat3/+ knat4 knat5 plants. Among F1 plants, self-fertile knat3/+ knat4/+ knat5 plants carrying the proFIL:miR164b transgene were selected, and F2 seeds were collected; proFIL:miR164b knat345 plants were identified in the resultant F2 population. To characterize the effects of the pro35S:amiR159-KNAT345–1 transgene in mutant backgrounds, the mutant plants were directly transformed with the pro35S:amiR159-KNAT345–1 construct, and transformants were selected by resistance to Basta. As stm null alleles are seedling lethal, heterozygous plants were used for transformation. More than twenty T1 plants for each background were examined, and phenotypes consistently observed among independent lines were reported.
RNA was extracted, using the RNeasy Plant Mini Kit (Qiagen), from 10-day-old seedlings grown on half-strength MS medium supplemented with 0.5% sucrose. RNA samples were treated with on-column DNaseI (Qiagen) and purified. SMARTScribe reverse transcriptase was used for cDNA synthesis (Clontech), and PCR reactions were performed using Ex Taq (Takara). Oligo sequences used for PCR reactions are described in S3 Table.
amiRNAs were designed using the Arabidopsis pre-miR159a backbone (S7 Fig.) and synthesized (GenScript). For construction of the proKNAT5:KNAT5-GUS reporter construct, the genomic sequence spanning the KNAT5 locus (from the next upstream annotated gene [At4g32030] to the next downstream annotated gene [At4g32050]) was used, and the stop codon was replaced with the GUS coding sequence. For construction of the proKNAT4:GUS reporter construct, an approximately 6.6-kb region of the sequence directly upstream of the KNAT4 coding sequence was amplified using BAC T5K6 as PCR template and cloned into pCRII-TOPO (Invitrogen). The KNAT4 upstream sequence was subcloned into the pRITA vector, which contains the GUS coding sequence and the terminator sequence from the nopaline synthase gene. For constitutive expression, the amiRNA sequences or the KNAT3 coding sequence were cloned into the ART7 vector, which contains the Cauliflower mosaic virus pro35S sequence and the terminator sequence from the octopine synthase gene. KNAT5, SAW2, and PNY coding sequences were amplified from Ler cDNA and cloned downstream of an Lac Op array [33] to generate responder cassettes used in the transcription activation system. All constructs were subcloned into pMLBART or pART27 binary vector and were introduced into Agrobacterium tumefaciens strain GV3001 by electroporation. Transgenic lines were generated by Agrobacterium-mediated transformation, and transformants were selected on soil on the basis of resistance to the BASTA or kanamycin. Primers used to clone the various cDNAs and promoters are described in S2 Table.
Scanning electron microscopy was performed according to Alvarez and Smyth [71]. For light microscopy, cleared samples were prepared. Leaf samples were fixed overnight in 9:1 (v:v) ethanol:acetic acid at room temperature. After rehydration in a graded ethanol series, samples were rinsed with water and were cleared with chloral hydrate solution [1:8:2 (v:w:v) glycerol:chloral hydrate:water]. For histochemical analysis of GUS activity, samples were infiltrated with GUS staining solution [0.2% (w/v) Triton X-100, 2 mM potassium ferricyanide, 2 mM potassium ferrocyanide, and 1.9 mM 5-bromo-4-chloro-3-indolyl-β-glucuronide in 50 mM sodium phosphate buffer, pH 7.0] and incubated at 37°C.
Publically available KNOX and BELL coding nucleotide sequences representing taxa across land plants were manually aligned as amino acid translations using Se-Al v2.0a11 (http://tree.bio.ed.ac.uk/software/seal/). We excluded ambiguously aligned sequence to produce alignments for subsequent Bayesian analysis. Bayesian phylogenetic analysis was performed using Mr. Bayes 3.2.1 [72,73]. Three separate analyses were performed. The first included Chlorophyte algal and land plant KNOX sequences (S1 Fig.); the second included only land plant KNOX2 sequences (S2 Fig.); and the third included land plant BELL sequences (S17 Fig.). The fixed rate model option JTT + I was used based on analysis of the alignments with ProTest 2.4 [74]. Sequence alignments and command files used to run the Bayesian phylogenetic analyses are provided upon request.
|
10.1371/journal.pntd.0002325 | TNF-α and CD8+ T Cells Mediate the Beneficial Effects of Nitric Oxide Synthase-2 Deficiency in Pulmonary Paracoccidioidomycosis | Nitric oxide (NO), a key antimicrobial molecule, was previously shown to exert a dual role in paracoccidioidomycosis, an endemic fungal infection in Latin America. In the intravenous and peritoneal models of infection, NO production was associated with efficient fungal clearance but also with non-organized granulomatous lesions. Because paracoccidioidomycosis is a pulmonary infection, we aimed to characterize the role of NO in a pulmonary model of infection.
C57Bl/6 wild type (WT) and iNOS−/− mice were i.t. infected with 1×106 Paracoccidioides brasiliensis yeasts and studied at several post-infection periods. Unexpectedly, at week 2 of infection, iNOS−/− mice showed decreased pulmonary fungal burdens associated with an M2-like macrophage profile, which expressed high levels of TGF-β impaired ability of ingesting fungal cells. This early decreased fungal loads were concomitant with increased DTH reactions, enhanced TNF-α synthesis and intense migration of activated macrophages, CD4+ and CD8+ T cells into the lungs. By week 10, iNOS−/− mice showed increased fungal burdens circumscribed, however, by compact granulomas containing elevated numbers of activated CD4+ T cells. Importantly, the enhanced immunological reactivity of iNOS−/− mice resulted in decreased mortality rates. In both mouse strains, depletion of TNF-α led to non-organized lesions and excessive influx of inflammatory cells into the lungs, but only the iNOS−/− mice showed increased mortality rates. In addition, depletion of CD8+ cells abolished the increased migration of inflammatory cells and decreased the number of TNF-α and IFN-γ CD4+ and CD8+ T cells into the lungs of iNOS−/− mice.
Our study demonstrated that NO plays a deleterious role in pulmonary paracoccidioidomycosis due to its suppressive action on TNF-α production, T cell immunity and organization of lesions resulting in precocious mortality of mice. It was also revealed that uncontrolled fungal growth can be overcome by an efficient immune response.
| Paracoccidiodomycosis is a human systemic mycosis endemic in Latin America that has a wide spectrum of manifestations ranging from localized to fatal disseminated forms. Both in humans and experimental models, immunoprotection is mediated by T cell immunity whereas immunosuppression is associated with the severe forms of the disease. The literature shows that nitric oxide (NO) produced by the enzyme nitric oxide synthase-2 (NOS2 or iNOS) is the major fungicidal component of phagocytic cells. The role of NO production was previously investigated in the intra-peritoneal and intravenous murine models of Paracoccidioides brasiliensis infection. The human paracoccidioidomycosis is believed to be acquired by the respiratory route, thus our study aimed to characterize the role of NO production in a pulmonary model of infection. We verified that, paradoxically, absence of NO production by iNOS- deficient mice resulted in less severe disease and increased survival times. This was associated with increased development of cellular immunity and enhanced synthesis of TNF-α which enhances cell migration to the site of infection and contributes to the better organization of lesions. Our work highlighted the deleterious effect of excessive NO production in pulmonary paracoccidioidomycosis, and demonstrated that uncontrolled fungal growth can be overridden by an efficient immune response.
| Phagocytes are important effector cells of innate and adaptative immunity and use several mediators and mechanisms to control pathogen growth. The production of nitric oxide (NO) has been shown to be an important microbicidal mechanism of macrophages in the protective immune responses against different pathogens [1], [2], [3], [4]. NO is generated from the amino acid L-arginine by the catalytic action of the inducible isoform of NO synthase (iNOS or NOS2) [5]. The synergistic interaction of NO with hydrogen peroxide (H2O2) or superoxide (O2−) anion can generate extremely potent oxidizing compounds resulting in cell damage and microbicidal activity [6], [7]. Besides its action on pathogens viability, there are also evidences that NO has an inhibitory effect in the innate and adaptive immunity of hosts. For example, it reduces the antigen-presenting ability of pulmonary dendritic cells, inhibits MHC class II antigen expression, controls the production of cytokines and expression of costimulatory and adhesion molecules [8], [9]. The iNOS gene expression is regulated by an ever-increasing number of agonists, especially proinflammatory cytokines such as IFN-γ and TNF-α and bacterial products such as lipopolysaccharides from Gram-negative bacteria [10], [11]. On the other hand, type 2 cytokines, especially IL-4, IL-10 and TGF-β, were shown to inhibit NO production [12], [13], [14].
Similarly with other microorganisms, the production of NO has been associated with the protective immunity against several fungal pathogens [15], [16]. Recent evidences, however, suggested that certain fungal species such as Cryptococcus neoformans and Aspergillus fumigatus developed ingenious mechanisms to evade nitric oxide-dependent death [17], [18].
Paracoccidioidomycosis (PCM), a fungal disease caused by the inhalation of P. brasiliensis spores, presents a wide spectrum of immunopathological manifestations [19]. Patients with benign PCM usually develop adequate cellular immune responses and their antigen-stimulated leukocytes preferentially secrete type 1 cytokines; in contrast, patients with the severe form of the disease show impaired cell mediated immunity and type 2-skewed immune response [20], [21], [22]. Recent investigations, however, indicate that other regulatory mechanisms, not involving Th1/Th2 cells, play an important role in the immunopathogenesis of PCM [21].
Although the mechanisms involved in resistance to P. brasiliensis infection are not completely understood, it appears that alveolar macrophages have a fundamental role, acting as the first line of host defense. The enhanced fungicidal ability of cytokine-activated macrophages was shown to be mainly mediated by NO [23], [2]. Despite this protective activity, in some studies NO production was associated with suppression of lymphoproliferation and MHC class II expression [24], [25]. Interestingly, in P. brasiliensis infection we could detect an inverse correlation between TNF-α synthesis and NO production. Peritoneal and alveolar macrophages from resistant A/J mice in vitro infected with P. brasiliensis yeasts secreted high TNF-α levels, low NO amounts and displayed low fungicidal ability associated with enhanced TGF-β secretion. In contrast, macrophages from susceptible B10.A mice secreted high NO levels, presented efficient fungal killing but produced low levels of TNF-α [26], [27]. NO deficiency was also associated with organized granulomas of i.p. infected mice [28], while exacerbated inflammatory reactions and cytokines production was described in i.v. infected mice [29].
Because P. brasiliensis infection is acquired by the respiratory route and the role of NO was never investigated in the pulmonary model of PCM, we aimed to further understand the immunoregulatory function of this mediator using i.t. infected iNOS-deficient (iNOS−/−) and normal (WT) C57BL/6 mice. We could characterize the temporal effects of NO synthesis in the control of fungal growth. At week 2 of infection, absence of NO results in lower fungal loads but at week 10, increased numbers of yeasts were detected in the lungs of iNOS−/− mice. Unexpectedly, the deficient mouse strain showed increased survival times and this behavior was associated with high levels of TNF-α production increased and persistent delayed type hypersensitivity reactions and enhanced migration of activated T cells and macrophages into the lungs of infected mice. In addition, the increased fungal loads lately developed by iNOS-deficient mice appeared to be contained by better organized granulomatous lesions. Furthermore, in vivo depletion experiments showed that the protective effect of iNOS deficiency was mainly mediated TNF-α and the expansion of IFN-γ and TNF-α CD4+ and CD8+ T cells.
Animal experiments were performed in strict accordance with the Brazilian Federal Law 11,794 establishing procedures for the scientific use of animals, and the State Law establishing the Animal Protection Code of the State of São Paulo. All efforts were made to minimize suffering, and all animal procedures were approved by the Ethics Committee on Animal Experiments of the Institute of Biomedical Sciences of University of São Paulo (Proc.76/04/CEEA).
Breeding pairs of homozygous iNOS-deficient (iNOS−/−) and wild type (WT) control C57BL/6 mice (intermediate susceptibility to P. brasiliensis) were bred at the University of São Paulo animal facilities under specific-pathogen-free (SPF) conditions in closed-top cages. Clean food and water were given ad libitum. Mice were 8 to 11 weeks of age at the time of infection, and procedures involving animals and their care were approved by the Ethics Committee on Animal Experiments of our Institution.
The P. brasiliensis 18 isolate, which is highly virulent, was used throughout the study. To ensure the maintenance of its virulence, the isolate was used after three serial animal passages [30]. P. brasiliensis 18 yeast cells were then maintained by weekly subcultivation in semisolid Fava Netto culture medium [31] at 35°C and used on the seventh day of culture. The fungal cells were washed in phosphate-buffered saline (PBS; pH 7.2), counted in a hemocytometer and the concentration was adjusted to 20×106 fungal cells ml−1. The viability of fungal suspensions, determined by Janus Green B vital dye (Merck, Darmstadt, Germany) [32], was always higher than 80%. Mice were anesthetized and submitted to i.t. P. brasiliensis infection as previously described [33]. Briefly after intraperitoneal anesthesia, the animals were infected with 1×106 P. brasiliensis 18 yeast cells, contained in 50 µL of PBS, by surgical i.t. inoculation, which allowed dispensing of the fungal cells directly into the lungs. The skins of the animals were then sutured, and the mice were allowed to recover under a heat lamp. Mice were studied at several time points after infection.
The number of viable microorganisms in infected organs (lung, liver and spleen) from experimental and control mice were determined by counting the number of CFU. Animals (n = 6–8) from each group were sacrificed, and the enumeration of viable organisms was done as previously described [34]. Briefly, aliquots (100 µL) of the cellular suspensions and serial dilutions were plated on brain heart infusion agar (Difco, Detroit, USA) supplemented with 4% (vol/vol) horse serum (Instituto Butantan, São Paulo, Brazil) and 5% P. brasiliensis 192 culture filtrate, the latter constituting a source of growth-promoting factor. The plates were incubated at 35°C, and colonies were counted daily until no increase in counts was observed. The number (log10) of viable P. brasiliensis colonies per gram of tissue were expressed as means ± standard errors (SEs).
Mice (n = 6–8) were infected i.t. with P. brasiliensis, their right lung and liver of anti-TNF-α and IgG treated mice, were removed aseptically and individually disrupted in 5.0 mL of PBS. Supernatants were separated from cell debris by centrifugation at 2,000× g for 15 min, passed through 0.22 µm pore-size filters (Millipore, Bedford, Mass, USA), and stored at −70°C. The levels of IL-2, IL-12, IFN-γ, TNF-α, IL-4, IL-5, IL-10 and TGF-β were measured by capture enzyme-linked immunosorbent assay (ELISA) with antibodies pairs purchased from Pharmingen (Pharmingen, San Diego, CA, USA). The ELISA procedure was performed according to the manufacture's protocol. The concentrations of cytokines were determined with reference to a standard curve for several twofold dilutions of murine recombinant cytokines. As an addition control, lung homogenates were added to recombinant cytokines used to obtain standard curves; no interference was detected, indicating the absence of inhibitory substances (e.g., soluble cytokine receptors).
Two and ten weeks after i.t. infection, lungs of mice were lavaged by repeated injections of 0.5 ml of sterile PBS (final volume 2.0 ml) after cannulation of the trachea with polyethylene tubing which was attached on a tuberculin syringe. An aliquot of the recovered bronchoalveolar lavage fluid (BALF) was assayed by CFU to determine the presence of viable yeasts. Then, the remaining BALFs obtained from individual mice were spun at 1200 rpm, the supernatants removed and alveolar macrophages cultivated to characterize the presence of viable fungal cells. Cell pellets were resuspended in RPMI containing 10% fetal calf serum, 2 mM L-glutamine, 100 U/ml penicillin and 100 µg/ml streptomycin, adjusted at 4×105/ml of culture medium and 0.5 ml dispensed in 24-well tissue culture plates for a 2 h adhesion step. Non-adherent cells were discarded and some cultures treated with 0.5 ml of culture medium supplemented with 100 U/ml of IFN-γ (Pharmingen, San Diego, CA, USA) and cultivated for 48 h. Plates were then centrifuged, supernatants stored, cells disrupted by five washes with 0.5 ml of distilled water and suspensions collected in individual tubes. Pellets were resuspended in culture medium, and aliquots (100 µL) and their serial dilution were assayed for the presence of viable yeasts. In addition, NO and H2O2 levels were determined in the supernatants of alveolar macrophages cultures.
Peritoneal macrophages from WT and iNOS −/− C57BL/6 mice were induced by i.p. injection of brewer thioglycollate medium (Difco, Detroit, MI, USA). Macrophages were isolated by adherence (2 h at 37°C in 5% CO2) to plastic-bottom tissue-culture plates (1×106 cells/well in 24 well plates), cultivated overnight with fresh complete medium in the presence or absence of recombinant IFN-γ (40 ng/ml, BD-Pharmingen San Diego, CA, USA), TNF-α (20 ng/ml, BD-Pharmingen San Diego, CA, USA) or 1-methyl-DL-tryptophan (1 MT, 1 mM in culture medium, Sigma Aldrich, St. Louis, MO, USA), a specific inhibitor of 2,3 indoleamine dioxygenase. Macrophage cultures were infected with P. brasiliensis yeasts in a macrophage∶yeast ratio of 12.5∶1. After 48 h of culture, plates were centrifuged and supernatants removed. The wells were washed with distilled water to lyse macrophages, the suspensions collected and assayed for the presence of viable yeasts. All assays were done with five wells per condition in over three independent experiments.
For phagocytic assays, macrophages from WT and iNOS −/− mice were infected with heat-inactivated, FITC labeled P. brasiliensis yeasts at a macrophage∶yeast ratio of 1∶1 for 2 h at 37°C in 5% CO2 to allow fungi adhesion and ingestion as previously described [35]. Some macrophage cultures were treated with IFN-γ (40 ng/ml, BD-Pharmingen), or TNF-α (20 ng/ml BD-Pharmingen) overnight, before infection. Macrophages were gently washed twice with PBS and cells detached from plastic with fresh cold medium and a rubber cell scraper on ice. The cells were transferred to tubes, centrifuged (400×g. 10 min., 4°C), and the pellets were labeled with anti-F4/80 (APC) antibodies (eBioscience, San Diego, CA, USA). The cells were washed twice in PBS, the pellets were suspended in 200 µL of PBS 1% FCS and were immediately read on FACScalibur (Becton Dickinson, Franklin Lakes, NJ, USA) and data analyzed using the FlowJo software program (Tree Star, Ashland, OR, USA).
For the distinction between internalized and surface-bound yeasts (FITC- labeled P.brasiliensis particles), trypan blue (TB, 250 µg/mL, Sigma Aldrich, St. Louis, MO, USA) was used for quenching the green surface-bound fluorescence on macrophages. TB quenching technique was performed as described by Busetto et al. [36] with minor modifications. Phagocytic assays were performed as above described and adherent/ingested cells measured using the FL1 and FL4 channels of a FACscalibur cytometer. Cell suspensions were then treated in an ice bath with 0.1 ml of a TB solution prepared in 0.1 M citrate buffer, pH 4.0, lowering samples pH to nearly 4.0, thereby optimizing the TB quenching effect. After 1 min of incubation in ice bath, the samples were again analyzed. APC-labeled macrophages were gated, and FL1 and FL3 channels used to discriminate ingested (green fluorescent, FL1) from adherent (red fluorescent, FL3) yeasts.
Total RNA was extracted from cultures of normal or P. brasiliensis-infected macrophages using the TRIzol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's instructions. The RNA concentrations were determined by spectrophotometer readings at an absorbance of 260 nm. First-strand cDNAs were synthesized from 2 µg RNA using the High Capacity RNA-to-cDNA kit (Applied Biosystems, Foster City, CA, USA) according to the manufacturer's instructions. Real-time polymerase chain reaction (RT-PCR) was performed using the TaqMan real-time PCR assay (Applied Biosystems) for the following molecules: ARG1 (Mm00475988_m1), NOS2 (Mm00440502_m1), IL-12 p40 (Mm00434174_m1), TGF-β (Mm00441727_m1), TNF-α (Mm99999068_m1), IDO (Mm00492586_m1). Cycling conditions were as follows: 10 min at 95°C, followed by 45 cycles of 20 s at 95°C, 20 s at 58°C, and 20 s at 72°C. Analysis was performed with the ABI PRISM 7000 sequence detection system (Applied Biosystems). GAPDH was used as an internal control. All values were normalized to GAPDH, and the relative gene expression was calculated using the Pfaffl method [37]
Nitric oxide production was quantified by the accumulation of nitrite in the supernatants by a standard Griess reaction. Briefly, 50 µL of supernatants was removed from 24-well plates and incubated with an equal volume of Griess reagent (1% sulfanilamide, 0,1% naphthylene diamine dihydrochloride, 2,5% H3PO4) at room temperature for 10 min. The absorbance at 550 nm was determined with a microplate reader. The conversion of absorbance to micromolar NO was deduced from a standard curve by using a known concentration of NaNO2 diluted in RPMI medium [38]. All determinations were performed in triplicates and expressed as micromolar NO.
DTH reactions were evaluated employing a footpad test previously described [34]. Briefly, mice were inoculated with 25 µl of a soluble P. brasiliensis antigen [31] and footpad thickness was measured with a caliper (Mitutoyo, Tokyo, Japan) immediately before and 24 h after antigen inoculation. The increase in thickness was calculated and expressed in millimeters. Non-infected mice submitted to the footpad test were used as controls.
Lungs from each mouse were excised, washed in PBS, minced, and digested enzymatically for 1 hour in 15 mL/lung of digestion buffer [RPMI, 5% fetal calf serum, 1 mg/mL collagenase and 30 µg/mL DNase (Sigma Aldrich, USA)]. After erythrocyte lysis using NH4Cl buffer, cells were washed, resuspended in complete media, and centrifuged for 30 minutes at 2,000× g in presence of 20% Percoll (Sigma) to separate leukocytes from cell debris and epithelial cells. Total lung leukocyte numbers were assessed in the presence of trypan blue using a hemocytometer; viability was always higher than 85%. The absolute number of a leukocyte subset was equal to the percentage of that cell subset multiplied by the total number of leukocytes recovered from the digested lung/100.
For surface staining alone, leukocytes were washed and resuspended at a concentration of 1×106 cells/mL in staining buffer (PBS 1×, 2% serum calf bovine and 0,5% NaN3). Fc receptors were blocked by the addition of unlabeled anti-CD16/32 (Fc block; BD Pharmingen, San Diego, CA, USA). The leukocytes were then stained for 20 min at 4°C with the optimal dilution of each antibody. Anti-CD4, CD8, CD69, CD25, CD40 CD80, CD86, CD11b and CD11c FITC or PE-conjugated antibodies were from BD Pharmingen. Cells were washed twice with staining buffer resuspended in 100 µl, and an equal volume of 2% formalin was added to fix the cells. The stained cells were analyzed immediately on a FACScalibur equipment using the Cell-Quest software (Becton & Dickinson, Sparks, MD, USA) gating on macrophages or lymphocytes as judged from forward and side light scatter. Ten thousands cells were counted and the data expressed as the percentage or the absolute number of positive cells which was calculated trough the percentage obtained by FACS and the number of cells determined in Neubauer chambers. The intracellular detection of FoxP3, the X-linked forkhead/winged helix transcription factor, in leukocytes obtained from the lung lesions was performed in fixed and permeabilized cells using Cytofix/Cytoperm (BD Biosciences, San Diego, CA, USA). Initially, the cells were labeled with antibodies for cell surface molecules such as FITC-conjugated anti-CD4 and PE-conjugated anti-CD25. Next, the cells were fixed, permeabilized, and stained with Cy-conjugated anti-FoxP3, for 1,5 h at 4°C. Cells were then washed twice with staining buffer, resuspended in 100 µl, and an equal volume of 2% formalin was added to fix the cells. A minimum of 20,000 events was acquired on FACScalibur flow cytometer (BD Pharmingen) using the Cell-Quest software (BD Pharmingen). The graphs represent the number of Foxp3+ cells in the gate of CD4+ CD25+ T cells. For intracellular cytokine (IL-4, IFN-γ and TNF-α) staining, cells were stimulated for 6 h in complete medium in the presence of 50 ng/ml phorbol 12-myristate 13-acetate, 500 ng/ml ionomycin (both from Sigma-Aldrich) and monensin (3 mM, eBioscience). After surface staining for CD4 (Pacific Blue anti-CD4) and CD8 (Alexa Fluor 488 anti CD8), cells were fixed, permeabilized, and stained by PerCP- Cy5.5 anti-IFN-γ, Pe-Cy7 anti-IL-4 and PE anti-TNF-α antibodies (eBioscience, San Diego,CA, USA). The cell surface expression of leukocyte markers as well as intracellular expression of IL-4, IFN-γ and TNF-α in lung infiltrating leucocytes, were analyzed in a FACScalibur flow cytometer (BD Pharmingen) using the FlowJo software (Tree Star, Ashland, OR, USA).
In selected experiments, lungs from infected iNOS−/− and WT mice were removed and digested enzymatically as above described. DCs were purified by magnetic cell sorting with microbeads (Miltenyi, Bergisch Gladbach, Germany) conjugated to hamster anti-mouse CD11c monoclonal antibodies. Positively selected DCs contained more than 90% CD11c+ cells. Cell-surface markers of pulmonary DCs were characterized by flow cytometry using monoclonal antibodies anti-CD11c PE-Cy7, anti-CD11b PerCP-Cy5.5, anti-CD8a-Alexa Fluor 488, and anti- B220-PE.
P. brasiliensis-infected (iNOS−/− and WT) mice were given i.p. injections of 0.25 mg/0.5 mL of an anti-TNF-α MAb (MP6 XT 22.11), a rat IgG1 monoclonal antibody, against murine TNF-α [39], 4 h before the infection, and at days 6 and 12 postinfection. Normal rat IgG was given i.p. to mice as a control for antibody administration. Treated and untreated mice (n = 6–7) were studied at weeks 2 and 8 weeks after infection and mortality rates were also evaluated. H-35 (rat IgG1 anti-CD8α) hybridoma was grown in BALB/c nu/nu mice. Monoclonal antibodies (MAbs) were purified from ascites as previously described [40] and assayed for purity by sodium dodecyl sulfate-polyacrylamide gel electrophoresis. Groups (n = 6–7) of WT and iNOS−/− mice were given 200 µg of anti-CD8α mAbs or normal rat IgG (controls) by the i.p. route, 48 and 24 h before infection and 150 µg of the mAbs or rat IgG at days 6 and 12 postinfection. The severity of infection and lung infiltrating leukocytes were characterized at week 2 after infection.
Groups of iNOS−/− mice and their normal counterparts were killed at the second, eighth and tenth week postinfection. Lungs were collected, fixed in 10% formalin and embedded in paraffin. Histolopathogical studies were also performed with anti-TNF-treated and untreated mice at week 8 after infection. Five-micrometer sections were stained by the hematoxilin-eosin (H&E) for an analysis of the lesions. Pathological changes were analyzed based on the size, morphology and cell composition of granulomatous lesions, presence of fungi and intensity of the inflammatory infiltrates.
Mortality studies were done with groups (n = 6–8) of iNOS−/− and WT control mice inoculated i.t. with 1×106 yeast cells or PBS. Deaths were registered daily for a 350-day period, and the median survival time postinfection was calculated. Mortality of untreated and anti-TNF-α -treated mice (n = 6–7) of both mouse strains were also studied. Experiments were repeated three times.
All values are means ± SEM, unless otherwise indicated. Depending on the number of experimental groups, data were analyzed by Student's t test or two-way analysis of variance and the Bonferroni posttests to compare groups. Differences between survival times were determined with the LogRank test using GraphPad Prism software (GraphPad Software, San Diego, CA, USA). P value<0.05 was considered significant.
The evolution of the disease of i.t. infected iNOS−/− mice and their WT controls was monitored by CFU counts in the lung and liver at different postinfection periods (48 h, 2, 6 and 10 weeks) (Figure 1A and 1B). At the first 48 h of infection, an equivalent number of viable yeasts cells was recovered from lungs of both mouse strains. Interestingly, at the 2nd week, iNOS deficiency resulted in decreased CFU counts in the lungs (4.55±0.86 log10 CFU/g of tissue), compared with normal controls (5.23±0.29 log10 CFU/g of tissue). No differences were noted in the dissemination to liver (Figure 1B) and spleen (data not shown). Although at the 6th week both groups of mice showed equivalent pulmonary fungal loads, at week 10 postinfection the iNOS−/− mice presented augmented fungal burden in the lungs (5.99±0.79 log10 CFU/g of tissue) relative to WT mice (4.64±0.80 log10 CFU/g of tissue). Again, no differences in fungal dissemination were observed. These data clearly showed the opposite temporal effect of NO: early in the infection, the absence of nitric oxide synthesis resulted in a protective effect, but at the chronic phase led to a more severe disease (Figure 1A).
Because the differences in pulmonary fungal burdens of iNOS−/− and WT mice were detected at the 2nd and the 10th weeks postinfection, these periods were chosen to next determine the levels of type 1 (IFN-γ, TNF-α, IL-2 and IL-12) and type 2 (IL-4, IL-5 and IL-10) cytokines. As shown in Figure 1C, type 1 and type 2 cytokines were present in the lungs of both studied groups, but at the early phase of infection only TNF-α appeared in significantly higher levels in NO-deficient mice. Later in the infection, however, no differences in pulmonary cytokines were detected (Figure 1D). These data suggested that TNF-α could be involved in the early immunoprotection conferred by NO deficiency.
We have also evaluated the presence and the activation profile of leukocytes in the lungs from both mouse strains. A higher number of mononuclear phagocytes expressing activation molecules was detected in iNOS−/− mice when compared with WT mice. As shown in Figure 2A, at the 2nd week, the number of double positive CD11b+CD86+ and CD11b+CD40+ cells was higher in iNOS−/− mice. Further phenotypic characterization of macrophages (CD11bhighCD80high) and dendritic cells (CD11chighCD86high) demonstrated the increased presence of macrophages in the deficient mouse strain, while the number of dendritic cells was equivalent in both studied groups (Figure 2A). By week 10, no significant differences in the number of pulmonary macrophages were detected (Figure 2B).
In addition, to determine the lymphocyte influx and the activation profile of CD4+ and CD8+ T cells in the lungs of P. brasiliensis infected mice, we evaluated the expression of CD69 and CD25 by freshly isolated T cells. The marker CD69 is a very early activation antigen [41] as well as CD25, the α-chain of the interleukin-2 receptor [42], which is rapidly upregulated on activated T cells. Compared with the control group, at the 2nd week, NO-deficient mice presented an increased recruitment of activated CD4+CD69+, CD4+CD25+ and CD8+CD69+ T lymphocytes to the lungs (Figure 2C). Although in lower intensity, at the chronic phase, the recruitment/expansion of CD4+CD69+ and CD4+CD25+ T cells to the lungs of iNOS−/− mice remained higher than in WT mice. The same was not verified for CD8+CD69+ T cells that appeared in equivalent numbers in both mice strains (Figure 2D). Because Treg cells control the expansion of effector T cells, and the number and function of these cells were shown to be influenced by NO production [43], [44] we characterized the presence of CD4+CD25+FoxP3+ T cells in the pulmonary cell infiltrates (Figure 2C and 2D). Although at the week 2, no differences in the numbers of Treg cells were seen, by week 10 increased numbers of CD4+CD25+FoxP3+ cells were detected in iNOS−/− mice. This finding paralleled the increased pulmonary CFU counts, the diminished number of T cells and the impaired macrophage activation detected in the lungs of iNOS−/− mice at this late period of the infection.
After characterizing the main features of the infection, we aimed to clarify the mechanisms involved in the early immunoprotection conferred by iNOS deficiency. Thus, the behavior of alveolar macrophages was assessed at two opposed periods of infection. In agreement with lung CFU data, at the 2nd week a lower number of yeasts was recovered from the bronchoalveolar lavage fluid of iNOS−/− mice (Figure 3A). The microbicidal activity of alveolar macrophages was further determined after 48 h cultivation in the presence or absence of IFN-γ. Again, lower CFU counts were recovered from deficient macrophages (Figure 3B). As expected, only WT macrophages showed decreased fungal counts associated with increased NO production after IFN-γ treatment (Figure 3B and 3C, respectively). At week 10, iNOS-deficient macrophages showed higher fungal loads than WT cells, which increased their fungicidal ability after IFN-γ treatment (Figure 3D and 3E).
Studies with alveolar macrophages raised two important questions. Were the decreased fungal loads of iNOS−/− macrophages due to their increased fungicidal activity or decreased phagocytic ability? Why are iNOS−/− macrophages refractory to IFN-γ activation? Therefore, inflammatory peritoneal macrophages were used to better understand the behavior of iNOS-deficient phagocytes. Peritoneal macrophages were obtained, activated or not by IFN-γ, TNF-α or both cytokines and infected by P. brasiliensis. Some activated and non-activated cells were also treated with 1MT, a specific inhibitor o 2,3 indoleamine dioxygenase, an enzyme that catalyzes the degradation of tryptophan along the kynurenine pathway. In fungal infections, this enzyme was shown to exert an efficient fungicidal activity but also an important suppressive effect on the immune response [45]. Recapitulating the results obtained with alveolar macrophages, lower numbers of viable yeasts were recovered from iNOS−/− macrophages (Fig. 4A). Only WT cells increased their fungicidal ability when activated by IFN-γ, although TNF-α was not able to modify the fungicidal activity of cells from both mouse strains. Similar result was observed with 1MT-treated macrophages. The IDO inhibitor was not able to modify the behavior of activated and non-activated macrophages from WT and iNOS−/− mice (Fig. 4A).
We have then used FITC-labeled P. brasiliensis yeasts to discriminate adherent and ingested cells. As shown in Fig. 4B, iNOS−/− macrophages have reduced capacity to adhere/ingest fungal cell. When the green fluorescence of adhered yeasts was quenched by trypan blue treatment, a decreased ingestion activity was observed with WT and iNOS-deficient macrophages, although more evident with the latter cells. Thus, iNOS−/− macrophages have a reduced ability to adhere and ingest fungal cells, and this behavior appears to explain the reduced CFU counts detected in vitro and in vivo with iNOS−/− cells.
To better characterize the differentiation of iNOS−/− and WT macrophages, the expression of iNOS, ARG1, TGF-β, IL-12, TNF-α and IDO mRNA was measured in uninfected and P. brasiliensis infected cells (Fig. 5). iNOS-deficient macrophages showed reduced expression of IL-12 associated with increased production of ARG1 and TGF-β mRNA, demonstrating a prevalent anti-inflammatory behavior and some characteristics of M2, healing, or alternatively activated macrophages. An opposite result was observed with WT macrophages which expressed high levels of iNOS and IL-12 mRNA. Interestingly, no differences in IDO expression were detected between WT and iNOS−/− cells, but the latter showed increased expression of TNF-α.
The delayed type hypersensitivity (DTH) reactions developed by iNOS−/− and WT mice were evaluated at weeks 2 and 10 after infection using a soluble P.brasiliensis antigen. As depicted in Figure 6A, both mouse strains developed increased footpad reactions, but these were significantly higher in iNOS−/− than in WT mice. To better typify the effect of iNOS deficiency in the severity of pulmonary PCM, histopathological analysis of lung sections from iNOS−/− and WT mice at both periods of infection was also performed. By week 2, iNOS−/− and WT mice showed diffuse inflammatory reactions composed of monocytes and lymphocytes surrounding interlobular spaces localized around the bronchi, bronchioles and blood vessels. However, an increased presence of inflammatory cells accompanied the lower CFU counts observed in the lungs of iNOS−/− mice (data not shown). Surprisingly, at week 10 remarkable histopathological differences were detected; iNOS−/− mice presented a large number of well-organized granulomas (Figure 6B, lower panel) containing an elevated number of yeast cells surrounded by epithelioid and multinuclear giant cells, and a well-defined lymphocytic mantle. Plasma cells and eosinophils were scarce. The fungi, detected in great numbers, were large and have multiple buds. Compared with iNOS−/− mice, WT mice presented more extensive, non-organized lesions, and decreased fungal loads irregularly distributed in the lung parenchyma (Figure 6B, upper panel). Importantly, despite the higher number of yeasts recovered late in the infection, iNOS−/− mice showed decreased mortality rates (Figure 6C). Thus, the increased T cell immunity as evidenced by increased DTH reactions, and the more organized lesions (possibly mediated by the increased TNF-α secretion) appear to have compensated the poor control of fungal multiplication resulting from iNOS deficiency
Supported by the remarkable increase of TNF-α levels observed early in the infection in the lungs of iNOS−/− mice, and to further understand the mechanisms of immunoprotection used by this mouse strain to control P.brasiliensis infection, iNOS−/− and WT groups were in vivo depleted of TNF-α and the severity of infection analyzed by CFU counts and pulmonary inflammation. At week 1 after infection, a significant difference in fungal burdens were observed between IgG-treated WT and iNOS−/− mice. The anti-TNF-α treatment, however, abolished this difference (Figure 7A). Anti-TNF-α treatment was shown to be effective, since reduced levels of this cytokine were detected in the treated groups of both mouse strains (Figure 7B). In addition, in both mouse strains at weeks 2 and 8 after infection TNF-α depletion did not result in significant increases in pulmonary fungal burdens (Figure 7C, D). Yet, no significant differences were noticed in the dissemination to liver and spleen. Cytokines measurements in lung homogenates at week 2 after infection demonstrated that, although anti-TNF-α treatment remained until day 12 postinfection, TNF-α rapidly reached the pre-treatment levels in the liver and lungs of both mouse strains (Figure 7E and 7F). On the other hand, a higher concentration of TGF-β was detected in the lungs, while decreased levels of hepatic IFN-γ were seen in anti-TNF-α treated iNOS−/− mice (Figure 7E, F).
Unexpectedly, when lung infiltrating leukocytes were characterized at week 2 postinfection, an impressive increase in the numbers of lymphocytes (Figure 8A) and mononuclear phagocytes (Figure 8B) were seen in the lungs of anti-TNF-α-treated mice. Thus, a higher influx of activated CD4+CD69+, CD8+CD69+ T cells into the lungs of TNF-depleted WT and iNOS−/− mice was observed, although CD4+CD25+ T appeared in higher numbers only in the former strain (Figure 8A). In addition, compared with IgG-treated controls, iNOS−/− and WT-depleted groups showed elevated numbers of CD11b+CD80+ and CD11b+CD40+ mononuclear phagocytes (Figure 8B).
To further characterize the role of TNF-α in our model, TNF-α-depleted and IgG-treated controls were studied regarding mortality and histopathology of lungs. As depicted in Figure 8C, the effect of TNF-α neutralization was much more prominent in the iNOS-deficient strain. Indeed, 100% of TNF-α depleted iNOS−/− mice died within 70 days of infection while only 25% of WT mice died in the same period. This result demonstrated that, at least partially, the relative protection of NO-deficient mice was due to the enhanced TNF-α production. In vivo neutralization of TNF-α did not appear to cause clear alterations in the already poor-organized and confluent lesions of WT mice (Figure 8D, left panels). As described for untreated-infected mice, IgG-treated iNOS−/− mice presented better-defined granulomas, and a high influx of inflammatory cells, which appears to restrain fungal spreading (Figure 8D, upper right panel). In contrast, anti-TNF-treated iNOS−/− mice lose this organized pattern of lesions and non-organized, confluent inflammatory reactions containing fungal cells were scattered through the pulmonary tissue (Figure 8D, lower right panel).
The lower pulmonary fungal growth detected at week 2 postinfection of iNOS−/− mice paralleled the enhanced presence of inflammatory cells (Figures 1A and 2C). Importantly, among T cells, only the CD8+ T cell subset lost their activation profile at week 10 of infection when iNOS−/− mice were unable to control of fungal growth. In order to study the involvement of CD8+ T cells in the early immunoprotection of NO-deficient mice, CD8α+ cells were in vivo depleted by monoclonal antibodies, and the severity of infection characterized at week 2 postinfection. As depicted in Figure 9A, anti-CD8 treatment abolished the differences in CFU counts previously observed in the lungs of mice. Moreover, this treatment also abolished the higher presence of activated T cells and macrophages observed in the lungs of iNOS-deficient mice at this early period of infection. Compared with IgG-treated controls, CD8-depleted iNOS−/− mice presented significantly diminished influx of CD4+CD69+ and CD8+CD69+ T cells to the lungs (Figure 9B). The same occurred with CD11b+CD80+, CD11b+CD40+ and CD11b+CD86+ mononuclear phagocytes (Figure 9C).
Subsequently, we asked whether this difference in cell influx caused by anti-CD8α treatment was due to the depletion of the CD8α+ subset of dendritic cells. Thus, IgG-treated and anti-CD8-depleted WT and iNOS−/− groups were i.t. infected and at the second week post-infection the lungs were removed, digested and pulmonary dendritic cells obtained with anti-CD11c magnetic beads. The phenotype of these cells was then assessed by flow cytometry. The Figure 9D demonstrates that no differences in the numbers of conventional (CD11chighCD11blow), lymphoid (CD11c+CD8α+) and plasmacytoid (CD11c+B220+) DCs were detected between anti-CD8-treated and untreated mice of both mouse strains. In summary, this result suggested that the effect of anti-CD8 treatment was mainly mediated by CD8α+ T lymphocytes and not lymphoid DCs.
Next, we characterized the phenotype of IFN-γ, TNF-α and IL-4-secreting lymphocytes in the lungs of anti-CD8-treated and IgG-treated iNOS−/− and WT mice by intracellular cytokine staining. Comparing IgG-treated controls, enhanced numbers of TNF-α + CD4+ and CD8+ T cells were seen in the lungs of iNOS−/− than in the WT group (Figure 10). Anti-CD8 treatment led to increased numbers of TNF-α+, IFN-γ+ and IL-4+ CD4+ T cells in WT mice. However, reduced numbers of TNF-α+ and IFN-γ+ CD8+ T cells appeared in the lungs of this strain. Importantly, in iNOS−/− mice the depletion of CD8+ cells resulted in a remarkable reduction of TNF-α+ and IFN-γ+ CD8+ T cells, besides a decreased presence of IFN-γ+ CD4+ T cells (Figure 10). Therefore, at the 2nd week of P. brasiliensis infection, the reduction of pro-inflammatory CD8+ T cells observed in WT mice appeared to be compensated by the increased presence of TNF-α- and IFN-γ-secreting CD4+ T cells whereas in iNOS-deficient mice a prevalent reduction of pro-inflammatory CD4+ and CD8+ T cells was seen.
In the present work, we investigated the temporal significance of nitric oxide synthesis in the evolution of pulmonary paracoccidioidomycosis and the immunopathological mechanisms associated with iNOS deficiency. Early in infection, the protective effects of iNOS deficiency was associated with decreased fungal burdens, enhanced secretion of TNF-α augmented DTH reactions, and increased migration of activated T cells and macrophages to the lungs, which subsequently organize as well-compact granulomas. On the other hand, at later periods, increased fungal loads were concomitant with sustained T cell immunity allied with increased presence of regulatory T cells at the site of infection. Unexpectedly, the mortality rates of WT mice were higher than those of iNOS−/− mice. In vivo depletion of TNF-α and CD8+ T lymphocytes demonstrated a division of labor carried out by these two components in the immunoprotection developed by iNOS−/− mice. While upregulated TNF-α secretion avoided precocious mortality and organized pulmonary lesions, the increased expansion of CD8+ T cells controlled fungal growth and secretion of pro-inflammatory cytokines. Both, TNF-α and CD8+ T cells were involved in the enhanced recruitment of inflammatory cells to the lungs. This protective effect was persistent, but excessive inflammatory reactions were possibly controlled by the increased expansion of Treg cells at late stages of immunity developed by NOS−/− mice. Altogether, these mechanisms appear to confer sustained protection to iNOS−/− mice, which, despite the elevated fungal loads, presented increased survival time and better disease outcome.
In our model, we could verify that iNOS deficiency seems to be compensated by the deviation of the immune response to a more pronounced Th1 pattern. The lower fungal loads were concomitant with the high levels of pulmonary TNF-α produced by P.brasiliensis infected NO-deficient mice at the 2nd week of infection. Interestingly, it has been reported that low concentrations of NO enhanced Th1 immunity by increasing the expression of IL-12 receptor in T cells although high NO concentrations are cytotoxic [44], [46]. Interestingly, alternative mechanisms for the immunosuppressive activity of NO production have recently been described. It was demonstrated that NO suppresses NALP3 inflammasome activation by nitrosylation of NALP3 proteins resulting in decreased synthesis of mature IL-1 beta and IL-18, impaired Th1 immunity and NK cell activation [47]. Considering this information, it is tempting to speculate that the enhanced Th1 immunity developed by iNOS-deficient mice could be associated with increased NALP3 inflammasome activity and enhanced IL-1/IL-18 production. Consistent with a prevalent Th1 pattern of immunity, by week 2, increased levels of IFN-γ-regulated isotypes (IgG2a and IgG3), were produced by iNOS−/− mice (data not shown). Thus, these data are consistent with the benign forms of murine and human PCM, which are associated with prevalent type-1 immunity [21]. Allied with type-1 cytokine production, an increased number of activated lymphocytes and macrophages was found in the lungs of iNOS-deficient mice suggesting the development of enhanced DTH reactions at the site of infection.
iNOS-deficient alveolar macrophages showed fungal loads equivalent to those detected in lung tissue and did not present an increased fungal ability when activated by IFN-γ. Several investigations showed the fundamental role of NO in P. brasiliensis killing by IFN-γ activated macrophages [2], [23], although NO-independent mechanisms were reported to be associated with TNF-α-activated human and murine macrophages [48], [49]. Our studies with inflammatory macrophages confirmed that iNOS−/− macrophages are refractory to IFN-γ and TNF-α activation. Furthermore, the reduced CFU numbers displayed by these cells could not be ascribed to increased IDO expression, because the CFU counts were not modified by 1MT treatment. Importantly, iNOS−/− macrophages were also shown to have a decreased ability to adhere to and ingest yeast cells, which possibly explains the low CFU numbers detected both in the in vitro and in vivo infections. Absence of iNOS expression was also associated with an M2-like behavior of macrophages. Indeed, these cells expressed high levels of arginase-1 and TGF-β and were associated with low levels of IL-12. This anti-inflammatory behavior appears to explain why iNOS-deficient cells were not activated by IFN-γ and TNF-α and explains why the iNOS-deficient cells do not express elevated levels of IDO, an enzyme primarily induced by IFN-γ activation [45]. However, iNOS-deficient cells expressed high levels of TNF-α, which is potentially associated with an efficient activation of dendritic cells and increased migration of inflammatory cells to the site of infection.
Interestingly, the behavior of iNOS−/− macrophages is similar to the behavior of macrophages and dendritic cells from mice (strain A/J) resistant to P. brasiliensis infection [27], [35]. A/J cells are poorly activated by IFN-γ and IL-12 and show an impaired NO production and killing ability. Concomitant with elevated production of TGF-β, A/J macrophages produce high levels of TNF-α that contribute to the late but consistent cellular immunity and immunoprotection developed by this mouse strain [27], [35], [50].
Our data have also demonstrated that in pulmonary PCM, early NO production inhibits the activation and migration of CD4+ and CD8+ T cells to the site of infection. This finding was not previously described in murine PCM, but was reported in other experimental models where less severe infections, mainly due to increased IFN-γ production and CD4+ Th1-skewed immune responses, were observed in iNOS−/− mice [4], [44], [46], [51], [52]. The increased influx of T cells and macrophages was concomitant with elevated levels of pulmonary TNF-α, a proinflammatory cytokine that enhances the maturation of antigen presenting cells and induces increased expression of adhesion molecules on endothelial cells [53]–[55]. Thus, the enhanced secretion of TNF-α appeared to have amplified the afferent and the efferent phases of immunity, by increasing T cell sensitization and further migration to the site of P. brasiliensis infection.
Despite the less severe infection at the 2nd week, iNOS−/− mice were not able to sustain this behavior, and elevated fungal burdens were seen in their lungs at later periods. Despite the yet significantly increased presence of activated CD4+ T cells in iNOS−/− mice, no differences in the number or activation of CD8+ T cells were detected. This decreased influx of effector CD8+ T cells was parallel to the decreased activation of macrophages, indicating, as we previously demonstrated, the important role of this T cell subset in the immunoprotection of pulmonary PCM [21], [56].
Our studies appear to indicate that at an early phase of infection NO does not affect the expansion of CD4+CD25+Foxp3+ regulatory T cells. However, late in infection (week 10), iNOS deficiency supported the expansion of this regulatory T cell subset. These findings suggest that absence of NO production led to an early enhanced T cell immunity, but the excessive lung inflammation was lately avoided by increased expansion of Treg cells. The M2-like profile of inflammatory macrophages appears to have contributed to the expansion of Treg cells and the controlled tissue pathology observed in iNOS−/− mice. This Treg-associated mechanism controlling effector immunity was not previously described for the NO-induced immunosuppression in murine PCM, and can be added to the regulatory mechanisms mediated by unbalanced NO production. Moreover, our data on the presence of CD4+CD25+Foxp3+ Treg cells at the site of infection are in good agreement with ours [57]–[59], and others [60], [61] reports showing a late enhancement of Foxp3+ Treg cells associated with increased immunological responses and pathogen burden.
Interestingly, the histopathological examination of lungs was consistent with the important role of TNF-α in the organization of granulomatous lesions [62], [64]. Indeed, the high influx of inflammatory T cells and macrophages into the lungs of iNOS−/− mice, associated with the locally increased levels of TNF-α resulted in more organized lesions at week 10, which appeared to have overridden the elevated fungal loads due to the lack of NO synthesis. Furthermore, this pattern of lesion organization appeared to be able to restrain fungal dissemination to distant organs, since despite the lately increased pulmonary fungal burdens, no differences were detected in the liver and spleen of iNOS−/− mice. The increased survival time of iNOS-deficient mice appears to underline the possible protective effect of well-organized lesions.
It is known that arginine catabolism is mediated by two types of enzymes: the nitric oxide synthases convert arginine to citrulline and NO, while the arginases hydrolyze arginine to urea and ornithine. The latter component is necessary for the production of proline and is essential for the synthesis of collagen, which is the main component of the extracellular matrix (ECM) observed in granulomatous lesions [65]. As here shown, iNOS-deficient macrophages only express arginase-1, which may have contributed to the expression of ECM components and the compact organization of granulomas observed in iNOS-deficient mice. It is also known that NO has a modulatory effect on the expression and activity of zinc-dependent metalloproteases (MMPs), which decrease the deposition and accumulation of extracellular matrix (ECM) proteins [66]. Some studies have demonstrated that NO induces the activation of MMP2 and MMP9, which are two enzymes that have an anti-fibrotic effects through the degradation of ECM proteins (laminin, collagen, elastin, etc.) and pro-cytokines involved in fibroblast activation and granuloma organization (e.g., pro-TNF-α and pro- TGF-β) [66], [67], [68]. In the intraperitoneal model of PCM using WT and iNOS−/− mice [69], NO production was associated with increased MMP9 activity and the loose organization of granulomas developed by WT mice. In the pulmonary model employed in the present study we suppose that the NO produced by the WT mice inhibited the production of TNF-α, which is required for granuloma organization, and of TGF-β, which is needed for tissue repair and Treg cell expansion. Consistent with this possibility, infected iNOS−/− macrophages expressed elevated levels of TGF-β (Fig. 5), which is a cytokine involved in fibroblast activation, enhanced synthesis of ECM, and Treg cell differentiation. The TGF-β-induced Treg cell differentiation apparently contributed to the controlled tissue pathology even in the presence of the augmented immune response of the iNOS−/− mice. The study by Nishikako et al. [69] revealed additional important information on the influence of NO in granuloma formation and disease outcome. At late stages of infection (120 days post-infection, not evaluated in our model), i.p. infected iNOS−/− mice presented decreased fungal loads in their well-organized granulomas, which demonstrated the efficiency of their inflammatory reactions. Furthermore, as observed here, iNOS−/− mice showed decreased mortality rates when compared with WT mice.
In vivo depletion of TNF-α abrogated important advantages conferred by NO deficiency since only the iNOS−/− strain showed precocious mortality rates associated with non-organized pulmonary lesions. Importantly, early after interrupting anti-TNF treatment, the levels of TNF-α returned to normal levels and a massive influx of activated T cells and macrophages into the lungs occurred in both mouse strains. Therefore, it became clear that besides mortality rates and organization of lesions, TNF-α also controlled the migration of inflammatory cells to the site of infection. As a whole, these results led us to demonstrate that the concomitant deficiency of NO and TNF-α is fatal to iNOS-deficient mice, whereas the still preserved ability of NO synthesis by TNF-depleted WT mice appeared to rescue this mouse strain from precocious mortality.
Depletion experiments of CD8+ T cells revealed the important role of this T cell subset in the early immunoprotection of iNOS−/− mice. Thus, the early differences in fungal loads were abrogated, and the influx of inflammatory cells was markedly impaired only in CD8-depleted mice iNOS−/− mice. Studies on the phenotype of DCs at the site of infection showed that anti-CD8 treatment did not alter the presence of CD8+ lymphoid DCs, suggesting that CD8+ T lymphocytes, and not lymphoid DCs, played an important control of fungal growth and inflammatory reactions mediated by T cells and macrophages. We could verify by intracellular cytokine staining that in iNOS−/− mice, the depletion of CD8+ T cells resulted in decreased numbers of CD4+ (IFN-γ) and CD8+ (IFN-γ and TNF-α) T cells, supporting the proinflammatory feature of this T cell subpopulation. This fact was consistent with the diminished influx of inflammatory cells observed in the lungs of CD8-depleted iNOS−/− mice. In WT mice, however, the depletion of CD8+ cells had a negligible effect, and this appears to reflect the concomitant increase of IFN-γ and TNF-α CD4+ T cells with decreased numbers of CD8+ T cells secreting the same proinflammatory cytokines in the lungs. The increased presence of IL-4+ CD4+ T cells appeared to have exerted a negligible effect in the inflammation mounted by WT mice.
In conclusion, this work brought new information regarding the role of NO synthesis in experimental PCM. We demonstrated the protective effect of iNOS deficiency in pulmonary PCM. This protective effect appeared to be mediated by increased type-1 inflammatory reactions regulated by TNF-α production and expansion of IFN-γ and TNF-α-producing T cells.
|
10.1371/journal.pgen.1004048 | p53 and TAp63 Promote Keratinocyte Proliferation and Differentiation in Breeding Tubercles of the Zebrafish | p63 is a multi-isoform member of the p53 family of transcription factors. There is compelling genetic evidence that ΔNp63 isoforms are needed for keratinocyte proliferation and stemness in the developing vertebrate epidermis. However, the role of TAp63 isoforms is not fully understood, and TAp63 knockout mice display normal epidermal development. Here, we show that zebrafish mutants specifically lacking TAp63 isoforms, or p53, display compromised development of breeding tubercles, epidermal appendages which according to our analyses display more advanced stratification and keratinization than regular epidermis, including continuous desquamation and renewal of superficial cells by derivatives of basal keratinocytes. Defects are further enhanced in TAp63/p53 double mutants, pointing to partially redundant roles of the two related factors. Molecular analyses, treatments with chemical inhibitors and epistasis studies further reveal the existence of a linear TAp63/p53->Notch->caspase 3 pathway required both for enhanced proliferation of keratinocytes at the base of the tubercles and their subsequent differentiation in upper layers. Together, these studies identify the zebrafish breeding tubercles as specific epidermal structures sharing crucial features with the cornified mammalian epidermis. In addition, they unravel essential roles of TAp63 and p53 to promote both keratinocyte proliferation and their terminal differentiation by promoting Notch signalling and caspase 3 activity, ensuring formation and proper homeostasis of this self-renewing stratified epithelium.
| The mammalian epidermis is a stratified self-renewing epithelium, in which cell loss at the surface is properly balanced by cell proliferation in basal layers to ensure tissue homeostasis. But how is this balance genetically controlled? Here, we address this question in zebrafish breeding tubercles, epidermal appendages in which keratinocytes undergo more advanced differentiation processes than in regular fish epidermis, sharing crucial features with the cornified mammalian skin. We identify a linear pathway consisting of the transcription factor p53 and its close relative TAp63, which activate Notch signalling and thereby caspase 3 to promote terminal differentiation and eventual shedding of keratinocytes in upper tubercle layers, while at the same time employing non-cell autonomous mechanisms to promote keratinocyte proliferation at the tubercle base, thereby ensuring proper development and homeostasis of this self-renewing tissue. Such a two-fold function of the pathway is consistent with the formerly reported dual role of a caspase during wing regeneration in the fruitfly. Our findings will help to better understand the seemingly contrary effects described for TAp63 in different mammalian systems, while demonstrating partial functional redundancy between p53 and TAp63 during epidermal development in fish.
| The mammalian epidermis is a self-renewing stratified epithelium on the outer surface of the skin. During embryogenesis, it develops from the surface ectoderm, which is initially a single-layered epithelium. Stratification is initiated (E10 in mouse) with the formation of the outer periderm, leading to a bi-layered epidermal organization in which peridermal cells are attached to each other via tight junctions to protect the embryo against the amniotic fluid [1], [2]. Further epidermal maturation (E12.5–E17.5 in mouse) takes place beneath with the consecutive formation of the spinous, granular and cornified layers, establishing the later epidermal barrier, while the periderm is sloughed off [3]–[6]. Crucial contributions to this later epidermal barrier come from the granular layer, in which cells are sealed to each other via tight junctions, and from the outer cornified layers, providing physical resistance and preventing dehydration. Formation of this water barrier is essential for the adaptation to terrestrial life both during the ontogeny [7] and the evolution [8], [9] of land-based vertebrates. Accordingly, fish lack epidermal cornification.
In zebrafish, the embryonic and larval epidermis is bi-layered, consisting of an outer enveloping layer (EVL) segregating from inner cells before surface ectoderm specification [10], [11], and a basal keratinocyte layer, resembling the bi-layered organization of the mammalian epidermis at midgestation stages. Further stratification of the zebrafish epidermis only commences with the onset of metamorphosis (after three weeks of development), during which the epidermis becomes multi-layered. However, keratinocytes remain metabolically active throughout, including the outer-most layer, and lack morphological signs of cornification [12], [13]. This is in contrast to the organization of breeding tubercles, contact and secondary sex organs on the head and pectoral fin rays. Based on morphological studies in several other fish species, they were described as “keratinized epidermal appendages covered by a layer of dead cells with altered content” [14], suggesting that in breeding tubercles, keratinocytes might undergo more advanced, cornification-like differentiation processes.
Mammalian keratinocyte cornification is a multi-step process initiated by a switch in the expression of particular keratin genes, followed by the expression of the keratin-bundling protein filaggrin and proteins such as involucrin and loricrin, which together with keratins become cross-linked by transglutaminases (Tgm1-4) to reinforce the formation of a cornified envelope [15]–[17]. Furthermore, lipids stored in lamellar bodies are extruded into the extracellular space to form a lipid envelope. During these later steps, cells enter an apoptotic-like phase, lose cytoplasmic organelles including the nucleus, and are eventually sloughed from the skin surface. This loss of corneocytes by desquamation is tightly balanced by keratinocyte proliferation in basal layers, allowing constant self-renewal during epidermal homeostasis.
A key regulator of mammalian epidermal stratification and keratinocyte proliferation and differentiation is p63, a homolog of the tumour suppressor and transcription factor p53 [18]–[24]. Use of alternative promoters of the Tp63 gene gives rise to two isoform categories: those with an N-terminal transactivation domain (TAp63 isoforms), as also present in p53, and those lacking this domain (ΔNp63 isoforms) and acting as inhibitory competitors of their TA counterparts. In addition, alternative splicing gives rise to at least three different C-terminal isoforms (α,β,γ) in each category, and to different N-terminal isoforms of TAp63 [25]. Recent analyses of mice specifically lacking the ΔN isoforms indicate that ΔNp63 is required for maintaining the proliferative potential of basal keratinocytes in embryonic epidermis while preventing their premature entry into terminal differentiation [26], consistent with findings obtained in cultured keratinocytes [27]–[29]. In contrast, TAp63 and p53 might promote keratinocyte differentiation. Thus, in inducible cell lines, TAp63 activates genes involved in keratinocyte differentiation, including different tgms [30]. Also p53, via direct and indirect activation of Notch signalling, has been shown to be required for keratinocyte differentiation in cell culture systems and during squamous cell carcinoma suppression in mouse tumour models [31], [32]. siRNA-mediated knockdown studies in organotypic cultures of human keratinocytes further suggest that it is primarily p53 that antagonizes the proliferation-stimulating effect of ΔNp63, and that the contribution of TAp63 to keratinocyte differentiation is minor compared to that of ΔNp63 [29], [33]. Furthermore, TA-specific p63 mutant mice lack an abrogation of keratinocyte differentiation [34]–[36], leaving the in vivo role of TAp63 during keratinocyte development unclear.
Notch signalling promotes different steps of mammalian keratinocyte differentiation in vivo [37]–[41]. Upon binding of Delta or Jagged ligands, the Notch receptor is cleaved and releases its intracellular domain (NICD), which binds to the transcriptional repressor RBP-J, enabling it to activate target genes that are repressed in the absence of Notch signals. Findings concerning the interconnection between p53/p63 and Notch signalling are controversial. Thus, Jagged1/2 and Notch1 have been reported to be positively regulated by p53 and TAp63γ [31], [42]–[46], but also by ΔNp63 [26], [43], while in other systems, the Notch target Hes1 was negatively regulated by both TAp63 and ΔNp63 [43]. How Notch signalling promotes keratinocyte differentiation is also incompletely understood [38], [47]. One described Notch/RBP-target gene in embryonic mouse keratinocytes encodes caspase 3 [39]. This cysteine protease not only executes apoptosis, but also promotes terminal differentiation processes in a range of cell types [48], leading to delayed basal keratinocyte differentiation in caspase 3 mutant mouse embryos [39].
Here, we identify the zebrafish breeding tubercles as sites with higher keratinocyte proliferation in basal layers as well as more advanced, cornification-like keratinocyte differentiation in upper layers, including transglutaminase expression and stronger keratinization, exclusive presence of tight junctions in second-tier keratinocytes, and rudimentary lipid envelope formation and constant desquamation and renewal of surface keratinocytes. In addition, we identify zebrafish TAp63 isoforms and a TA-specific p63 loss-of-function mutant, revealing essential and partially redundant roles of TAp63 and p53 to promote both keratinocyte proliferation at the base and terminal keratinocyte differentiation in upper layers of breeding tubercles. Both effects are mediated via Notch signalling and activated caspase 3, although these mediators are restricted to upper breeding tubercle layers, pointing to a combination of cell autonomous and non-cell autonomous effects. These findings will help to better understand the seemingly controversial roles described for TAp63 and p53 in different systems.
Even after metamorphosis (approximately 30 days post fertilization; dpf), when the zebrafish epidermis has become multi-layered, superficial epidermal cells display crucial similarities to the embryonic enveloping layer (EVL), like the presence of tight junctions and distinct microridges at their outer surface [12]. To determine the developmental origin of superficial cells of the adult zebrafish epidermis, we carried out transgenic lineage tracing experiments, using promoter elements that drive embryonic expression confined to the outer EVL (krt4) [49] or the basal layer (krt19) [50], respectively (Figure 1A; 5 dpf). Employing these promoters in an inducible binary transgenic Cre/Lox-system, embryonic EVL cells or basal keratinocytes were stably labeled via tamoxifen application from 1–4 dpf (Figure 1B–F). At 60 dpf, superficial cells of the regular body epidermis consisted of a mixture of derivates of embryonic EVL cells (Figure 1B) and basal keratinocytes (Figure 1C,D). However, both expressed the same marker genes (Figure 1B–D; and data not shown) and displayed identical morphological characteristics (Figure 1E). This suggests that in regular epidermis, embryonic EVL cells can persist beyond metamorphosis and are only slowly replaced by derivatives of basal keratinocytes. However, renewed superficial cells seem to have the same properties and functions as persisting EVL cells.
A different pattern was observed in spike-like epidermal structures that according to their location and fine structure (see below) were identified as breeding tubercles [14]. Here, superficial cells consisted solely of derivatives of basal keratinocytes and failed to express markers of the embryonic EVL and of the superficial cells of the regular epidermis (Figure 1D; n = 13/13), suggesting that they have been renewed completely and have acquired a different fate.
In adult zebrafish, breeding tubercles were exclusively present in a disc- and row-like structure on each side of the lower jaw (Figure 2A), as well as in rows along the bony rays of the pectoral fins of males (Figure 2B), but not females (Figure 2C). Scanning electron microscopy (SEM) revealed that superficial cells of breeding tubercles, also called cap cells [14], lacked the microridges present at the outer surface of regular epidermis (Figure 2D,E). In addition, they lacked expression of tg(krt4:GFP) (Figure 2G) and other specific markers of superficial cells of regular epidermis (data not shown), and displayed a different cytoplasmic composition in AFOG trichrome stainings (Figure 2F). Furthermore, immunofluorescence revealed a higher keratin content of tubercle keratinocytes compared to regular epidermis (Figure 2H). Correspondingly, expression levels of the type II keratin gene krt8 were significantly higher in breeding tubercles than in regular epidermis (Figure 2I). Furthermore, the type I keratins krtt1c11a (ZFIN: zgc:136902) and krt17 (ZFIN: zgc:92061) were exclusively expressed in lower tubercle layers, but not in regular epidermis (Figure 2J and data not shown), while expression of type II keratin krt5 and type I cytokeratin cki [51] was shared by regular epidermis and lower tubercle layers, but absent in upper layers of the tubercles (Figure 2K and data not shown). Strikingly, tgm1 [52], encoding the cross-linking enzyme transglutaminase 1 involved in cornification of the mammalian epidermis [53], was exclusively expressed in upper layers of the breeding tubercles (Figure 2L).
Striking differences were also observed at the structural level. In regular epidermis, basal and intermediary keratinocytes were of similar shapes and organized in a rather irregular pattern. In contrast, basal cells of tubercles were more regularly aligned to each other, while intermediary cells displayed a looser and spinous-like organization (Figures 2F and 3A,B), connected to each other via local desmosomes, but with wide extracellular spaces in between (Figure 3C). Tight junctions were only found in the second tier layer directly beneath the cap layer (Figure 3D), similar to their confinement to granular cells beneath the cornified layers in mammals. Desmosomes were also present between second tier and outer cap cells. However, in many samples, they were in the process of regression, leading to a partial dissociation of the two layers (Figure 3E). In addition, cap cells contained large vesicles reminiscent of lamellar bodies [54], extruding their content into the space between cap and second tier cells (Figure 3F), which was filled with material resembling the extracellular lipid lamellae in cornifying layers of the mammalian epidermis [54] (Figure 3G). Gradual loss of desmosomes was also observed between adjacent cap cells, but accompanied by cell membrane deterioration and cellular fusion (Figure 3H). Intracellular, cap and second tier cells contained large amounts of electron-dense granules (Figure 3G,H), while apart from nuclei (see below), cell organelles like mitochondria were largely absent (data not shown).
In sum, these data indicate that breeding tubercles display a more pronounced stratification than regular epidermis, consisting of different layers with distinct morphological properties and some cornification-like features in the superficial layer.
Regular epidermis becomes multi-layered during metamorphosis [12], with final thicknesses between 3 and 10 layers, depending on the position on the body (data not shown). Breeding tubercles develop at the same time. At 21 dpf (6.0–6.5 mm body length) and 24 dpf (6.5–7.0 mm), the future breeding tubercle domain on the lower jaw was still covered by microridge-bearing peridermal cells (Figure 4A,B). At 28 dpf (7.5–9.0 mm), first elevations with microridge-free outer cells were present (Figure 4C), which had acquired the mature spiky shape at 31 dpf (10.0–11.0 mm) (Figure 4D).
Already at early elevation stages, tubercles were observed in which some of the smooth outer cells had been lost, exposing second tier cells with a rougher surface (Figure 4E), possibly reflecting former contact points to the lost outer cells. During later stages, entire sheets of outer cells were lifted up (Figure 4F) and shed. Corresponding spiky discs (Figure 4G–I) or rows were found in the water, in particular after spawning. Confocal analyses after DAPI-staining revealed that the spikes of these sloughed structures were hollow, indicating that they consisted solely of cap layers. To investigate whether shed cap layers are renewed, we took advantage of their unique property to be readily and permissively stained with externally applied dyes like calcein (Figure 4J) or methylene blue, reflecting their loss of cell membrane integrity, a hallmark of cell death [55]. In contrast, second tier and deeper cells remained unstained. In daily analyses of the same fish over several weeks, calcein permeability of individual discs or rows was randomly lost within one day and only regained after 7–14 days (Figure 4K,L; n = 12), suggesting that shed cap cells are replaced by second tier cells and that it takes the latter 7–14 days to terminally differentiate.
In contrast to mammalian corneocytes, cap cells of breeding tubercles still contain their nuclei. However, compared to the nuclei of basal and spinous keratinocytes (Figure 5A,B), their chromatin was strongly condensed (Figure 5C), similar to the pyknosis apparent during cell death [56]. Nevertheless, cap cells were TUNEL (Terminal deoxynucleotidy transferease dUTP nick end labelling)-negative (Figure 5D,E), pointing to the absence of DNA fragmentation, whereas they showed high levels of activated caspase 3 (aCasp3; Figure 5F). This cysteine protease, which does not only execute apoptosis, but also promotes differentiation of embryonic keratinocytes [39], [48], was also present in spinous cells in intermediary tubercle layers (Figure 5F), which had normal nuclei (Figure 5B) and excluded externally applied dyes (Figure 4J), thus lacking all hallmarks of cell death. In contrast, aCasp3 was absent from basal tubercle layers and regular epidermis (Figure 5F). This aCasp3 distribution pattern was largely complementary to that of p63, which was present in all layers of regular epidermis, but confined to lower layers of breeding tubercles (Figure 5G). Cell proliferation, assayed via BrdU incorporation, displayed the same pattern like p63, complementary to aCasp3 (Figure 5H). However, proliferation rates in lower tubercle layers were significantly higher than in regular epidermis, both in fully grown (Figure 5H,J) as well as in developing tubercles (Figure 5I,J).
In sum, in contrast to regular epidermis, where keratinocytes of all layers display comparable p63 and proliferation levels while lacking activated caspase 3, at least two domains can be distinguished in breeding tubercles: aCasp3-positive cells in the upper layers that are postmitotic and undergo more advanced differentiation, and aCasp3-negative keratinocytes at the base, with proliferation rates even higher than in regular epidermis. Furthermore, aCasp3 in upper layers is not correlated with apoptosis.
In light of the described observations, we next studied breeding tubercle development in zebrafish loss-of-function mutants in p53 and TAp63, potential regulators of keratinocyte differentiation and proliferation in mammalian cell culture systems [29], [31]–[33]. The p53zdf1 allele bears an (M214K) exchange of a conserved amino acid residue in the DNA-binding domain that compromises p53 activity [57]. For zebrafish p63, only ΔN isoforms had been described thus far [58], [59]. However, via exon prediction of genomic sequences upstream of the ΔNp63-specific exon (Ensembl accession number ENSDARG00000044356) and validation via RT-PCR analyses and cDNA sequencing, two N-terminal TAp63 isoforms, TA1 and TA4, were identified (Figures 6A and S1) that are similar to the corresponding isoforms in mammals [25] (39% aa identity; 55% aa similarity; Figure S2A,B; GenBank accession numbers KF682365, KF682366). TA1 corresponds to mammalian full-length TAp63, TA4 to Δ40TAp63, which was the initially described human isoform [18]. Furthermore, the exon-intron organization is conserved between mammals and fish (Figure S1), and sequences from shark to human segregate in the expected phylogenetic pattern (Figure S2C).
Both predicted TAp63 isoforms (TA1, TA4) were expressed in adult zebrafish skin (Figure 6B). Comparative regular (Figure 6C) and quantitative real-time RT-PCR (Figure 6D) analyses further revealed almost exclusive presence of ΔNp63 transcripts in early zebrafish embryos, similar to the situation in mouse [60]. At the onset of metamorphosis (20 dpf), TAp63 levels were approximately 5fold increased, but still constituted only 1.3% of the total p63 levels (Figure 6D). Also in adult zebrafish skin, TAp63 was expressed at much lower levels than ΔNp63 (approx. 2%), whereas higher TAp63 transcript levels were found in the ovary, the site of essential TAp63 function in mouse [34] (Figure 6D). In situ hybridizations with an isoform-specific RNA probe further revealed that TAp63 was expressed in regular epidermis as well as in all layers of the breeding tubercles (Figure 6E,F), similar to the expression pattern of p53 (Figure 6G,H).
To investigate whether the zebrafish TAp63 transcripts are translated into biologically functional proteins, we carried out over-expression studies in zebrafish embryos. Injected synthetic mRNA encoding the TA1 or TA4 isoform of zebrafish TAp63γ caused widespread apoptosis during gastrulation stages (Figure 6I,J), leading to embryonic death or severe malformations during further development (Figure 6K), similar to the previously reported effects of injected zebrafish p53 mRNA [61]. In addition, zebrafish TAp63γ mRNA significantly rescued the (headless) phenotype caused by ΔNp63α over-expression (Figure 6L–N), as previously reported for mouse TAp63 and p53 [58]. This indicates that zebrafish TAp63 transcripts give rise to a protein with p53-like pro-apoptotic and ΔNp63-antagonizing activities.
Using target-selected mutagenesis [62], a TA-specific zebrafish p63 mutant (Tp63hu2525) mutant was isolated that bears a TCA->TAA nonsense mutation (S48X) in the last TA-specific exon truncating the N-terminal isoforms of TAp63, while leaving ΔNp63 isoforms unaffected (Figures 6A and S3). RT-PCR followed by restriction fragment length polymorphism (RFLP) analysis of skin samples confirmed the presence of the premature stop codon in TAp63 transcripts of TAp63hu2525/hu2525 animals (Figure 6O). Furthermore, qRT-PCR revealed that in TAp63hu2525/hu2525 mutant skin, TAp63 mRNA levels were more than 7.5 fold reduced compared to wild-type siblings, whereas ΔNp63 mRNA levels were slightly up-regulated (Figure 6P), suggesting that the mutant TAp63 transcripts undergo nonsense-mediated mRNA decay and that TAp63 normally has a subtle negative effect on ΔNp63 transcription. Unfortunately, we could not investigate TAp63 proteins in TAp63hu2525/hu2525 mutants, as all tested p63 antibodies failed to detect p63 proteins after Western blotting of zebrafish extracts (data not shown). However, upon injection of synthetic mRNAs encoding N- or C-terminally Myc-tagged zebrafish TAp63γ, wild-type mRNA gave rise to full-length fusion protein, whereas hu2525 mutant mRNA yielded no products (Figure 6Q and data not shown), suggesting that the stop codon cannot be by-passed, that no internal start site is used, and that the truncated protein is unstable. Furthermore, in contrast to wild-type TAp63γ, hu2525 mutant mRNA lacked pro-apoptotic activity upon over-expression in early zebrafish embryos (Figure 6K; compare columns 2 and 4). Together, this suggests that the hu2525 mutant is a TAp63 null.
Like p53zdf1/zdf1 mutants, TAp63hu2525/hu2525 mutants were viable and fertile. Anti-p63 immunofluorescence analysis revealed normal staining in regular epidermis of TAp63 mutants (Figure 7A,B; Figure S4E–G), suggesting that ΔNp63 is the major p63 isoform of the adult zebrafish skin, consistent with our RT-PCR data. However, both TAp63 and p53 mutants displayed specific breeding tubercle deficiencies of variable strength (C1,C2). SEM (Figure 7C–E), calcein stainings (Figure 7F) and histological sections (Figure 7G–I; Figure S4A–D) revealed complete absence (C2) in some individuals and reduced numbers and/or sizes of tubercles (C1) in others, while frequencies of strongest phenotypes were significantly increased in TAp63/p53 double mutants (Figure 7F). At the molecular level, even tubercle remnants of more weakly affected (C1) TAp63 and/or p53 mutants displayed significant alterations compared to wild-type tubercles, such as reduced tgm1 (Figure 7J–L) and ectopic (ΔN)p63 expression (Figure 7A,B) in upper layers, as well as reduced proliferation rates at the base, which again was most prominent in TAp63/p53 double mutants (Figure 7M–P). In sum, this makes the molecular signature of breeding tubercle remnants of weaker TAp63/p53 mutants more similar to that of the regular epidermis.
Together, our data suggest that TAp63 and p53 are required for proper keratinocyte proliferation at the base of breeding tubercles, as well as for cornification-like differentiation processes in upper layers.
In light of their known roles in keratinocyte differentiation in mouse [39], we also explored the involvement of Notch signalling and caspase 3 during zebrafish tubercle development, as well as their epistatic relationships to TAp63/p53. Differentiating tubercle keratinocytes displayed strong Notch signalling, as revealed by Tg(TP1bglob:eGFP)um13, a transgenic in vivo reporter with 12 RBP-Jk binding sites [63], which was strongly expressed in upper layers of breeding tubercles, but not in lower layers and regular epidermis, complementary to the distribution of (ΔN)p63 (Figure 8A), but overlapping with aCasp3 (Figure 8B). Double labelling with calcein blue as a marker of differentiated cap cells revealed that during development, Notch signalling in the tubercle domain is initiated several days before the first cap cells have differentiated (Figure 8C,D). Furthermore, TAp63 and p53 mutants displayed strongly reduced Notch signalling and aCasp3 levels in their (significantly smaller) tubercles at 50 dpf (Figure 8F–H,L–N), although wild-type tubercles of similar sizes at 30 dpf were strongly positive for both (Figure 8E,K).
To study whether Notch signalling and caspase 3 are required for breeding tubercle formation, we treated wild-type fish from 20 to 50 dpf with the specific Notch/g-secretase inhibitor DAPT or the caspase 3 peptide inhibitor zDEVD-fmk. While both treatments did not affect the general conditions and growth of the fish (Figure S5), DAPT-treatment caused significant reductions in the numbers and sizes of tubercles, in the activity of the Notch reporter and in aCasp3 levels (Figure 8I,O,Q), as also seen in p53 and TAp63 mutants (Figures 7F and 8G,H,M,N). zDEVD-fmk treatment had similar effects on tubercle numbers and sizes and on aCasp3 levels (Figure 8P,Q), while leaving Notch signalling unaffected (Figure 8J). Furthermore, despite the restriction of Notch signalling to upper breeding tubercle layers (see above), DAPT treatment led to a significant reduction of keratinocyte proliferation at the base of the tubercles (Figure 8R–T), comparable to the effects caused by loss of TAp63 and p53 function (Figure 7P). Finally, we re-introduced Notch signalling into TAp63 mutants, using a binary transgenic approach for temporally controlled expression of the constitutively active intracellular domain of Notch1 (NICD) [64]. While heatshock-induced NICD expression from 20–50 dpf had little effect in wild-type fish, it significantly elevated tubercle numbers in TAp63 mutants back to wild-type conditions (Figure 8U–X).
Together, these data suggest that TAp63/p53, Notch signalling and aCasp3 constitute a linear pathway required for proper breeding tubercle formation.
Both the in vivo role of TAp63 isoforms during vertebrate keratinocyte development and the stratification status of the epidermis of adult zebrafish were not fully understood. Here, we present data shedding new light onto both questions, identifying the zebrafish breeding tubercles as epidermal appendages with more advanced stratification and keratinocyte differentiation, and with regular desquamation and self-renewal, while presenting genetic evidence for an in vivo involvement of a TAp63/p53->Notch->caspase 3 pathway in these processes.
Thus far, it had been unclear whether the epidermis of adult zebrafish undergoes self-renewal as in mammals. Our transgenic lineage analyses indicate that both in regular epidermis and in breeding tubercles, derivatives of basal keratinocytes can be found in the outer cell layer (Figure 1). Furthermore, long-term observations of individual fish show that outer layers of breeding tubercles are shed off, and that it takes cells of the next layer several days to fully develop outer layer properties before they undergo desquamation themselves (Figure 4).
Electron microscopy and marker analyses further revealed striking differences in the differentiation of keratinocytes in breeding tubercles versus regular epidermis. In regular epidermis, basal and intermediary cells display largely identical ultrastructural features [12]. In addition, they are (ΔN)p63-positive and mitotically active throughout all layers, and express the same keratin genes (Figures 2 and 5). Only when ending up in the superficial layer do they become strikingly different, express specific keratins and other structural proteins, including components of tight junctions, which are only present in this layer. Strikingly, basal keratinocyte-derived superficial cells of the regular epidermis are indistinguishable from persisting cells of the early embryonic enveloping layer, with which they form a uniform and continuous periderm-like sheet (Figure 1E).
In contrast, breeding tubercles exhibit a much more pronounced stratification (see Figure 9A for schematic drawing). They consist of more epidermal cell layers and display a keratin gene expression pattern strikingly different from that of regular epidermis (Figure 2). Cell proliferation is confined to basal layers, while the expression of transglutaminase 1, a crucial cross-linking enzyme during mammalian cornification [53], which is absent in regular zebrafish epidermis, is confined to upper layers (Figures 2 and 5). Structurally, at least 4 different layer types can be distinguished. The basal layer, in which cells are organized in a much more regular and columnar fashion than in regular epidermis, several spinous layers, a second-tier layer, in which cells are sealed to each other via tight junctions, and a heavily keratinized outer cap layer (Figures 3 and 9A). This organization is similar to that of the epidermis of adult mammals. In addition, the presence of extruding lamellar body-like vesicles and of lipid deposits at the second tier – cap layer interphase is reminiscent of lipid envelope formation during mammalian cornification [54], while the progressive degradation of desmosomes between second tier and cap layer cells resembles the fate of corneodesomosomes in mammalian corneocytes [65]. However, there are also differences between zebrafish breeding tubercles and the cornifying mammalian epidermis. Thus, despite their high keratin content, tubercle keratinocytes lack obvious intermediary filament bundles and a cornified envelope (CE), in line with the absence of genes encoding the keratin-bundling protein filaggrin and the CE components loricrin and involucrin in teleost genomes [66]. Furthermore, in contrast to mammalian corneocytes, sloughed cap cells of tubercles still contain their nuclei and display some (loss of cell membrane integrity, nuclear pyknosis), but not all hallmarks of apoptosis [55], [56] (Figures 4 and 5).
Overall, this makes differentiated zebrafish tubercle keratinocytes more similar to the “immature horny cells with nuclei” observed in fetal human epidermis, which are supposed to represent a “transition phase of keratinisation” [2]. In line with such a more basal state, keratinocyte differentiation in zebrafish tubercles requires Notch signalling and caspase 3, regulators that are also needed for early steps of epidermal cornification in mammals [37], [39].
There is compelling evidence in mouse and zebrafish that ΔNp63 is required during early steps of epidermal development, promoting the proliferation and stemness of basal keratinocytes, while possibly blocking differentiation processes [26], [58], [59]. However, the in vivo role of its counterpart TAp63 during keratinocyte development and differentiation remained elusive. Also, to our knowledge, no defects during regular keratinocyte development have been reported for p53 mutants as yet [31], [32].
Here, we revealed specific defects during keratinocyte development in breeding tubercles of zebrafish p53 and TA-specific TAp63 mutants. The used p53 allele, zdf1 (also called e7), bears a missense mutation in the DNA binding domain. Although the strongest of all available p53 mutants [57], [67], it might not be a complete functional null. However, inferences with other members of the p53/63/73 family seem very unlikely, as zdf1 homozygous embryos lack the phenotypes caused by loss of p63 or p73 [58], [59], [61], [67]. Antimorphic effects are also unlikely for the used TAp63 allele, hu2525, as the mRNA and the resulting truncated protein seem to be unstable (Figure 6P,Q). Furthermore, the protein would only contain part of the transactivating domain, while lacking DNA binding and oligomerization domains (Figure 6A). Therefore, we conclude that both TAp63 and p53 are per se essential for normal breeding tubercle development.
Similar tubercle defects as in TAp63 and p53 mutants were obtained upon chemical inhibition of Notch signalling or caspase 3 activity (Figure 8). aCasp3 levels are strongly reduced in TAp63 and p53 mutants, as well as after loss of Notch signalling, whereas Notch signalling is only lost in TAp63 and p53 mutants, but not after inhibition of caspase 3. In addition, re-introduction of Notch signalling into TAp63 mutants rescues their tubercle deficiencies. Together with their shared expression in differentiating keratinocytes of wild-type tubercles (Figures 6E,G and 8B), this provides in vivo evidence for the presence and requirement of a linear TAp63/p53->Notch->caspase 3 pathway.
Future studies have to elucidate the genetic control of breeding tubercle formation upstream of TAp63 and p53. As previously suggested [21], we believe that TAp63/p53 activity in zebrafish tubercles is indirectly promoted via negative interferences with its antagonist ΔNp63, which in skin is present in vast excess over TAp63 (Figure 6D), but, in contrast to TAp63 (Figure 6E), most likely restricted to the base of breeding tubercles (Figure 7A). This initial inhibition of ΔNp63 production in upper layers of the tubercle anlage might be reinforced by various negative feedbacks between TAp63/Notch/caspase and ΔNp63. Thus, in mammalian epidermal keratinocytes, Notch1 has been shown to repress ΔNp63 expression [43], while TAp63 induces caspase-dependent ΔNp63 degradation [68], consistent with the observed increased ΔNp63 transcript levels in the skin (Figure 6P) and the presence of ectopic ΔNp63 protein in upper tubercle layers of TAp63 mutant zebrafish (Figure 7A,B).
It might appear paradoxical that in addition to reduced keratinocyte differentiation in upper layers of breeding tubercles (Figure 7J–L), loss of TAp63 or p53 function also leads to reduced keratinocyte proliferation at the base of the tubercles (Figure 7M–P), pointing to both differentiation- and proliferation-promoting effects of these regulators. Reduced tubercle growth and basal keratinocyte proliferation was also obtained upon inhibition of the TAp63/p53 mediators Notch or caspase 3 (Figure 8E–T), although both are only active in post-mitotic keratinocytes in upper tubercle layers (Figure 8A,B), pointing to non-cell autonomous mitogenic effects. Interestingly, in regenerating wings of the fruitfly Drosophila melanogaster, the enzymatic activity of the caspase Dronc generated in apoptotic cells does not only execute cell death in a cell-autonomous manner, but also promotes proliferation of adjacent cells even when Dronc-positive cells are prevented from dying, pointing to the presence of a Dronc-dependent mitogenic signal that acts in an apoptosis-independent and non-cell autonomous manner [48], [69]. It is tempting to speculate that a similar mechanism might be at play in breeding tubercles, in which a TAp63/p53->Notch->caspase 3 pathway active in upper epidermal layers, while cell-autonomously promoting terminal keratinocyte differentiation, enhances proliferation of cells in lower layers in a paracrine fashion (via an unknown secreted factor X; Figure 9B), accounting for proper tubercle growth during development and for proper balancing between cell loss via desquamation and cell renewal during tissue homeostasis [70]. Such a proliferation-stimulating role seems in contrast to the known functions of Notch and p53 as tumour suppressors [71], [72]. However, it is in line with the initial identification of p53 as an oncogene [73] and its more recently described function in the context of metabolic control [74].
The breeding tubercle phenotype of zebrafish TAp63 and p53 mutants is not fully penetrant and variable in strength. Interestingly, however, phenotypic penetrance and average strength are significantly higher in TAp63/p53 double mutants, suggesting that the two structurally related transcription factors play partially redundant roles. Although other reasons cannot be ruled out, this suggests that a similar, possibly even more pronounced functional redundancy might also account for the apparent absence of epidermal defects in TA-specific p63 mutant mice [34]–[36]. Of note, we even identified a few TAp63/p53 double mutant zebrafish in which breeding tubercles were not completely lost. This could be due to some remaining p53 activity (see above). In addition, it might point to the existence of further partially redundant factors. TAp73, the third member of the family, is a candidate, which in mouse is expressed in all examined tissues and required side by side with TAp63 in oocytes to prevent genomic instability and female infertility [24], [34], [75]. Particular ΔNp63 isoforms might also be involved. Although in most cases they have dominant negative effects on TAp63 isoforms and p53, acting as transcriptional repressors, cases have been reported where they transactivate target genes [27] and positively cooperate with p53 [29].
Together, this demonstrates the complexity of the p53/p63/p73 system of transcriptional regulation, while revealing that it can be helpful to perform genetic analyses in different in vivo model systems, taking advantage of variations that have occurred during vertebrate evolution.
Unless stated otherwise, wild-type fish from a mixture of TL and EK were used. The mutant Tp63hu2525 (S48X) line was generated upon our request in the Hubrecht Institute, NL, using target-selected mutagenesis (TILLING) [62]. The stable transgenic lines Tg(krt4:creERt2)fr33, Tg(krt19:dTomato)fr34 and Tg(krt19:creERt2)fr35 were generated using the Tol2 kit [76], [77] with the described krt4 [49] or krttc19e [50] (here abbreviated as krt19) promoter fragments for construct generation, followed by standard injection and screening procedures.
The mutant line Tp53zdf1 (M214K) [57] and the transgenic lines Tg(krt4:GFP)gz7 [49], Tg(actb2:loxP-STOP-loxP-dsREDEx)sd5 [78], Tg(TP1bglob:eGFP)um13 [63], Tg(5xUAS-E1b:6xMYC-notch1a)kca3 [64] and Tg(-1.5hsp70l:Gal4)kca4 [63] have been previously described. For NICD expression, Tg(5xUAS-E1b:6xMYC-notch1a), Tg(-1.5hsp70l:Gal4)kca4 double transgenic fish were heat-shocked from 20–50 dpf once a day for 1 hour at 40°C.
The TAp63hu2525 allele was genotyped using the dCAPS (derived Cleaved Amplified Polymorphism Sequence) method [79] with PCR primers CTGACCCCGAGGTTGTCTAA (sense) and TGCTAATCTGTATAGTATTGGAAGCT (antisense) and subsequent HindIII digest. The Tp53zdf1 allele was identified via an RFLP (Restriction fragment length polymorphism) genotyping assay with PCR primers CCAGAGTATGTTCTGTCCA (sense) and TGATTGTGAGGATGGGCCTGCGGAATC (antisense) and subsequent BstyI restriction digest.
Fish carrying the Tg(-1.5hsp70l:Gal4)kca4 or Tg(5xUAS-E1b:6xMYC-notch1a)kca3 transgene were identified by PCR transgene amplification with the primers CGGGCATTTTACTTTTATGTTGC (gal4, sense), CATCATTAGCGTCGGTGAG (gal4 antisense), CATCGCGTCTCAGCCTCAC (NICD sense), CGGAATCGTTTATTGGTGTCG primer (NICD antisense), yielding a 1.2 or 0.3 kb amplification product, respectively.
All zebrafish experiments were approved by the national animal care committees (LANUV Nordrhein-Westfalen; 8.87-50.10.31.08.129; 84-02.04.2012.A251; City of Cologne; 576.1.36.6.3.01.10 Be) and the University of Cologne
Tg(krt19:creERT2)fr35, Tg(actb2:loxP-STOP-loxP-dsREDEx)sd5 double transgenic, Tg(krt19:creERT2)fr35, Tg(actb2:loxP-STOP-loxP-dsREDEx)sd5, Tg(krt4:GFP)gz7 triple transgenic, or Tg(krt4:creERT2)fr33, Tg(actb2:loxP-STOP-loxP-dsREDEx)sd5; Tg(krt4:GFP)gz7 triple transgenic embryos were treated with 5 µM 4-Hydroxytamoxifen (Sigma Aldrich; H7904) in the dark at 28°C from 24 hpf to 96 hpf before being returned to normal system conditions for growing up. For Figure 1A–D, larvae or adult fish were fixed with 4% paraformaledehyde (PFA)/PBS overnight at 4°C, followed by cryosectioning, mounting of sections in Mowiol (Carl Roth) containing DAPI and fluorophore analysis with a Zeiss Apotome. For Figure 1E fish were stained with rabbit anti-RFP (1∶100; MBL, PM005) (secondary = Alexa Fluor-488 anti-rabbit (1∶100; Invitrogen, A11008)), mouse anti-p63 (1∶100, Santa Cruz, sc-8431) (secondary = Alexa Fluor-647 anti-mouse (1∶100; Invitrogen, A21240)) and Rhodamine-Phalloidin (1∶100, Invitrogen, R415), and analyzed via confocal microcopy (Zeiss LSM710 Meta).
Epidermal cell proliferation was assessed by BrdU incorporation after incubating adult fish in 100 µg/ml BrdU (Sigma) in fish system water for 12 or 24 hours, followed by anti-BrdU immunolabelling. TUNEL assay was performed using the in situ Cell Death Detection Kit, POD (Roche) according to the manufacturers recommendations.
For in vivo calcein staining, fish were incubated for two hours in calcein green or calcein blue solution (100 mg/l; Sigma Aldrich). After extensive washings, fish were anaesthetized with Tricaine (ethyl-3-aminobenzoate methanesulfonate, Fluka) for fluorescence analysis of live whole mounts or after PFA fixation and cryosectioning.
For histological, immunofluorescence and in situ hybridization analyses, adult zebrafish were sacrificed by Tricaine overdose and fixed in 4% PFA overnight at 4°C. Samples for paraffin embedding were decalcified in 0.5 M EDTA (pH 7.4) at room temperature for 5 days, dehydrated in a graded series of alcohols, cleared in Roti-Histol (Carl Roth) and embedded in paraffin wax. 10 µm sections were cut using a Leica RM2255 microtome. Samples for cryosections were orientated in 15% sucrose with 1% agarose in PBS and mounted in tissue freezing medium (Leica). 10 or 12 µm sections were obtained using a Leica CM1850 cryostat.
Paraffin sections were stained with hematoxylin & eosin or acidic fuchsin orange G (AFOG) trichrome (Gennova) according to standard protocols. For immunofluorescence analysis of paraffin- or cryosections, antigen retrieval was performed with 10 mM sodium citrate (pH 6.0) at 70°C for two hours, followed by washes and primary and secondary antibody incubations in PBS supplemented with 10% fetal calf serum (FCS), and mounting of sections in Mowiol containing DAPI. Primary antibodies other from the ones described above were: rabbit anti-activated caspase3 (1∶1000, abcam ab-13847), mouse anti-BrdU (1∶200, Roche 1170376), mouse anti-pan Keratin Type II (1∶200, Progen 61006). Secondary antibodies used were: anti-mouse Cy3 (1∶1000, Invitrogen), anti-rabbit Cy3 (1∶1000, Invitrogen), Alexa Fluor-488 anti-rabbit (1∶1000, Invitrogen)
In situ hybridization on paraffin sections was performed according to [80]. Antisense RNA probes were generated via in vitro transcription with Dig RNA labelling mix (Roche) and the following templates and conditions: krt8 (GenBank BI875660): 1.8 kb cDNA fragment cloned from EST into pBluescript SK, linearization with HindIII, transcription with T3 RNA polymerase; cki (GenBank AF197880): 0.6 kb fragment in pSPORT, EcoRI, SP6 RNA pol; krt5 (GenBank AF197909): 0.4 kb fragment in pBluescript SK, KpnI, T3 RNA pol; krt17 (ZFIN-ID zgc:92061; GenBank BI850052): 1.5 kb fragment in pSPORT, EcoRI, SP6 RNA pol. For krtt1c11a and tgm1, 1.0 kb fragments were amplified via RT-PCR and cloned into pGMTeasy (Promega) (krtt1c11a: SpeI, T7 RNA pol) or pCRII (tgm1: XhoI, SP6 RNA pol). For a TA-specific TAp63 probe, a 408 bp TA1 cDNA fragment was amplified via RT-PCR with the primers 5′-CAGGGGCTAGCTTCTAGTGG-3′ (sense) and 5′-TGTAAGGGGCTCCTCAGGCTC-3′ (antisense) and cloned into pGEMTeasy. The plasmid was digested with SpeI and transcribed with SP6 RNA pol for antisense, and with NcoI and T7 RNA pol for sense probe. For p53, EST clone MPMGp609B127Q8 with the full-length p53 cDNA in pSPORTI was linearized with EcoRI and transcribed with SP6 RNA pol for antisense, and with BamHI and T7 RNA pol for sense probe.
Images were captured on a Zeiss Axiophot, Zeiss Apotome, Zeiss Confocal (LSM710 META) or Leica M165 FC stereo microscope.
Transmission electron microscopy (TEM) of adult zebrafish was carried out as described [81]. For Scanning electron microscopy (SEM), adult fish were sacrificed and fixed overnight in 4% PFA at 4°C, dehydrated and either cryo-fixed, sputter-coated (gold/palladium) and transferred onto the SEM cryo-stage while still frozen, or critical point dried (CPD), sputter-coated and evaluated at room temperature.
Fish were raised from 20–50 dpf in E3-medium containing 100 µg/ml of the γ-secretase inhibitor DAPT (N-N-(3,5-difluorophenacetyl)-L-alanyl)-S-phenylglycien t-butylester; Sigma-Aldrich 208255) [82], 5 µg/ml of the caspase 3 peptide inhibitor z-DEVD-fmk (Calbiochem 264155-80) [83] or 0.2% DMSO as control. Standard length (SL) of the fish was used to control equal development of each group.
RNA of whole zebrafish embryos at different developmental time points or from isolated tissues or organs of adults was isolated using the trizol reagent (Invitrogen). cDNA was generated using random hexamer primers. Regular PCR was carried out with the TA-specific sense primers TAS1-3 or the ΔN-specific sense primer ΔNs (see Figure S1), combined with a shared reverse primer 5-GTGACTGGGTGGGGCTATTT-3. Zebrafish actb2 (GenBank: BC0675676) specific primers were used as control (sense, 5′-AGTTTGAGTCGGCGTGAAGT-3′; antisense, 5′-AGGCTGTGCTGTCCCTGTAT-3′). PCR reactions were performed with an annealing temperature of 55°C for 35 cycles.
For cDNA RFLP analysis, the 629 bp fragment shown in Figure 6B was amplified with primers TA3 and the reverse primer 5′-GTGACTGGGTGGGGCTATTT-3′, followed by overnight digest with MboI (NEB) and electrophoresis in 4% agarose gel, revealing 379, 107, 101 and 42 bp cleavage products in wild-type, but only 279, 208 and 42 bp products in hu2525 mutant cDNA.
Quantitative RT-PCR was performed in triplicates (2 experiments each) with TaqMan primers (see below) and an Applied Biosystems 7500 Fast Real-Time PCR System under default PCR conditions, resulting in specific 65 bp (TAp63; shared by TA1 and TA4) and 72 bp (ΔNp63) products. Used primers were: TA-forward, 5′-GCCTGAGGAGCCCCTTACA-3′; ΔN-forward, 5′-CCAATGCTCCCTCATCCTACA-3′, TA-reverse and ΔN-reverse, 5′-CATTTTGATCCATGCTGTTGAGA-3′; TA-TaqMan probe, 5′-CTCAGTATACAAGCCTGGG-3′; ΔN-TaqMan probe, 5′-AGCCTCAGTATACAAGCC-3′; standard, rps23 (ribosomal protein S23; standard; Applied Biosystems; Dr.0343030371m1). Amplification efficiencies were determined with a dilution series of cDNA from adult skin, and were above 95% for all three amplificants (TAp63, 99,4%; ΔNp63, 95.9%; rps23, 95.2%). Data were analyzed using Biosystems Prism SDS and Excel software, applying ΔCT and ΔΔCT calculations.
To generate the TAp63 expression constructs pCS2-TA(1)p63γ and pCS2-TA(4)p63γ, a replacement strategy was used, amplifying the N-terminal fragments of the TA1 and TA4 isoforms of TAp63 via RT-PCR from adult skin of wild-type and hu2525 mutants with forward primers 5′-TTGGATCCACCATGACCTCTCCTTATGCAGC-3′ (TA1) or 5′-TTGGATCCACCATGTCACAGGGCCAGGGCTC-3′ (TA4), and reverse primer 5′-GTGACTGGGTGGGGCTATTT-3′, followed by BamH1/BspM1 digest and cloning into BamH1/BspM1-digested pCS2-ΔNp63γ [58]. To generate expression constructs for TAp63 with six N-terminal Myc tags, TAp63γ coding sequences were amplified from wild-type and hu2525 mutant pCS2-TA(4)p63γ plasmids with primers 5′-CGAATTCAACCATGTCACAGGGCCAGGGCTC-3′ (sense) and 5′- TTTCTAGATCACACTGATTGAGAACTCTTTTT G-3′ (antisense), digested with EcoRI and XbaI, and cloned into EcoRI/XbaI digested pCS2-MT (www.addgene.org/vector-database/2296/). For expression constructs with six C-terminal Myc tags, amplification was performed with primers 5′-TTGGATCCACCATGTCACAGGGCCAGGGCTC-3′ (sense) and 5′-CGATCGATTCACTGATTGAGAACTCTTTTTGTC-3′, followed by digestion with BamHI and ClaI, and cloning into BamHI/ClaI, digested pCS2-MT.
Capped RNA was prepared after restriction digest of these expression constructs or pCS2-ΔNp63α1 [58] with KpnI, using the Message Machine kit (Ambion, Austin, TX). RNA was dissolved in water, and 1 nl per embryo injected. TAp63γ mRNAs were injected at a concentration of 10 ng/µl, 6xMyc-TAp63γ mRNAs at a concentration of 5 ng/µl, and ΔNp63α1 mRNA at a concentration of 25 ng/µl. Apoptosis and resulting embryonic death or embryonic malformations were scored at 8 hpf and 24 hpf, respectively, ΔNp63α1-induced loss of eyes at 32 hpf, as described [58].
Zebrafish embryos were dechorionated and deyolked, and cells were collected as described [84]. Cell pellets or adult tissues were either directly dissolved in SDS loading buffer as described [84], or first lysed in chilled CSH buffer (50 mM Tris-HCl (pH 7.5), 250 mM NaCl, 1 mM EDTA, 1% Triton-X100, supplemented with cOmplete Protease Inhibitor Cocktail, Roche), followed by protein concentration determination. 10–12% SDS-PAGE, blotting on nitrocellulose membrane, Ponceau staining and immunodetection were carried out as described [84]. Used primary antibodies were: anti-Myc, 9B11 (mouse, Cell Signaling Technology; 1∶2000); anti-p63, 4A4 (mouse, Santa Cruz Technologies, against aa 1–205 of human ΔNp63), D-9 (mouse, Santa Cruz Biotechnology, against aa 15–151 of human ΔNp63), H-137 (rabbit, Santa Cruz Biotechnology, against aa 15–151 of human ΔNp63), H-129 (rabbit, Santa Cruz Biotechnology, against aa 513–641 at C-terminus of human TAp63α).
|
10.1371/journal.pcbi.1000130 | Optimal Compensation for Temporal Uncertainty in Movement Planning | Motor control requires the generation of a precise temporal sequence of control signals sent to the skeletal musculature. We describe an experiment that, for good performance, requires human subjects to plan movements taking into account uncertainty in their movement duration and the increase in that uncertainty with increasing movement duration. We do this by rewarding movements performed within a specified time window, and penalizing slower movements in some conditions and faster movements in others. Our results indicate that subjects compensated for their natural duration-dependent temporal uncertainty as well as an overall increase in temporal uncertainty that was imposed experimentally. Their compensation for temporal uncertainty, both the natural duration-dependent and imposed overall components, was nearly optimal in the sense of maximizing expected gain in the task. The motor system is able to model its temporal uncertainty and compensate for that uncertainty so as to optimize the consequences of movement.
| Many recent models of motor planning are based on the idea that the CNS plans movements to minimize “costs” intrinsic to motor performance. A minimum variance model would predict that the motor system plans movements that minimize motor error (as measured by the variance in movement) subject to the constraint that the movement be completed within a specified time limit. A complementary model would predict that the motor system minimizes movement time subject to the constraint that movement variance not exceed a certain fixed threshold. But neither of these models is adequate to predict performance in everyday tasks that include external costs imposed by the environment where good performance requires that the motor system select a tradeoff between speed and accuracy. In driving to the airport to catch a plane, for example, there are very real costs associated with driving too fast and also with being just a bit too late. But the “optimal” tradeoff depends on road conditions and also on how important it is to catch the plane. We examine motor performance in analogous experimental tasks where we impose arbitrary monetary costs on movements that are “late” or “early” and show that humans systematically trade off risk and reward so as to maximize their expected monetary gain.
| In the execution of any movement, there is always timing uncertainty. This uncertainty has two major consequences. First, it limits performance on any task for which there are costs associated with temporal imprecision. Second, it has implications for how the motor system should plan movements when the costs of temporal imprecision are asymmetric. In hurrying to catch a subway train, for example, the cost of arriving early is usually small compared to the cost of arriving late and missing the train. An optimal movement planner must take into account temporal reward asymmetries in forming movement plans.
The complexity of movement planning under risk is further increased because temporal uncertainty in the motor system changes constantly. Two major sources of variation in temporal uncertainty occur over different time courses and have different properties: One is a uniform, global shift in temporal uncertainty possibly due to aging, fatigue, injury or disease [1]–[9]. The second is a linear increase in the standard deviation of movement duration with increases in mean movement duration [10].
Here we use a model of optimal temporal movement planning to investigate the control of movement duration in the face of these two types of temporal uncertainty while human subjects attempted to touch a computer screen within a specified temporal window. We introduced asymmetries in the penalties imposed for early vs. late movement timing (Figure 1A), while at the same time increasing subjects' temporal uncertainty by adding Gaussian noise with 25 ms standard deviation (see Methods). As in all models of motor planning and motor control based on decision theory, we are concerned with the interplay of three elements: possible decisions (here planned movement time, τ), uncertainty in the mapping of motor decisions to motor outcomes (represented by the family of probability distributions p[t|τ]), and the costs/benefits resulting from those motor outcomes, G(t). The mathematical models considered here are part of a growing literature on Bayesian decision models of motor phenomena, such as models of motor adaptation [11]–[13] and motor planning/control e.g., [14]–[21], including the use of prior information in spatial [16],[18] and temporal [17] motor planning, the use of asymmetric cost functions in spatial motor planning [14]–[15],[19] and when selecting a speed-accuracy tradeoff [20]–[21]. The neural computation of decision variables such as those considered here and in previous work has also begun to be investigated [22]–[25].
Figure 1B illustrates the computations needed to maximize expected gain with temporally asymmetric penalties. When discussing movement duration, we must distinguish between the planned arrival time, denoted τ, and the actual arrival time, t. When movements are executed, the actual arrival time will be unpredictably earlier or later than τ. In Figure 1B we show four possible choices of τ and outline the calculation of expected gain for each. Note that the optimal planned arrival time need not fall within the temporal reward window.
Human performance will be optimal if the CNS learns its linear temporal uncertainty function,(1)as it relates to planned movement time (τ), and uses this information (ασ and βσ) to plan reach times that maximize expected gain. Human performance in our task could be sub-optimal in several ways, each depending on the type of information the CNS maintains about Equation 1. We consider 5 such sub-optimal models, denoted M1, …, M5. In the first three of these, subjects fail to take account of ασ, βσ, or both when planning reaches. In model M1, subjects fail to compensate for the experimentally imposed static increase in temporal uncertainty due to the added Gaussian noise (SD = 25 ms); in M2 subjects fail to compensate for the linear increase in temporal uncertainty with increasing reach duration; and in M3 subjects fail in both respects (for details, see Methods: Data Analysis and Model Comparison). Models M4 and M5 were analogous to models M2 and M3, respectively, but assumed the offset or slope were unknown and hence not fixed to match the training data or added 25 ms timing uncertainty. We compare subjects' performance to each of these sub-optimal movement strategies, and to the optimal strategy (M0) that results in maximum expected gain.
During training trials, subjects attempted to produce reaches with an experimenter-specified temporal duration; no rewards or penalties were imposed. In Figure 2A, we plot the mean movement duration as a function of the target duration for subject HT. The points lie near the identity line, indicating that the subject could accurately produce a wide range of movement times on command. Figure 2B shows the temporal uncertainty function (the standard deviation of arrival times as a function of target duration, with and without the added noise) measured during training for the same subject. As expected, unperturbed standard deviations (dot-dashed line, open symbols) increase linearly across this range. Estimated Weber-noise parameters (ασ) for all subjects' temporal uncertainty functions, and verification of the stationarity of those functions (across the training trials and the subsequent main experiment), are provided in Figure 3. Note that fitted functions obtained from training data (lines) and the standard deviations measured during main-experiment reaches (filled diamonds) were well-matched, consistent with the idea that subject performance did not change during the experimental reaches.
Each of the models makes predictions of reach durations that are based on the aspects of the temporal uncertainty function it incorporates. Because the optimal model (M0) incorporates both components of the temporal uncertainty function, it can take account of the temporal noise actually experienced by each subject when planning reaches, in turn allowing it to predict optimal movement times. Three of the sub-optimal models (M1–M3) each specify only a portion of the actual temporal noise experienced by subjects. Because these models cannot account for the full temporal uncertainty function, their predicted ‘best’ movement times are sub-optimal. For each subject and model, we derived predictions of the mean duration in each of the four conditions that would maximize expected gain in the task given that temporal uncertainty function (see Methods: Model Predictions; Figure 4 illustrates these calculations for an example subject). These predictions allow us to compare observed performance in the task to the theoretical performance of subjects who maximize expected gain under the constraints imposed by each of the four models. In addition to these four models, we considered two sub-optimal models that did not have fixed parameters (M4 and M5). In models of this type, the model likelihood (see Method: Data Analysis and Model Comparison) is calculated by integrating over the possible values of the unknown parameters (e.g., overall noise level).
The results of a Bayesian comparison of the performance of the four models (see Methods: Data Analysis and Model Comparison) favored the optimal model M0 over the sub-optimal models; yielding 11.5 dB in favor of M0, but −60.5 dB, −11.5 dB and −41.4 dB of evidence for M1, M2 and M3, respectively. Models M4 and M5 are less constrained, resulting in evidence below −100 dB. Negative evidence is evidence against a model relative to the other possible models. In our previous work [26] we have used 3 dB evidence, corresponding to odds of nearly 2∶1, as a minimal guideline for inferring an advantage for a model over its competitors. The 11.5 dB evidence for M0 is strong, corresponding to nearly 15∶1 odds in favor of the optimal model over the set of alternatives.
To assess inter-subject variability, we recomputed the evidence values for 5 subgroups of subjects, with each subgroup consisting of all subjects but one. The change in evidence that occurred as we left each subject out is a measure of how much the conclusions we draw are based on one subject alone. While the evidence decreases somewhat when each subject is removed (and it should since we are basing our conclusion on fewer data), it always favored M0, and always by at least 7.5 dB, consistent with the conclusions based on all subjects taken together. We note, in particular, that removing the non-naive subject who was an author (TEH) still resulted in evidence of 9 dB in favor of M0.
In addition, we plotted, for all subjects and conditions, the mean observed movement duration as a function of the duration predicted by each of the four models (Figure 5 plots the deviations of the actual from the predicted movement times). In such a plot, consistency of the data with the model corresponds to the data falling along the identity line. We computed linear regressions of observed mean duration as a function of predicted mean duration for each of the four models. Only M0 had a best-fit slope and intercept whose confidence intervals contained those of the identity line (Table 1), corroborating the result of the Bayesian model comparison. We conclude that the evidence favoring M0 over any of the competing models is overwhelming, implying that subjects compensated for their increased uncertainty at longer durations and also for the 25 ms added uncertainty imposed experimentally.
To investigate how the suboptimal models fail, we present differences between observed average temporal endpoints and model predictions for each of the four models (Figure 5). For each of the sub-optimal models, we describe how the pattern would appear if data were fit with that model.
Model M1 compensates for increased temporal uncertainty with increased movement duration but fails to compensate for the σ = 25 ms temporal noise added experimentally. Subjects conforming to this model will have temporal aim points closer to the center of the target region than they should be since they are based on an erroneously small estimate of temporal uncertainty. That is, compared with the optimal model (M0), model M1 predicts longer durations for predictions of durations shorter than the target duration (650 ms), and shorter durations for predictions longer than the target duration. Thus, we predict the left-hand cloud of residuals to move down and right and the right-hand cloud to move up and left, which is precisely what happened (upper-right panel, Figure 5).
Subjects employing model M2 (lower-left panel, Figure 5) would fail to take duration-dependent noise into account, but compensate for the s = 25 ms temporal noise added experimentally. Such subjects overestimate noise for short durations and underestimate it for long durations. Intuitively, the residuals should move up and left. This is true of most data points, but not all. The intuitive pattern is occasionally broken due to the complex, nonlinear calculation of expected gain (Figure 1B) and the switch from the veridical uncertainty function (M0) to an incorrect, flat function (M2). As expected, the predictions of M3 combine the shifts of the other two suboptimal models.
In summary, based on the comparison of the optimal and three suboptimal models, we conclude that subjects delayed or advanced their temporal endpoints in accordance with the calculated optimal times defined by M0. The Bayesian model comparison employed is novel and correct for comparison of non-nested models (see Method: Data Analysis and Model Comparison). We also carried out a set of statistical tests based on linear regression of actual versus predicted times. The conclusions based on these regressions tests are identical to those just reported: we reject models M1, M2 and M3 but not M0 (Table 1).
The gains earned by subjects potentially provide an additional dimension for testing the models. We have compared actual gains to expected gains predicted by each of the models. However, the gain functions are flat relative to the sampling variability of observed points earned, so that this analysis does not serve to differentiate the models.
To investigate the possibility that subjects used a hill-climbing strategy during the main experiment, instead of maximizing expected gain by taking account of their own temporal uncertainty function and experimentally imposed gain function, we performed a hill-climbing simulation using each subject's temporal uncertainty function. In the simulation, intended duration was moved away from the penalty region by 3Δt ms after each penalty and towards the center of the target region by Δt ms for each miss of the target that occurred on the opposite side from the penalty (corresponding to the 3∶1 ratio of penalty to reward). The value of Δt was initially set to be relatively large. With each change of direction of step, Δt was reduced by 25% to a minimum step size of 1.5 ms. While this simulation approximately reproduced the final average reach times observed experimentally, it does not provide a good model of subject performance. First, there were significant autocorrelations of reach durations beyond lag zero in the simulation data but not in the experimental data. Second, a learning algorithm would be expected to produce substantially higher σ values during test than those observed during training. This is what we found with our hill-climbing simulation. Using subjects' training σ values to produce the simulated data, the simulation produced 17 out of 20 main-experiment σ values that were above the training values, whereas our subjects' main-experiment σ values (Figure 3) were entirely consistent with temporal uncertainty functions measured during training.
To move accurately, an organism's motor system must generate an intricate series of precisely timed neural commands. The exact nature of these commands is not known. Whatever the format of the command signals [27]–[32], movement controlled by any physical controller-actuator system, including biological motor systems, will always exhibit some motor uncertainty. Nevertheless, it is possible to plan movements that will maximize expected gain in the face of that uncertainty. To do so, an organism must be capable of assessing both the probabilities of possible movement outcomes and their consequences.
One of the most thoroughly studied cases in which humans integrate the probabilities of possible movement outcomes and their consequences is the tradeoff between movement speed and spatial accuracy [20]–[21], [33]–[34]. However, in our experiment we were concerned with temporal accuracy, and faster movements are typically more temporally accurate (the opposite of the spatial speed-accuracy tradeoff). By imposing costs for early/late arrivals, we were able to determine whether the motor system is capable of picking movement times that maximize expected gain, taking into account temporal uncertainty.
We conclude that, in the timing task we examined, the motor system estimates and compensates almost perfectly for its own temporal uncertainty and correctly anticipates how that uncertainty interacts with the asymmetric reward structure of the environment. This outcome is plausible given the close neurophysiological links between motor timing and the assessment of probabilities and consequences [22]–[25], [35]–[37].
We note however that it has been argued that a representation of time plays no role in one of the most basic forms of motor learning: motor adaptation [38]. The current study provides evidence that the motor system is capable of using a representation of time in at least some circumstances where the consequences of the movement are unambiguously linked to the timing of the movement, and in addition that it does so optimally.
Several models of spatio-temporal movement control are based on optimizing an internal cost function that either includes or predicts movement timing. One such model of trajectory formation, the minimum variance model [39], assumes that the CNS selects a spatio-temporal reach trajectory by optimizing a cost function based on the movement's endpoint variance. In particular, the minimum variance model selects “…the temporal profile of the neural command … so as to minimize the final positional variance for a specified movement duration…” [39], p. 782. More recently the minimum-time model of trajectory formation has been proposed [40] based on the assumption that, subject to a constraint on movement accuracy, the CNS attempts to minimize movement duration. In both models, the speed-accuracy tradeoff is modeled by scaling the spatial variance of the reach with the amplitude of the motor control signal; that is, they assume signal-dependent spatial motor noise.
In the absence of signal-dependent noise, both models would predict a ‘bang-bang’ control scheme, where the control signal takes first a maximum positive and then maximum negative value producing alternating maximum forward and reverse accelerations leading to maximum movement speed and hence minimum duration. However, bang-bang control predicts trajectories that are inconsistent with typical motor behavior. By modeling spatial noise as signal-dependent, it is possible to predict a range of important behavioral results with both the minimum-variance and minimum-time models, such as the smooth variation in spatial and temporal reach profiles e.g., [41]–[42], Fitts' law [33], and the spatio-temporal details of saccadic trajectories [43].
Unlike these previous studies, here the emphasis is on accuracy of movement duration. This results in a reverse speed-accuracy tradeoff; slower movements have lower temporal accuracy (even though they have higher spatial accuracy). We show that, in a task where spatial uncertainty (and therefore signal-dependent spatial noise) plays essentially no role, reach durations are selected to nearly maximize expected gain in the presence of duration-dependent temporal uncertainty.
Duration-dependent temporal uncertainty constitutes a constraint on the temporal aspects of movement planning that is similar in many respects to the planning constraint imposed by signal-dependent spatial noise. Simultaneously minimizing temporal and spatial noise provides a method of solving the underconstrained problem of trajectory selection. Although several previous studies have proposed multiply-constrained models of movement planning [44]–[45] and the duration-dependence of temporal uncertainty is well known e.g.,[10]; [46]–[47], we provide the first demonstration of the CNS making use of its own temporal uncertainty in movement planning. While selecting the movement trajectory that minimizes spatial and/or temporal noise is a possible method of movement planning, the optimal movement planner carefully separates the constraints imposed on spatial and temporal accuracy (duration-dependent temporal noise and signal-dependent spatial noise) with the costs of spatial and temporal errors, which we discuss next.
In both the minimum-time and minimum-variance models [39]–[40], a trajectory is selected so as to optimize an internal cost for spatial variance or movement duration (respectively) in the presence of signal-dependent spatial noise. The cost is internal in the sense that it does not make reference to any externally imposed costs on movement errors, such as monetary rewards and penalties that may be imposed due to one's spatial precision or movement duration. There have been a large number of models of movement based on the optimization of internal cost functions that identify movement cost with an invariant kinematic or dynamic variable (time [48], spatial precision [39], torque-change [49]–[50], jerk [51], etc.). However, there are pitfalls inherent in identifying movement cost with an aspect of the movement itself, despite the current movement goals. For example, the minimum-variance model always chooses a movement with the best possible spatial precision, even when that level of precision is unnecessary for the task. Similarly, the minimum-time model always chooses the shortest duration movement that satisfies the constraint on spatial precision even when, as in some conditions of the current study, an external temporal cost function rewards longer-duration movements.
Recent models of optimal movement planning e.g., [14],[18],[26],[44] approach the problem somewhat differently. In these models, which have previously been used to predict spatial movement endpoints [14],[18] and movement trajectories [44], the difference between a constraint on movement planning and a cost incurred from movement error must be recognized. While duration-dependent temporal noise, signal-dependent spatial noise, energy consumption, biomechanics, etc. constitute constraints on movement planning and control, they are not properly costs. A cost essentially imposes a weighting on the available constraints, and is task dependent. By experimentally imposing costs [14]–[15], [18]–[21],[26] on spatial or temporal inaccuracy, it is possible to predict flexible movement strategies that incorporate task-relevant constraints (e.g., duration-dependent temporal uncertainty) while effectively ignoring (down-weighting) constraints that are not as important to the task at hand (signal-dependent spatial uncertainty). In the present study, we manipulated the temporal cost function by imposing penalties on too-short reach durations in some conditions, and too-long durations in other conditions, and determined whether subjects responded appropriately to these different cost functions.
We have modeled movement planning as minimizing an external gain function in the presence of task-relevant internal temporal noise. By identifying the to-be-minimized cost with the movement goal we have separated fixed kinematic/dynamic variables from the purpose of the movement. This allows us to predict flexible movement plans that may minimize spatial or temporal uncertainty, but only when that is relevant to the task at hand. A deeper understanding of movement planning and execution will result from models that similarly separate cost functions from fixed sets of kinematic/dynamic variables while simultaneously taking account of task-relevant spatial and/or temporal uncertainty.
Subjects were instructed to reach to a computer screen. Prior to each reach, a timer bar was presented on-screen, indicating the timing of the rewarded and penalized temporal windows, along with a circular spatial target. To earn rewards, subjects had to touch within the circular target area within a specified temporal window (“temporal target”). All spatial targets (12 mm radius) were presented along a circular arc 430 mm from the start position (Figure 1C). The timer bar was used to indicate the reward structure of each trial (described below), and also to signal to the subject the movement duration achieved following completion of each reach. All measurements (spatial and temporal) were made with an Optotrak 3020, sampling at 200 Hz. Reach initiation was defined as the moment when the fingertip moved (at least) 2 mm toward the computer monitor, and reach termination as the time when the fingertip arrived within 3 mm of the monitor and the forward fingertip velocity fell below 3 mm/s. Subjects were seated facing the center of the (upright) computer monitor.
The start position of the reach was on the tabletop, in front of the upright computer screen. Fingertip position was controlled at the start of each reach, and constrained to be within 1 mm of the start position. The start position was 350 mm in front of the center of the monitor's bottom edge (Figure 1C). Target locations were selected from a circular arc on the screen. The arc was centered on the projection of the start position to the bottom edge of the screen (Figure 1C). All points on this arc were equidistant from the start position. Reaches were made in a dimly lit room (the majority of the light coming from the CRT), and subjects could see their hands. No feedback was presented on the screen showing the fingertip landing point, although an auditory beep indicated that the target had been touched.
Subjects were not told that Gaussian noise with σ = 25 ms was added to all measured temporal endpoints. This added noise, in combination with subjects' natural duration-dependent variations in temporal uncertainty, allowed us to determine whether subjects were sensitive to changes in the two sources of variation in temporal uncertainty described above. The noise-added temporal endpoint was displayed after each reach, shown as a thin line intersecting the timer bar at the appropriate position.
Each subject completed two sessions, a training session and the main experiment. Both sessions were completed within the same hour on a single day.
Subjects were first given a training session in which temporal targets (width: 3 ms, no adjacent penalty region) were presented at six target durations (565, 595, 625, 655, 685 and 715 ms; 8 repetitions each, in separate blocks, followed by 50 repetitions each, in separate blocks) spanning the range of temporal aim points observed during pilot work. Although this window was too narrow for subjects to reliably hit, subjects were not scored during training, and were told simply to time their reaches as closely to each target time as possible. This session allowed us to estimate the standard deviation of each subject's movement durations for a set of precisely known target durations, and also allowed subjects to learn their own (noise-added) temporal uncertainties in the task. Standard deviations at each target time (Figures 2B and 3) were measured from the final 40 repetitions to avoid possible initial practice effects.
Immediately following training, subjects were given a temporal target centered at 650 ms, with a half-width of 0.6σ650, where σ650 was the estimated SD of movement duration for a mean duration of 650 ms. In this way, we equated the difficulty of the task across subjects based on their training performance.
Subjects were paid a bonus for touching the spatial target within the temporal target window (Figure 1A, green, cross-hatched bars), and penalized for touching the spatial target within a temporal penalty window (Figure 1A, red, striped bars) or for failing to touch the spatial target. Four blocked conditions were employed (Figure 1A), two early temporal penalty conditions and two late penalty conditions (64 trials each). The two early temporal penalty regions began at 0 ms and ended either 0.6σ650 or 1.35σ650 ms prior to 650 ms. The two late temporal penalty regions began either 0.6σ650 or 1.35σ650 ms following 650 ms, and were open-ended.
The outcome of each trial was signaled by distinct auditory tones notifying the subject that a reward was earned or a penalty assessed. The possible reward earned on any trial was $0.12 and the penalty was −$0.36 (or −$0.60 for missed spatial targets). Note that the ratio of penalty to bonus magnitudes was 3∶1. Trials in which the spatial target was not touched were re-run (fewer than 1% of all trials) to equate the number of touched-target trials in each condition. The untouched-target trials were not analyzed.
Subjects were four students at New York University who were not aware of the purpose of the experiment and one author (TEH). All subjects gave informed consent before the experiment. The experimental protocol had been approved by the Institutional Review Board at New York University.
As described in the Introduction, decision theoretic models of motor behavior are concerned with the interplay of three elements: movement strategy, uncertainty, and the gain or loss from possible movement outcomes. The interplay of these three elements is represented graphically in Figure 1B for the optimal model, M0. Calculation of the temporal endpoints predicted by each of the models to be considered required that the expected gain, in terms of average bonus earned per reach, be computed based on the constraints supplied by the hypothetical system. For example, the optimal neuromotor controller would make use of information concerning both Weber-like increases in temporal uncertainty with increasing reach time, and the experimentally increased overall temporal uncertainty.
A given motor strategy or plan, s, determines the critical states of the system. Although motor plans are complex sequences of control signals in time, the only consequence of the choice of motor plan in our task is to select an expected temporal endpoint, τs. The expected gain from s is then given by (Figure 1B):(2)where G(t) describes the gain or loss associated with a particular temporal endpoint (Figure 1A and Figure 1B, middle panel). The term p(t | τs) describes the probability density of temporal endpoints expected from any chosen movement strategy s. Note that these are planned durations, not reaction times, and hence we have no a priori expectation that these distributions will be skewed. We model the duration distribution as a Gaussian with mean arrival time τs and a standard deviation σ(τs)(3)(QQ plots of these distributions confirm that the Gaussian distribution models the data well). The temporal uncertainty function, σ(τs) is able to capture the well-known Weber-like scaling of temporal standard deviation with mean arrival time τs (Figure 1B, top panel). We used values estimated from each subject's training data to compute individual σ(τs) functions for models M0–M3.
In Figure 1B (bottom panel), for the rightmost choice of τ, the probability of arrival in the penalty zone is nearly as high as that of arrival in the reward zone. This choice of τ is likely to lead to nearly as many penalties as rewards. Given that the penalty/reward ratio was 3∶1, expected gain is negative for this choice of τ. The distribution associated with the leftmost choice of τ is primarily in the uncolored time zone where the subject earns nothing. This choice of τ is likely to lead to rare rewards and extremely rare penalties, resulting in only a small total reward across many trials. Interestingly, a third choice of τ, centered on the temporal reward region, earns even less than the previous choice of τ because of a combination of its proximity to the temporal penalty, the magnitude of temporal movement noise, and the ratio of the reward to penalty magnitudes.
The best of the four choices shown is therefore the τ located at the left edge of the rewarded temporal region. Of the four shown, it makes the best compromise between the width of the probability distribution for t and its distance from the centers of the reward and penalty regions, given the widths of those regions and the ratio of gains to losses. Of course, there are infinitely many possible choices of τ. The lower panel shows the expected gain as a function of τ, with the maximum expected gain (MEG) point indicated with a circle at the peak of the expected gain function. If observers select this value τopt, they maximize their expected gain.
We computed τopt for each of the four penalty conditions and each subject based on an estimated temporal uncertainty function σ(τs) that was specific to each subject. In all cases the optimal (maximum expected gain) value of τs was shifted away from the penalty region.
The optimal Bayesian model (M0) makes full use of the temporal uncertainty function σ(τs) from each subject's training session. The five sub-optimal models use less information. M1 uses the σ(τs) calculated from each subject's training data without the experimentally added σ = 25 ms noise. M2 uses each subject's constant σ for all τs that includes the overall added σ = 25 ms noise; it uses the square root of the average of perturbed variances about the target durations measured during training. M3 uses the subject's constant σ without the experimentally added noise. M4 and M5 use a constant offset and constant offset and slope, respectively, but assume that the values of these parameters are unknown. Of course, some subjects are more accurate than others but this is explicitly taken account of in our analysis. Each model's predictions are defined in terms of performance relative to an individual's temporal uncertainty function. Subjects who are inherently poorer timers are being compared to a standard (defined by each model) that is tailored to (defined in terms of) the limits of that subject's abilities. So while there are in fact individual differences between subjects, these were removed in the design and analysis of the experiment. Because we equated subjects in this way we could analyze group data.
The predicted movement strategy, s, is therefore a function of the type(s) of temporal uncertainty information incorporated by each model Mm, the reward structure defined by the jth experimental condition (j = 1 to 4), and the temporal uncertainties measured during training for the kth subject (k = 1 to 5). Let denote the value of τ predicted by model Mm based on an estimate of timing uncertainty calculated from the assumptions of each model. For convenience, we denote the temporal uncertainty for an attempt to produce a movement duration of (using the full temporal uncertainty function based on the training trials), , as .
The models we considered are not all nested and consequently we chose a method of model comparison for non-nested models [52]–[54] that we describe next. Let denote the ith arrival time (of the 64 trials per condition) in condition j for the kth subject. The likelihood of model Mm is given by:(4)where.(5)Note however that for M4 and M5, the model likelihood must be calculated by integrating over the unknown parameters: the constant offset, , and constant offset and slope, , of the temporal uncertainty function, respectively, where the prior probability distributions over the parameters are taken to be bounded Jeffreys (uninformative) priors [55].
Let π(Mm) denote the prior probability of the mth model. Then the posterior probability of the mth model given the data is(6)and(7)is a comparison of the posterior probability of the optimal model M0 to the combined posterior probabilities of sub-optimal models: it is a measure of evidence [53] favoring the optimal model (the factor of 10 allows us to express evidence in decibels, denoted dB). A similar evidence measure can be computed for each of the sub-optimal models using the odds ratio of the probability of each sub-optimal model to the combined probability for the remaining five models (four sub-optimal and one optimal). We set the prior probabilities of the six models to be equal and computed these evidence measures.
|
10.1371/journal.pcbi.0030030 | PPARα siRNA–Treated Expression Profiles Uncover the Causal Sufficiency Network for Compound-Induced Liver Hypertrophy | Uncovering pathways underlying drug-induced toxicity is a fundamental objective in the field of toxicogenomics. Developing mechanism-based toxicity biomarkers requires the identification of such novel pathways and the order of their sufficiency in causing a phenotypic response. Genome-wide RNA interference (RNAi) phenotypic screening has emerged as an effective tool in unveiling the genes essential for specific cellular functions and biological activities. However, eliciting the relative contribution of and sufficiency relationships among the genes identified remains challenging. In the rodent, the most widely used animal model in preclinical studies, it is unrealistic to exhaustively examine all potential interactions by RNAi screening. Application of existing computational approaches to infer regulatory networks with biological outcomes in the rodent is limited by the requirements for a large number of targeted permutations. Therefore, we developed a two-step relay method that requires only one targeted perturbation for genome-wide de novo pathway discovery. Using expression profiles in response to small interfering RNAs (siRNAs) against the gene for peroxisome proliferator-activated receptor α (Ppara), our method unveiled the potential causal sufficiency order network for liver hypertrophy in the rodent. The validity of the inferred 16 causal transcripts or 15 known genes for PPARα-induced liver hypertrophy is supported by their ability to predict non-PPARα–induced liver hypertrophy with 84% sensitivity and 76% specificity. Simulation shows that the probability of achieving such predictive accuracy without the inferred causal relationship is exceedingly small (p < 0.005). Five of the most sufficient causal genes have been previously disrupted in mouse models; the resulting phenotypic changes in the liver support the inferred causal roles in liver hypertrophy. Our results demonstrate the feasibility of defining pathways mediating drug-induced toxicity from siRNA-treated expression profiles. When combined with phenotypic evaluation, our approach should help to unleash the full potential of siRNAs in systematically unveiling the molecular mechanism of biological events.
| Approaches for discovering mechanisms of action and for identifying molecular biomarkers in biomedical research are evolving today, as the growing symbiosis with computational sciences becomes more widely appreciated. In fact, the combination of various new technologies has been pushing forward both frontiers. Here, we present an example of the combined use of in vivo siRNA knock-down technology, genome-wide gene expression profiling, and computational reasoning to unveil regulatory causal relationships and the sufficiency network of identified genes for compound-induced toxicity. Unlike previously reported approaches, our method requires only one targeted perturbation for genome-wide de novo pathway discovery. Hence, our method can be directly applied to animal models in which it is still technically challenging to perform genome-wide genetic perturbation or RNAi screening. The independent application of our derived model to compounds with unrelated mechanisms of action suggests the existence of a universal molecular module that mediates liver hypertrophy. The resulting sufficiency network for induction of liver hypertrophy will have an immediate impact on the progress of toxicogenomics. When combined with phenotypic evaluation, our approach should help to unleash the full potential of siRNAs in systematically unveiling the molecular mechanisms of biological events.
| Determining the pathways underlying drug-induced toxicity from genome-wide expression profiles is essential for drug development. Knowledge of the components of the identified pathways and their sufficiency order for toxicity is crucial for selecting mechanism-based biomarkers to assess the safety of novel agents. Numerous efforts have been made to establish toxicogenomics databases, most of which consist of thousands of expression profiles produced in response to treatment with dozens or hundreds of compounds and their associated toxicities [1–5]. Classifiers based on genes showing altered expression in association with the observed toxicities have been developed to predict a variety of potential toxicities in target tissues, such as liver, kidney, and heart [1]. Although these metrics built on different machine learning algorithms could be useful in estimating the severity of potential toxicities induced by compounds, the applications of these classifiers in understanding the mechanisms of drug-induced toxicity are not straightforward. In particular, this approach is unlikely to distinguish the upstream causal genes from the downstream responsive genes among all the genes associated with an induced toxicity. Without knowledge of the causal sufficiency order, designing experiments to test predicted toxicity in animal models remains difficult.
Genome-wide RNA interference (RNAi) has emerged as an effective tool in uncovering components of pathways for specific cellular functions and biological activities, such as the innate immune response, the DNA-damage response, Golgi organization, and lifespan regulation in Saccharomyces cerevisiae, Caenorhabditis elegans, and Drosophila [6–12]. However, eliciting the causal sufficiency relationships among identified genes for a biological activity is still challenging. Phenotypic screening by RNAi has successfully identified regulators of a gene on a single-gene basis [13]; however, it is unrealistic to exhaustively examine all potential interactions by concomitant perturbations targeting each gene within a pathway, given the exponential number of potential interactions between these genes and the phenotype [6–12]. Therefore, a high-throughput approach to identify the most likely sufficiency order among these genes is critical for the application of RNAi screening in defining phenotype-orientated pathways. Such an approach is particularly important for pathway discovery in rodents, the most frequently used animal models in preclinical studies. Unfortunately, introducing genome-wide targeted genetic modifications in rodents is technically challenging.
Computational approaches have been reported to infer regulatory networks from expression profiles [14–20]. A variety of statistical modeling approaches, including Bayesian-based networks, graph theory, and ordinary differential equations, have been used for depicting or uncovering molecular networks. Collins and colleagues first developed a model for identifying the mode of action for a compound after perturbing each of the genes in the network [17]. Friedman and colleagues reported a Bayesian network approach for de novo inference of potential transcriptional networks based on 300 expression profiles obtained from yeast strains with 276 deletion mutations [18]; this approach did not require prior knowledge about the modeled network, as Collins' approach did. Approaches based on physical interactions, such as DNA–protein or protein–protein interactions, were reported by Yeang et al., Gifford, and a couple of other groups as well [20,21]. Although implemented successfully in S. cerevisiae, these published approaches depend on the accessibility of genome-wide targeted perturbations in individual model systems. Even in models in which genome-wide perturbation by genetic engineering can be easily achieved, such as yeast or fly models, it is still quite labor intensive and prohibitively expensive to use a genome-wide overexpression or deletion mutation to obtain the data required for these analytic approaches. In rodents, it is unlikely that such a dataset from genome-wide genetic permutation could be obtained with current technology. Some progress has been made with the chromatin immunoprecipitation (ChIP)–chip technology for eliciting potential transcriptional regulatory relationships; however, screening for all the transcription factors on the genome scale is impeded by the lack of specific antibodies for each of the known transcription factors.
Recent work by Segal and colleagues in identifying regulatory modules from five strains of yeast shows promise [14,15]. Using precompiled putative transcription factors, they identified potential transcriptional modules that are associated with enriched functionality, suggesting the potential impact of these genes on certain biological processes. Others have attempted to associate the derived network with specific biological processes as well [18,22–24]. Recent progress toward identifying a phenotype-oriented network in rodent models was initially reported by Schadt et al. [23]. Taking advantage of the genetic heterogeneity introduced by more than 300 crossbreeding F2 mice, the authors inferred causal genes for obesity from a combination of genetic quantitative trait loci and their associated expression profiles. The sensitivity and specificity of their approach depend on three factors: the frequency of recombination across the genome, the density of the measured genetic markers, and the coverage of the phenotypic alteration of interest among the F2 subpopulations. To ensure adequate statistical power, their approach requires large-scale cross-breeding and genotyping, yet it may not be flexible enough for inferring causal genes for drug-induced toxicities in any target tissue.
To uncover a phenotype-orientated network for any type of drug-induced toxicity in rodents, we developed a two-step relay approach to infer causal genes and their sufficiency order from small interfering RNA (siRNA) response expression profiles. In the first step, we used both the on- and off-target effects from RNAi to infer pairwise regulatory relationships from the expression profiles. This strategy allowed us to avoid the need for large-scale targeted genetic perturbations required by the majority of previously developed methods. In the second step, we used the inferred regulatory relationships among the identified gene pairs to derive causal genes for any type of drug-induced toxicity from any previously established toxicogenomics database in the rodent. Our approach provides a practical and flexible strategy for uncovering potential molecular mechanisms for drug-induced toxicity.
To derive causal genes and their sufficiency order in inducing PPARα activation–induced liver hypertrophy (AILH), we profiled mouse liver after modifying Ppara mRNA levels by treatment with two Ppara siRNAs [25]. The mRNA levels of Ppara and two of its known downstream targets, carnitine palmitoyltransferase 1A (Cpt1a) and long-chain acyl-CoA dehydrogenase (Acadl), were examined in the siRNA-treated mice as well as in Ppara knockout mice, which served as a positive control for the silencing efficacy of the Ppara siRNAs (Figure 1A). Expression of both Ppara and its known target genes decreased in a coordinated fashion among all Ppara siRNA–treated mice, and regulation between Ppara and its targets, Cpt1a and Acadl, was confirmed in the Ppara knockout mice. These data suggest that Ppara siRNA effectively downregulates transcription of the targeted gene, Ppara, and its known downstream genes in vivo. The genome-wide expression profiles in livers from the Ppara siRNA–treated mice also revealed an extensive off-target effect when compared with the profiles in livers from the Ppara knockout mice, as reported separately [26]. A large number of genes that were downregulated in the Ppara siRNA–treated liver profiles either were not downregulated or were upregulated in the Ppara knockout liver profiles, and vice versa. Hence, Ppara siRNA, delivered by tail vein, has a significant effect on both on- and off-target genes in vivo. The transcriptional response to Ppara siRNA treatment provides an ideal dataset for inferring pairwise regulatory relationship genes by local constraint algorithm (LCA) in that the external perturbation, or the instrumental variable, on the system can be measured.
Figure 2 summarizes the three major causality relationships for the impact of siRNA on PPARα, PPAR downstream genes, PPAR siRNA off-target genes, and cellular responses. The causal models corresponding to all illustrated scenarios and their combinations share a common regulatory relationship between X and Y (designated as X → Y), representing genes and/or cellular responses affected by the perturbation variable, W, or a hidden variable, H, which is not measurable in the current system (Figure 2B, 2D, and 2F). Because the mRNA levels for both the on- and off-target genes are controlled by the siRNA, we considered the effect of Ppara siRNA as the instrumental variable (W), which is manifested and can be measured by the mRNA level of Ppara in response to the siRNA treatments. Without genome-wide perturbation induced by the off-target effect of Ppara siRNA, expression of the genes associated with Ppara, such as gene X and gene Y, in the Ppara siRNA–treated profiles should highly correlate with expression of Ppara. Hence, it would be impossible to determine the regulatory relationship between gene X and gene Y. However, the off-target effects of different Ppara siRNAs often introduce additional perturbations on X and Y differentially, making it possible to derive the regulatory relationship between gene X and gene Y. For example, when two conditions—(1) the correlation between Ppara and gene Y is disrupted by the off-target effects, but the correlation between Ppara and gene X remains in some profiles, and (2) the disruption of the correlation between Ppara and gene X is parallel to the disruption of correlation between Ppara and gene Y in the other profiles—are met simultaneously for Ppara, gene X, and gene Y, then the observation, which statistically satisfies the Ppara and Y conditional independency given X, indicates that it is more likely that X regulates Y than that Y regulates X. Consequently, when W, an instrumental variable whose value is controlled by a defined input or signal, is known, the causal relationship between X and Y can be inferred computationally [27,28]. More comprehensive statistical tests, such as the LCA, to estimate the likelihood of X regulating Y given an instrumental variable W over all other possible relationships among X, Y, and any other hidden variables, have been previously established for general data-mining purposes [27]. Our two-step relay approach used LCA to infer all pairwise causal relationships between a top gene (X) and a bottom gene (Y) in the first step, and a causal gene (X) and its cellular response (Y) in the second step.
Specifically, we used LCA to identify gene pairs in which one was conditionally independent of Ppara when the other was conditioned on Ppara (Figure 3). When conditioned on the identified top genes, the correlation between the bottom gene and siRNA effect, manifested by the mRNA level of Ppara, vanished (Figure 3A, red dots). However, when conditioned on the bottom genes, the correlation between Ppara and the selected top genes remained significant (Figure 3B, red dots). Among approximately 25,000 genes profiled, 5,650 transcripts regulated by Ppara siRNA in at least one mouse liver (p < 0.01 revealed by a platform error model) were tested for potential pairwise relationships by LCA (Table S1). A total of 22,256 gene pairs that satisfied all of the six criteria specified in Materials and Methods were selected as the top–bottom, or upstream–downstream, gene pairs (X → Y) from the mouse expression profiles. These gene pairs with inferred regulatory relationships accounted for about 0.07% of all potential pairs among the 5,650 transcripts tested.
In the second step of our approach, we inferred the causal genes for liver hypertrophy induced by PPARα activation by using the derived regulatory relationship within these gene pairs. Of the 5,650 mouse transcripts tested, 2,569 had reciprocal orthologs on the rat chip used for profiling the liver responses to PPARα agonists (Table S2). A total of 4,727 gene pairs with inferred regulatory relationships were identified among the 2,569 mouse genes with rat orthologs. Similarly, these gene pairs represented 0.07% of all potential pairs among the 2,569 genes. Assuming that the regulatory relationships derived in the mouse liver hold in the rat liver, we selected gene pairs only if there existed conditional independency between gene X and hypertrophy given gene Y in the PPAR minicompendium [29] and not conditional independency between gene X and gene Y given hypertrophy (Figure 3D–3F). Specifically, for all of the selected gene pairs, the dependency between the top genes and liver hypertrophy was abolished when conditioned on the bottom genes (p < 0.01) (Figure 3D, red dots). The dependency between the top genes and bottom genes remained when conditioned on liver hypertrophy (Figure 3E). So did the dependency between liver hypertrophy and the bottom genes (p < 0.01) when conditioned on the top genes (Figure 3F). The Y genes in the selected gene pairs were identified as the causal genes for PPARα–AILH if they occurred in at least one gene pair. We identified 39 transcripts as candidates most likely mediating PPARα–AILH (p < 0.01), and 16 of these transcripts—corresponding to 15 genes and one expression sequence tag (EST)—were significant at the level of p < 5 × 10−4 (Table 1).
To check for false-positives among the derived relationships after multiple tests, we conducted a Monte Carlo simulation. The false-positive rate was estimated based on 1,000 simulations testing all 4,727 gene pairs. The chance of false discovery for the inferred causal genes mediating the PPARα–AILH was 0.002.
To validate our assumption that the regulatory relationship between gene pairs inferred in the mouse was retained in the rat, the conditional independencies between PPARα activity (measured by PPARα direct targets) and bottom genes given top genes for all 16 identified transcripts were examined in the rat liver PPAR minicompendium. Nine (56%) of the 16 transcripts satisfied the statistical tests. The remaining seven transcripts may have been excluded as a consequence of the different criteria used for measuring PPARα perturbation between the mouse and rat studies. For example, in the mouse study, PPARα activity was measured directly by the mRNA level in siRNA profiles, whereas in the rat study, it was measured indirectly by the mRNA levels of the PPARα target genes. Nevertheless, 11 of the 15 identified genes have comprehensive functional annotations, and eight of these 11 genes (73%) have been reported either to be associated with or to cause liver hypertrophy or hepatomegaly [30–32].
We assumed that liver hypertrophy is mediated by a common essential molecular mechanism, even if it can be induced by differing top signals. Therefore, there should be a set of core genes that are essential for liver hypertrophy regardless of whether the hypertrophy results from PPARα activation or other conditions. If some of the 16 identified transcripts are essential for liver hypertrophy, we would expect them to have predictive power for compound-induced liver hypertrophy regardless of the compounds' mode of action; otherwise, the derived causal effect of these 16 transcripts for liver hypertrophy cannot be sustained. To test this hypothesis, we profiled 211 rats that were treated with 30 non-PPAR compounds and determined their liver/body weight ratio, as an indicator of liver hypertrophy (Table S3). A logistic regression-based classifier for liver hypertrophy was built based on nine of the 16 identified causal transcripts from the 211 profiles as the training set. The optimal number of transcripts among the 16 inferred transcripts was determined by their contribution to the fit of the model for predicting liver hypertrophy. The accuracy of the model was measured by the root-mean-squared error estimated by leave-one-out validation. Nine transcripts were selected to build the best-fitted model with the least difference between the measured and predicted ratios of liver to body weight in the training set. The built model predicted liver hypertrophy in the training set with 96% sensitivity and 80% specificity (Figure 4B, Table 2).
To further validate this model, we assessed its performance in an independent testing dataset. This dataset consisted of the nine genes and liver/body weight ratios from another set of 107 rat liver profiles obtained in response to treatment with 17 non-PPARα compounds (Table S4). These compounds have mechanisms of action or targets that differ from those of the 30 non-PPARα compounds used in the training set. The nine-gene model built in the training set achieved 84% sensitivity and 76% specificity in predicting liver hypertrophy in the independent testing set (Figure 4C and Table 3). This finding suggests that we can predict liver hypertrophy based on this model with an acceptable level of confidence regardless of the type of top signals that induce the liver hypertrophy.
Of note, the selection of the 16 candidate transcripts on which the model was built was not dependent on either of the datasets used for training and testing. To determine if the causal information is necessary for such performance in predicting liver hypertrophy, we examined the possibility of achieving an equivalent level of accuracy in predicting liver hypertrophy based on genes selected solely by their association or correlation with the liver/body weight ratio in the PPAR minicompendium. Based on correlation with liver/body weight ratio, we identified 757 genes that had a value equal to or greater than that of the 16 inferred causal transcripts in the PPAR minicompendium. The identified genes were used as a candidate pool to build a correlation-based model. In total, we built 10,000 models based on 10,000 sets of 16 genes randomly selected from the candidate pool. Using the area under the curve (AUC) from a receiver operating characteristic (ROC) plot (a common indicator for model performance), we estimated the distribution of the performance for these 10,000 models. This in silico simulation revealed that none of the models in the training set, and only 37 of the models in the testing set, achieved a level of performance at least equal to that of the model based on inferred causal genes in predicting liver hypertrophy. The probability of building a model with at least the same level of performance as that built from the inferred causal transcripts was exceedingly small for both the training set (p < 0.001, Figure 4E) and the testing set (p < 0.005, Figure 4F). Hence, correlation alone is insufficient for extrapolating the capability of the 16 inferred PPARα–AILH causal transcripts to predict liver hypertrophy induced by non-PPAR compounds. The predictive power of the inferred PPARα–-AILH causal transcripts for non-PPAR–induced liver hypertrophy supports the assumption that essential biological components for liver hypertrophy exist. The inferred causality relationship between the 16 transcripts and liver hypertrophy is essential for the extrapolation of this prediction model.
Finally, we determined the sufficiency order for mediating PPARα–AILH among the 16 transcripts by exhaustive pairwise conditional independent (CI) tests (Figure 5 and Figure S1). There were 15 genes with known or predicted proteins among the 16 derived causal transcripts. CI tests identified nine of these genes as the most sufficient ones in inducing liver hypertrophy. For instance, as revealed by conditional dependency between Pck1 and PPARα–AILH given the constraint of the inferred causal relationship between Acadm and PPARα–AILH, Pck1 is more sufficient than Acadm in mediating PPARα–AILH (Figure S1A).
This order is also in agreement with observations in Pck1 null mice and Acadm null mice. Acadm, also called medium-chain acyl-CoA dehydrogenase (MCAD), encodes an essential enzyme for fatty acid oxidation. MCAD deficiency in mice or humans can cause liver hypertrophy and hepatic steatosis [32,33]. Meanwhile, Pck1, also known as phosphoenolpyruvate carboxykinase (PEPCK), encodes a rate-limiting enzyme for gluconeogenesis. PCK1 deficiency is lethal in both mice and humans [34,35]. However, liver hypertrophy and steatosis have been observed in liver-specific PCK1-deficient mice [34]; after 14 hours of fasting, such mice have a 71% increase in liver weight along with significant hepatic lipid accumulation. Furthermore, although these mice have elevated expression of MCAD, this does not reverse the gain in liver weight induced by PCK1 deficiency. Similarly, an exhaustive search over all potential gene pairs between Pck1 and the other inferred causal genes for PPARα–AILH revealed that Pck1 is more sufficient in mediating liver hypertrophy than are Pex19, Car, Pitrm1, and 3110048E14Rik.
Based on our knowledge or lack of knowledge about their functions, the nine genes derived to be most sufficient in mediating liver hypertrophy represent four groups: a fatty acid, lipid genesis, or energy metabolism group, including Pck1, Apoa4, Scd1, Acac, and Lpin2; a cell proliferation group, Cdc2a, also known as cdk1; a Golgi group, Golga4; and a group with unknown function, 2010309E21Rik and 9430083G14Rik. Five of these genes, Pck1, Apoa4, Scd1, Acac and Cdc2a, have been reported to be modified by gene-targeting or antisense oligonucleotide treatment, and in all five cases the resulting phenotypic alterations in the liver are consistent with the gene's inferred causal effect on liver hypertrophy as determined by our method [31,33,34,36–40].
In a final analysis, we examined potential reduction of statistical power associated with multiple tests for the genes with the maximal number of downstream genes mediating PPARα–AILH. For the nine genes that exhibited the highest sufficiency in mediating liver hypertrophy, the estimated probability that an additional causal gene from the rest of the 16 transcripts could be more sufficient for the hypertrophy was p = 0.07.
Overall, our genome-scale approach provides a powerful tool for unveiling components of potential pathways and the order of their sufficiency relationships in mediating a drug-induced toxicity. In selecting an example, we applied our approach to identify the key molecular components and their sufficiency order in PPARα–AILH. The proposed 15 causal genes not only reveal the molecular mechanism of this phenomenon, which is in agreement with that shown in previous studies, but also suggest potential unknown common mechanisms underlying liver hypertrophy. The inferred causality hypotheses orientated to a certain type of toxicity, in this case, liver hypertrophy, are essential in discovering the potential mechanism of drug-induced toxicity.
Computational approaches for de novo discovery of regulatory networks have been previously reported [14–20]. A few have attempted to associate the derived network with specific biological outcomes [18,22–24]. Nevertheless, application of these approaches in rodents is impeded by their requirement for a large number of targeted genetic perturbations or prior knowledge of the pathway. Our approach overcomes such hindrance by taking advantage of the siRNA off-target effects as additional genome-wide perturbations. Using the mRNA level of an siRNA-targeted gene, in the current case, Ppara, as the instrumental variable, in the siRNA-treated expression profiles, we were able to infer pairwise regulatory relationships by a simple but well-established LCA originally developed for data mining [27].
Instead of approximating a global regulatory structure, we focused on the sufficiency order among the causal genes identified by the two-step relay approach. Taking advantage of the derived causal structure from the causal gene to the liver hypertrophy as local constraints, the searching space for the 16 derived transcripts was reduced from O(2g) to O(g2), where g is the number of derived causal genes. In our case, the searching space was reduced by 272-fold. This dramatic reduction enabled us to exhaustively search for the sufficiency network, which was, otherwise, computationally expensive or simply infeasible. In contrast, most of the previously reported approaches require Monte Carlo Markov chain or other sampling techniques to search for a local optimal network as an approximation of the global optimal structure.
Moreover, the derived pairwise regulatory relationships from siRNA expression profiles can also benefit most of the currently used computational approaches for pathway de novo discovery. For example, application of these relationships as local constraints in Monte Carlo Markov chain–based Bayesian methods will reduce the searching space and consequently improve the chance that any resulting local optimal structure is the real global solution.
The two-step relay strategy offers unprecedented flexibility, although the assumption of retained pairwise relationships among the gene pairs derived from the first step is essential and needs to be experimentally examined. Because the derived regulatory relationship between the top (upstream) gene and the bottom (downstream) gene reflects a transcriptional regulatory relationship within a gene pair, these relationships are independent from the instrumental signal, in our case, Ppara. As a result, such regulatory relationships can be used to infer causal genes for any phenotypic changes induced by compounds or other types of perturbation. This portability of the inferred regulatory relationship within gene pairs is extremely helpful for toxicogenomics study because the majority of toxicity studies, including gene expression profile studies, are performed in rats, whereas knockout or RNAi silencing is most commonly undertaken in mice.
In this work, we used perturbation introduced by siRNA targeting one gene. It is conceivable that the coverage and reliability of pairwise relationship across the genome can be further improved if multiple siRNAs targeting different genes are applied. Meanwhile, other strategies can be applied to experimentally evaluate the accuracy of the inferred causal relationship. For example, genetic engineering of top genes in C. elegans or Drosophila can be used to verify the regulatory effect of top genes on bottom genes, the key assumption of our method. Monitoring the temporal relationship between the expression regulations of these inferred causal genes and the phenotypic outcomes in liver will enable further validation of the inferred causal network in rodents as well. Consequently, a genome-wide phenotype-oriented network can be inferred and tested in a relatively high throughput fashion.
Classification does not prove a causal relationship. Yet, only a causal relationship will enable extrapolation of an established model into prediction. Hence, capability of extrapolation can be used as a test of the derived causal relationship. Many classifiers or gene sets, established by vigorous machine learning procedures, have been reported to describe the extent of drug-induced toxicity or the outcome of cancer with reasonable accuracy. However, they do not allow us to distinguish whether these identified gene sets have a functionally important role in mediating those observed outcomes or are merely indicators for the underlying propensity of those outcomes. The performance of such types of classifiers for predicting drug-induced toxicity will be affected by the diversity of compounds used in the training set and testing set. When toxicity has to be predicted for compounds affecting targets different from the targets affected by compounds in the training set, the performance of a classifier not based on causality information will deteriorate dramatically. In other words, a classifier that is not based on causality will rarely have the capability of extrapolation. Our comparison between classifiers based on genes selected by inferred causality and by correlation alone demonstrates that causality information is necessary for compound-independent extrapolation in toxicity prediction (Figure 4). The performance of classifiers built on genes selected from the 757 genes correlated with liver/body weight ratio in the PPAR minicompendium became substantially inferior when these classifiers were used to predict liver hypertrophy induced by non-PPAR compounds. Our simulation indicates that classifiers based on correlated genes alone are unlikely to successfully predict or extrapolate drug-induced toxicity independently from the mode of action of compounds used in training set (p < 0.005). In contrast, the classifier we built based on inferred causal genes can predict liver hypertrophy independent from compounds' modes of action in the original training set. The capability of extrapolating a model that was initially built on PPAR compounds into predicting liver hypertrophy regardless of a compound's primary targets has not been demonstrated in any previous reports and has not been examined for classifiers based on association markers.
The legitimacy of using the capability of extrapolation as an analytic validation approach for the inferred causal genes is further supported by experimental evidence for the inferred 16 causal transcripts in in vivo models [31,33,34,36–40]. More than 85% (six of seven) of the inferred causal genes that have been genetically modified or specifically targeted in in vivo models are associated with phenotypic abnormalities in the liver that are in agreement with the role of these genes in mediating liver hypertrophy (Table 1). Five of them are among the genes with highest sufficiency in causing liver hypertrophy. In contrast, the rate of discovering functionally related genes among the top-ranked markers in an association-based classifier is quite low. For instance, Furey and colleagues detected only three genes that are cancer-related among the top ten most discriminative genes or ESTs for classifying ovarian cancer [41]. The rate for cancer-related genes or ESTs may be underestimated because they exclude noncoding sequences, which have been recently shown to play an important role in tumorigenesis. Nonetheless, the lack of evidence for the capability of extrapolating their association-based model into predicting outcomes of other solid tumors is in agreement with the contention that causality is a prerequisite for the capability of extrapolation in predicting outcomes.
Although we have made every effort to analytically validate the derived results of causal relation in general, several factors may affect our results. First, the assumption that the derived pairwise regulatory relationship between a top gene and a bottom gene is valid in both mouse and rat liver may not hold all the time. To gain confidence in this assumption, we examined the statistical conditional independence between PPARα activity, measured by direct targets of PPARα, and bottom genes given top genes for all 16 identified transcripts using the PPAR minicompendium. We found that nine of the 16 transcripts were more likely to retain the same top–bottom relationship as derived from the siRNA profiles. Of note, the original 16 gene pairs were established by treating the mRNA level of PPARα as a direct measurement of perturbation on Ppara in siRNA-treated profiles, whereas the nine gene pairs were reestablished by treating mRNA levels of Cpt1a and other genes known to be PPARα downstream genes as an indirect measurement of perturbations on Ppara in the PPAR minicompendium. Second, statistical power could be reduced with multiple tests. For example, false-positives may occur among the inferred causal genes for PPARα–AILH after testing all the 4,727 gene pairs. The estimated false-positive rate for the 16 identified transcripts was 0.002. Third, a measurement error associated with expression profiles could affect the sensitivity and specificity of our analysis. To account for this error, we used error-weighted mean logarithmic ratios in our analysis. Cross-validation from different platforms or independent experiments will help further address this issue.
The five identified most sufficient causal genes for PPARα–AILH suggest that three major molecular mechanisms underlie the increases in liver weight mediated by this receptor: increased cell proliferation, altered fatty acid or lipid metabolism, and affected function of the Golgi apparatus. In agreement with observations in humanized PPARα mice treated by PPARα ligand Wy-14643, our results suggest that Cdc2a (cdk1), a cell cycle control gene, may increase liver weight by enhancing cell proliferation. As shown by Gonzalez and colleagues, replacing murine PPARα with human PPARα dramatically decreases the number of proliferating cells and inhibits the level of cdk1 in the presence of Wy-14643 [37]. Although they clearly demonstrated the rodent specificity for the effect of PPARα activation in cell proliferation and tumorigenesis, the downstream genes mediating the proliferation effects of Wy-14643 remained largely unknown [37,38]. Our study provides a testable hypothesis for the role of Cdc2a or cdk1 in mediating the rodent-specific tumorigenic effect of PPARα activation. A cross between humanized PPARα mice and liver-specific Cdc2a knockout mice will be a perfect model to evaluate the hypothesized contribution of Cdc2a to PPARα activation–induced cell proliferation and tumorigenesis.
The contribution of fatty acid or lipid metabolism to PPARα–AILH, suggested by our observations, is also supported by phenotypic changes reported in several genetically modified mouse models, including the mice with humanized PPARα. For instance, both Acadm, a gene critical for fatty acid metabolism, and the extent of liver hypertrophy, measured by either the absolute liver weight or the liver/body weight ratio, are increased by PPARα ligand Wy-14643 in mice with or without humanized PPARα despite the fact that cell proliferation is dramatically reduced with the humanized receptor PPARα [37,38]. Furthermore, for inferred causal genes downregulated by PPARα ligands, such as Pck1 and Apoa4, knocking out these genes results in significant increases in liver weight and liver/body weight ratio or in fatty liver [34,39]. The increased liver weight or size resulting from deficient mutations of these two genes, and the fact that PPARα activation reduces their expression, supports their inferred role in liver hypertrophy. Additional evidence comes from the response to the high-fat diet, the endogenous ligand for PPARα, in mice treated with an antisense oligonucleotide for Acac1 and Scd1 [36,40]. As the key regulators for fatty acid oxidation and fat synthesis in the liver, Acac1 and Scd1 are upregulated by PPARα activation as well as by a high-fat diet. Increased expression of these genes results in increases in liver size or weight by lipid accumulation in hepatocytes. Antisense oligonucleotide inhibitors of Acac1 and Acac2 totally reverse this induced lipid accumulation, suggesting that Acac1 mediates the effect of PPARα activation on lipid accumulation in liver [40]. Similar results have been reported in mice treated with an antisense oligonucleotide inhibitor of Scd1 [36]. Injection of 15 mg/kg of this oligonucleotide twice a week significantly reduced the lipid accumulation in the liver induced by a high-fat diet. Taken together, these reported observations are not only in concert with our inferred role of fatty acid metabolism and lipid synthesis in mediating PPARα–AILH, but may also reveal a potential key pathological change, lipid accumulation, linking liver hypertrophy with fatty liver.
The role of Golga4 in mediating PPARα–AILH remains unknown because functional studies of this gene are lacking. However, the essential role of Golgi traffic in regulating the activation of sterol regulatory element–binding protein (SREBP), another key regulator of lipid metabolism, points in the same direction.
Distinguishing the proliferation component from the fatty acid and lipid metabolism component for liver hypertrophy is important for drug development. For example, compounds are being developed to target Pck1 as novel therapeutic agents for type II diabetes or obesity. Distinguishing the two components enables us not only to predict the potential effect on liver weight by compounds directly targeting Pck1 (Pepck), a potential therapeutic target regulated by PPARα and the rate-limiting enzyme for gluconeogenesis, but also to monitor any cellular proliferation associated with increased liver weight.
The causal sufficiency order among the 15 derived genes suggests that liver hypertrophy induced by compounds with drastically different mechanisms of actions may be mediated by common molecular components. For example, we found that Car, another nuclear receptor whose activation is also known to cause liver hypertrophy [31], was more sufficient than Pex19, a gene required for the synthesis of the peroxisome membrane, in mediating PPARα–AILH. To our knowledge, the effect of PPARα ligands on the liver in mice with a Car-deficient mutation has never been reported; however, liver hypertrophy induced by PPARα activation and by Car activation might share the same mechanism, given the extensive crosstalk between PPARα and Car ligands and their very similar effects on some of the genes regulating fatty acid or lipid metabolism. Finally, the capability to extrapolate a model built on causal genes for PPARα–AILH to predict liver hypertrophy induced by a set of non-PPAR compounds with different mechanisms of action supports the existence of common essential molecular machinery that mediates liver hypertrophy independent of upstream signals.
In conclusion, our approach enables systematic inference of testable hypotheses about potential molecular mechanisms underlying drug-induced toxicity. The validity of any inferred hypotheses will still be subject to experimental verification. Nonetheless, the derived causal sufficiency order provides a way to prioritize all these hypotheses, which otherwise cannot be tested exhaustively across the prohibitive number of potential interactions and combinations among the genes associated with drug-induced effects.
Two different siRNAs targeting Ppara and two control siRNAs targeting secreted alkaline phosphatase (SEAP) and GL3 were delivered to the mouse liver by hydrodynamic tail vein injection [25]. Four animals per group were treated 48 hours before livers were collected. RNAs were extracted and profiled on a custom-designed mouse oligonucleotide microarray with approximately 25,000 genes according to a published procedure [29]. In total, the Ppara siRNA mouse expression profiles included eight profiles for the Ppara siRNAs targeted to two different fragments of the Ppara mRNA sequence, four profiles for the SEAP control siRNA, four profiles for the GL3 control siRNA, and eight profiles for a vehicle control (Ringer solution).
In brief, the PPAR minicompendium used in this study consists of expression profiles of livers from 147 rats treated with activators of PPARs and other nuclear receptors, including the PPARα agonists fenofibrate (400 mg/kg per day; Sigma-Aldrich, http://www.sigmaaldrich.com) and Wy-14643 (100 mg/kg per day; ChemSyn Sciences, http://www.chemsyn.com), the PPARγ agonist rosiglitazone (100 mg/kg per day), the pan-PPAR agonist bezafibrate (250 mg/kg per day; Sigma-Aldrich), and the retinoic acid receptor agonist all-trans retinoic acid (25 mg/kg per day; Sigma-Aldrich). Female Sprague Dawley rats (Charles River Laboratory, http://www.criver.com) 7–8 wk of age were dosed daily for 2, 3, 4, 5, and 6 days by oral gavage. Six hours after the last dose, animals were euthanized by isoflurane inhalation followed by blood collection from the vena cava and exsanguination. Liver weight and body weight were measured for all animals, and the ratio of liver to body weight was used as the index for liver hypertrophy. RNAs were extracted from liver and profiled on a custom-designed rat oligonucleotide microarray with approximately 25,000 genes according to a published protocol [29]. The error-weighted ratios for the 2,569 transcripts used in our study are summarized in Table S2. A similar study was conducted using 318 liver samples from rats treated with 47 compounds whose primary targets were not PPARs, hereafter referred to as non-PPAR compounds. Data used in our analyses are summarized in Table S3. All the animals used in these experiments were handled in laboratories accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International. The animal studies were conducted with Institutional Animal Care and Use Committee approval. A comprehensive summary of related data and analyses for this PPAR rat liver minicompendium were reported separately [29].
The three causal models corresponding to the major scenarios for the impact of siRNA on other genes and cellular responses (Figure 2) share a common regulatory relationship between X and Y (denoted as X → Y). Cooper [27] and Spirtes and colleagues [28] have demonstrated that when W is an instrumental variable, or when W is not caused by any measurable variables, such as X and Y, the causal relationship between X and Y can be inferred by a constraint-based causal discovery method. Specifically, when six dependent and independent constraints among W, X, and Y hold—(1) D(W, X), representing dependence between W and X, (2) D(X, Y), (3) D(W, Y), (4) D(W, X given Y), representing dependence between W and X when conditional on Y, (5) D (X, Y given W), and (6) I (W, Y given X), representing independence between W and Y when conditional on X—it was sufficient to identify the causal relationship X → Y from 96 possible regulatory models among the triplet W, X, and Y elicited by Cooper [27]. Because the mRNA levels of siRNAs' on- and off-target genes are presumably controlled by the siRNA, the mRNA level of the siRNA-targeted genes could be considered a good measure for the on-target effect of the siRNA. Therefore, the siRNA can serve as the instrumental variable (W). Furthermore, it was assumed that the relationship between X and Y was not confounded by W or any hidden process in the possible scenarios for the impact of the siRNA. Therefore, those pairwise causal relationships between a top gene (X) and a bottom gene (Y), or a causal gene (X) and its cellular response (Y), affected by siRNA targeting a single gene can be inferred by LCA from expression profiles on the genome scale [27].
Genes mediating PPARα–AILH were inferred by a two-step relay method as illustrated in Figure 2G. In the first step, the pairwise causal relationship (X → Y) between genes was inferred by LCA from the expression profiles obtained in response to Ppara siRNA treatment. In the second step, the derived relationship X → Y was relayed to identify the local causal structure among the top gene (X), the bottom gene (Y), and the phenotypic response (R), liver hypertrophy. The top gene (X) was then considered as the instrumental variable W, the bottom gene (Y) was considered as X, and the response (liver hypertrophy) was considered as Y in our causal model (Figure 2). Similarly, as specified by Cooper's LCA, when the six previously defined constraints hold, it is sufficient to infer the causal relationship Y → R, with which the bottom gene Y causes R (i.e., liver hypertrophy), in the second step from the PPAR minicompendium. The key assumption that permits the relaying is that the regulatory relationship, X → Y, within the gene pairs derived from the mouse liver expression profiles is valid in the rat liver expression profiles.
The sufficiency order among causal genes (Y, Y′, Y″, …) was determined by a series of CI tests between two causal genes and the cellular response. Specifically, when two criteria are met—(1) a causal gene Y and a cellular response R are independent given another causal gene Y′, and (2) gene Y′ is not independent with cellular responses conditioning on gene Y—gene Y′ was considered to be more sufficient than gene Y in causing the cellular response R. By exhaustively examining all pairwise relationships among causal genes (Y, Y′, Y″, …), the sufficiency order network among all identified key genes can be established.
Gene ID numbers from the Entrez Gene database (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=gene) for the genes mentioned in this paper are: Acac (31), Acadl (33), Acadm (34), Apoa4 (337), Cdc2a (12534), Cpt1a (1374), Golga4 (2803), Lpin2 (9663), Nrli3 (9970), Pck1 (5105), Pex19 (5824), Pitrm1 (69617), Ppara (5465), Scd1 (20249), 2010309E21Rik (66488), 2700017I06Rik (209630), 3110048E14Rik (73225), and 9430083G14Rik (68817). |
10.1371/journal.ppat.1006520 | The role of co-opted ESCRT proteins and lipid factors in protection of tombusviral double-stranded RNA replication intermediate against reconstituted RNAi in yeast | Reconstituted antiviral defense pathway in surrogate host yeast is used as an intracellular probe to further our understanding of virus-host interactions and the role of co-opted host factors in formation of membrane-bound viral replicase complexes in protection of the viral RNA against ribonucleases. The inhibitory effect of the RNA interference (RNAi) machinery of S. castellii, which only consists of the two-component DCR1 and AGO1 genes, was measured against tomato bushy stunt virus (TBSV) in wild type and mutant yeasts. We show that deletion of the co-opted ESCRT-I (endosomal sorting complexes required for transport I) or ESCRT-III factors makes TBSV replication more sensitive to the RNAi machinery in yeast. Moreover, the lack of these pro-viral cellular factors in cell-free extracts (CFEs) used for in vitro assembly of the TBSV replicase results in destruction of dsRNA replication intermediate by a ribonuclease at the 60 min time point when the CFE from wt yeast has provided protection for dsRNA. In addition, we demonstrate that co-opted oxysterol-binding proteins and membrane contact sites, which are involved in enrichment of sterols within the tombusvirus replication compartment, are required for protection of viral dsRNA. We also show that phosphatidylethanolamine level influences the formation of RNAi-resistant replication compartment. In the absence of peroxisomes in pex3Δ yeast, TBSV subverts the ER membranes, which provide as good protection for TBSV dsRNA against RNAi or ribonucleases as the peroxisomal membranes in wt yeast. Altogether, these results demonstrate that co-opted protein factors and usurped lipids are exploited by tombusviruses to build protective subcellular environment against the RNAi machinery and possibly other cellular ribonucleases.
| Positive-strand RNA viruses build membranous replication compartment to support their replication in the infected hosts. One of the proposed functions of the usurped subcellular membranes is to protect the viral RNA from recognition and destruction by various cellular RNA sensors and ribonucleases. To answer this fundamental question on the putative role of co-opted host factors and membranes in protecting the viral double-stranded RNA replication intermediate during replication, the authors took advantage of yeast (Saccharomyces cerevisiae), which lacks the conserved RNAi machinery, as a surrogate host for TBSV. The reconstituted RNAi machinery from S. castellii in S. cerevisiae was used as an intracellular probe to study the effect of various co-opted cellular proteins and lipids on the formation of RNAi-insensitive replication compartment. Overall, the authors demonstrate the interaction between the RNAi machinery and the viral replicase complex, and the essential roles of usurped host factors in protecting the viral dsRNA replication intermediate from RNAi-based degradation.
| One of the hallmark features of positive-strand (+)RNA viruses, including tomato bushy stunt virus (TBSV), is to assemble numerous membrane-bound viral replicase complexes (VRCs) that leads to replication of the viral genomic RNA inside the infected cells. These viruses co-opt subcellular membranes and alter lipid metabolism in addition to usurping host proteins to form replication compartment or organelle. For several viruses, including TBSV, the extensive replication compartments contain many membranous vesicle-like structures, also called spherules, which are 50–100 nm invaginations with a narrow opening towards the cytosol. Other viruses have membranous, protrusion-type structures with single- or double-membrane structures formed via major membrane rearrangements [1–5]. Regardless of the structure of these replication organelles, it has been proposed that these elaborate membranous structures serve as platforms to assemble VRCs and to concentrate viral and host components for more efficient viral RNA synthesis. In addition, VRCs might also hide the viral RNAs from recognition by the antiviral surveillance system and protect against degradation by cytosolic ribonucleases.
In case of plant and insect viruses, the viral RNA-triggered adaptive innate immune response, called RNAi or RNA silencing response limits viruses to replicate and spread in infected tissues [6–11]. RNAi also contributes to viral RNA recombination and defective viral RNA production [12,13]. The viral double-stranded (ds)RNA replication intermediate or long structured portions of ssRNAs are recognized by the core components of RNAi, which consist of the Dicer-like (DCL) ribonuclease and Argonaute (AGO)-like proteins with RNA slicing activities [7,14–17]. The Dicer-like nucleases process these RNAs into 21–24 nt dsRNAs, called small-interfering siRNAs, via their RNase III activities. Then, the siRNAs are incorporated into the RNA-induced silencing complex (RISC) that contains the AGO proteins. Then, the RISC recognizes the target ssRNAs, followed by slicing and destruction of the target RNA by the RNase H-like activity of AGO [7,8,14]. The core RNAi components and pathways are conserved from some fungi (Neurospora and others) and plants to invertebrates and mammals [7,10,18,19].
TBSV is a well-characterized plant (+)RNA virus with one 4.8 kilobase genomic (g)RNA, which codes for only five proteins, two of which are replication proteins, namely p33 and p92pol. The p92pol RNA-dependent RNA polymerase (RdRp) is expressed from the gRNA via a translational readthrough mechanism of the p33 stop codon [20–22]. The pre-readthrough protein p33 is an RNA chaperone, and a membrane-associated RNA-binding protein that functions as a master regulator of TBSV replication. Accordingly, p33 plays a role in every step of replication, including in viral RNA recruitment, VRC assembly and RNA synthesis [23–26]. Through its interactions with other p33 and p92 molecules, cellular lipids and 50–100 host proteins, p33 is involved in the formation of the replication compartment [24,27–30]. The tombusvirus VRC contains several host proteins [30–32], including heat shock protein 70 (Hsp70), glyceraldehyde-3-phosphate dehydrogenase (GAPDH), eukaryotic elongation factor 1A (eEF1A), eEF1Bγ, DEAD-box RNA helicases, and the ESCRT (endosomal sorting complexes required for transport) family of host proteins [32–39]. These proteins are required for VRC assembly or affect viral RNA synthesis [27,34,37,40–42]. The TBSV replication process also depends on phospholipids, mainly phosphatidylethanolamine (PE), and sterols, which are actively enriched within the viral replication compartment [43–47].
It is assumed that (+)RNA viruses avoid the powerful RNAi response in plants by forming VRCs and replication compartments hidden away from RNAi and also via expression of suppressors of RNAi, which are part of counter defense strategies against RNAi [9,48,49]. Accordingly, the dsRNA replication intermediate formed during TBSV replication [50], is part of the membrane-bound VRCs. A long-standing unanswered question is that which co-opted host components are involved in VRC assembly that protect (+)RNA viruses during replication from the potent RNAi response. The protection provided by VRCs seems significant, because plant (+)RNA viruses could replicate and accumulate even in the absence of RNAi suppressors [51,52]. Moreover, the Dicer-like enzymes could not completely degrade the viral dsRNA in infected cells, indicating that the viral dsRNA enjoys significant protection during replication.
To answer this fundamental question on the putative role of the co-opted host factors in protecting the viral dsRNA during replication, we took advantage of yeast (Saccharomyces cerevisiae), which lacks the RNAi machinery, as a surrogate host for TBSV. Co-expression of a TBSV replicon (rep)RNA with p33 and p92pol replication proteins leads to robust TBSV replication in yeast [53,54], helping the dissection of the roles of subverted host factors in virus replication and virus–host interactions [27,28,55]. We have used the reconstituted RNAi machinery from S. castellii, which consists of the two-component DCR1 and AGO1 genes [56], as a simple, easily tractable system to study the effect of RNAi on TBSV in yeast. Based on this surrogate host system, we show evidence that deletion of a selected group of host factors, namely ESCRT proteins, and alteration of lipid levels, including PE and sterols, greatly affect TBSV accumulation when RNAi activity is induced. Based on our results, we propose that the co-opted host factors are critical for TBSV to assemble membranous VRCs that protect against RNAi activity.
Many co-opted host components are involved in tombusvirus VRC assembly [27,28,55] and they likely protect TBSV RNA from nucleases and antiviral responses during replication. To test this model, we adapted the simple RNAi machinery from S. castellii [56] as an intracellular probe to measure if a given co-opted host factor contributes to the protection of the viral RNA. Briefly, constitutive co-expression of S. castellii DCR1 and AGO1 from TEF1 promoter led to complete inhibition of TBSV repRNA accumulation in wt yeast (S1A Fig), whereas separate expression of DCR1 and AGO1 did not interfere with TBSV repRNA accumulation, suggesting that co-expression of the two components is required for the RNAi machinery. A more suitable strategy for this research was based on the inducible co-expression of S. castellii DCR1 and AGO1 from GAL1 promoter from plasmids, which could be suppressed by the addition of glucose and induced by the addition of galactose to the culture media (Fig 1A, lanes 5–6). In this system, after the induction of the expression of both DCR1 and AGO1, the induced RNAi pathway moderately inhibited viral RNA accumulation in wt yeast (Fig 1A, lanes 1–6), likely due to the protection provided by the membranous VRCs. The accumulation of the expected 23 bp vsiRNA demonstrates the operation of the RNAi machinery in this yeast strains expressing either DCR1 alone or co-expressing DCR1 and AGO1 (S1B and S1C Fig).
TBSV recruits the cellular ESCRT machinery to deform membranes and build spherules containing VRCs within the replication compartment [57,58]. First, the Vps23p ESCRT-I or Bro1p ESCRT accessory protein are recruited via direct binding to p33 replication proteins, followed by recruitment of the ESCRT-III proteins [38,58,59]. Then, Vps4p AAA ATPase is bound by p33, which likely stabilizes the neck structure of the spherule, to prevent scission and closure of the neck [57].
To test if cellular ESCRT factors required for TBSV-induced spherule formation are important for protecting the viral RNAs from RNAi-based degradation, we launched TBSV replication in vps23Δbro1Δ yeast [58]. Since the co-opted ESCRT factors affect the absolute level of TBSV replication in yeast [38,58,59], we calculated the extent of reduction in TBSV RNA protection level in the presence of the RNAi machinery based on the repRNA accumulation in the corresponding yeast strain control, not expressing the RNAi machinery (100%). We found that TBSV repRNA accumulation was inhibited by the RNAi machinery by almost three-times more efficiently in vps23Δbro1Δ yeast when compared with wt yeast expressing the full-set of the ESCRT components (Fig 1A, lanes 11–12 versus 5–6). Single expression of DCR1 had only small effect on TBSV RNA accumulation in both yeast strains (Fig 1A). The expressions of both DCR1 and AGO1 were comparable in vps23Δbro1Δ and wt yeasts (Fig 1A). Moreover, the expression of the tombusvirus p33 and p92pol replication proteins in vps23Δbro1Δ or wt yeasts was not affected by the co-expression of DCR1 and AGO1 (Fig 1A, bottom images), suggesting that enhanced susceptibility of tombusviral RNA in vps23Δbro1Δ yeast was likely due to the increased antiviral effect of the reconstituted S. castellii RNAi machinery on the viral RNAs.
Detection of vsiRNA(+) abundance revealed reduced level in vps23Δbro1Δ yeast in comparison with the wt yeast expressing DCR1 (Fig 1B). However, the reduction in vsiRNA(+) abundance is likely due to the reduced target viral RNA level in vps23Δbro1Δ yeast, which supports only ~20% TBSV repRNA level in comparison with the wt yeast (lacking the RNAi machinery) [58]. Indeed, comparison of vsiRNA(-) level (generated by DCR1 from dsRNA replication intermediate) revealed that vps23Δbro1Δ yeast generated almost as much vsiRNA(-) as the wt yeast did (Fig 1D), whereas the repRNA(-) level (representing the dsRNA replication intermediate of the repRNA) in vps23Δbro1Δ yeast was half of the level of repRNA(-) detected in wt yeast (Fig 1C).
To examine if the membranous VRCs in vps23Δbro1Δ yeast indeed provide less protection to the viral RNAs, we used cell-free extract (CFE)-based assay in the presence or absence of ribonucleases (the single-stranded ssRNA-specific RNase A and the dsRNA-specific RNase III). The TBSV dsRNA was ~2-fold more sensitive to RNases when CFE was prepared from vps23Δbro1Δ yeast in comparison with the CFE from wt yeast (Fig 1E). The TBSV ssRNAs, which are continuously released from VRCs as replication goes on, were fully degraded in both CFE assays.
In a second assay to test the level of protection provided by VRCs, we performed in vitro replicase assembly with purified recombinant viral proteins and (+)repRNA transcripts as schematically shown in Fig 1F, followed by viral RNA replication in the presence of micrococcal nuclease (MNase) to destroy the unprotected viral RNAs. The MNase was added at different time points (as shown) for 20 min and then it was inactivated by EGTA, followed by TBSV repRNA replication on the protected TBSV repRNAs up to 3 hours (Fig 1F). When CFE was prepared from wt yeast, then the VRC partially protected the viral dsRNA [produced by minus-strand synthesis on the (+)RNA template] after 40 min, whereas the protection of viral dsRNA was high after 60 min of incubation (Fig 1G, lanes 11 and 12 versus 9). In contrast, the in vitro assembled VRC based on CFE prepared from vps23Δbro1Δ yeast did not provide any detectable level of protection after 60 min of incubation. We also tested the VRCs assembled in CFEs prepared from vps23Δ or bro1Δ yeasts. These MNase protection experiments revealed poor dsRNA protection in CFEs from vps23Δ or bro1Δ yeasts at both 40 and 60 min time points (Fig 1G). However, the protection of TBSV dsRNA in CFEs from vps23Δ or bro1Δ yeasts were more significant at the 60 min time point than the lack of dsRNA protection provided in CFE prepared from vps23Δbro1Δ yeast (compare lanes 4 and 8 with 16, Fig 1G). These data suggest that due to the partially overlapping roles of Vps23p and Bro1p in supporting the formation of VRCs [58], the CFEs prepared from single deletion yeast strains provided better dsRNA protection in vitro than the CFE from double-deletion yeast strain. Altogether, the results from two separate in vitro replication assays with CFEs prepared from vps23Δbro1Δ yeast showed that the TBSV dsRNA is not well protected from ribonucleases even after the VRC assembly step, thus indicating that the dsRNA inside the VRCs in the absence of Vps23p and Bro1p is continuously exposed to the RNAi machinery or ribonucleases, likely due to incomplete VRC assembly.
Formation of complete vesicle-like structures induced by TBSV in yeast and plants also requires ESCRT-III factors [58]. In the absence of ESCRT-III factors, crescent-like membrane invaginations are formed in yeast replicating TBSV repRNA [58]. To test if Snf7p and Vps20p ESCRT-III factors are important for protecting the viral dsRNA from RNAi-based degradation, we launched TBSV replication either in snf7Δ or vps20Δ yeasts. We found that TBSV repRNA accumulation was reduced by ~2-fold more in both snf7Δ and vps20Δ yeasts when AGO1 and DCR1 were co-expressed in comparison with the wt yeast (Fig 2A). Similarly, the dsRNA replication intermediate was protected from MNase treatment by 2-to-4-fold less effectively by the in vitro assembled TBSV replicases prepared with CFEs from snf7Δ and vps20Δ yeasts than by the CFE from WT yeast (Fig 2B). The CFE-based replication assay using the double-deletion strain (snf7Δvps20Δ yeast) revealed that the level of dsRNA protection was comparable to that provided by the CFE from single deletion strain (snf7Δ yeast) against MNase treatment (Fig 2B), suggesting that the ESCRT-III factors do not have complementary roles in protecting the viral RNA from RNAi activities. Altogether, the results from the in vitro replication assay with CFE prepared from single deletion or snf7Δvps20Δ yeasts indicated that the dsRNA inside the VRCs in the absence of Snf7p or Vps20p ESCRT-III factors is continuously exposed to the RNAi machinery or ribonucleases, likely due to incomplete VRC assembly.
Formation of tombusvirus VRCs is greatly affected by sterols in vitro, in yeast and plants [45,47]. The co-opted sterols, which are enriched within the replication compartment [45], likely enhance the stability of vesicle-like structures and facilitate tighter packing of phospholipids in the membranes used by TBSV for VRC assembly. These features of sterols might also contribute to the protection of viral dsRNA replication intermediate provided by the membranous VRCs against cellular nucleases. This theory was first tested using a yeast strain deficient in ergosterol (yeast version of cholesterol) biosynthesis due to deletion of C-24 sterol reductase (erg4Δ yeast) [47,60]. Induction of the RNAi machinery in erg4Δ yeast inhibited TBSV repRNA accumulation by ~2-fold more efficiently than in wt yeast (Fig 3A and 3B).
Detection of vsiRNA(+) abundance revealed reduced level in erg4Δ yeast in comparison with the wt yeast expressing DCR1 (Fig 3C). However, the reduction in vsiRNA(+) abundance is likely due to the reduced target viral RNA level in erg4Δ yeast, which supports only ~20% TBSV repRNA level in comparison with the wt yeast (lacking the RNAi machinery) [60]. Therefore, we also measured vsiRNA(-) level (generated by DCR1 from dsRNA replication intermediate), which showed only slight reduction in erg4Δ yeast in comparison with vsiRNA(-) level in wt yeast (Fig 3E). However, the repRNA(-) level (representing the dsRNA replication intermediate of the repRNA) in erg4Δ yeast was only ~20% of the level of repRNA(-) detected in wt yeast (Fig 3D), which is comparable to the reduction in repRNA(+) level in erg4Δ versus wt yeasts [60]. Thus, the vsiRNA(-) was generated from 5-fold less dsRNA templates by DCR1 in erg4Δ yeast than in wt yeast, suggesting that DCR1 could produce vsiRNA(-) in ~3-fold higher ratio in erg4Δ yeast than in wt yeast.
Moreover, the CFE-based TBSV replication assay showed poor protection of TBSV dsRNA against the dsRNA-specific RNase III within VRCs assembled in erg4Δ yeast in comparison with wt yeast (Fig 3F). This suggests that the dsRNA within the VRCs formed in sterol-depleted cells is continuously exposed to RNAi/ribonucleases.
To further confirm the critical role of the co-opted sterols in the formation of more ribonuclease resistant VRCs, we also studied the effect of co-opted cellular proteins involved in membrane-contact site (MCS) formation [45]. The tombusvirus-induced MCSs are formed between the ER and peroxisome membranes, when the two organellar membranes are juxtaposed, with the help of co-opted ER-resident VAP proteins (Scs2 in yeast, see below) and the oxysterol-binding proteins (OSBP-related protein or ORP). The tombusvirus-induced MCSs help the enrichment of sterols within the replication compartment. First, we targeted four members of the oxysterol-binding proteins, namely Osh3, 5, 6 and 7, which are recruited by TBSV via interaction with p33 replication proteins to the site of replication. These ORP/Osh proteins are involved in the formation of MCSs and enrichment of sterols within the replication compartment [45]. Induction of AGO1 and DCR1 in osh3,5,6,7Δ yeast inhibited TBSV repRNA accumulation by ~2-fold more efficiently than in wt yeast (Fig 4A and 4B). In addition, the CFE-based in vitro replication assay revealed that the VRCs assembled in osh3,5,6,7Δ yeast could not protect efficiently the TBSV dsRNA intermediate against RNase III (Fig 4C). Second, we tested the ER-resident Scs2p VAP, which is also recruited by TBSV via interaction with p33 replication proteins to help the formation of MCSs and enrichment of sterols within the replication compartment [45,61]. Induction of AGO1 and DCR1 in scs2Δ yeast led to ~3-fold decreased TBSV repRNA accumulation than in the absence of RNAi, while the RNAi was less efficient in the parental yeast strain carrying wt copy of SCS2 (S2 Fig). The VRCs assembled in scs2Δ yeast provided negligible level of protection against RNase III in the in vitro replication assay (Fig 3F and S2C Fig). Overall, reduction of sterol biosynthesis (in erg4Δ yeast) or inhibition of virus-induced MCS formation (in osh3,5,6,7Δ or scs2Δ yeasts) that hinders the local sterol enrichment at replication sites greatly inhibited the assembly of RNase-resistant VRCs in yeast or in vitro. Thus, co-opted sterols play important roles in formation of RNAi-resistant replication compartment.
Tombusvirus replication greatly depends on phospholipid levels, especially on phosphatidylethanolamine (PE), which is highly enriched within the replication compartment [43]. PE is required for TBSV replication in an artificial vesicle- (liposome)-based in vitro assay [43] and it enhances the activation of the p92 RdRp in vitro [62]. Moreover, the PE level is increased during TBSV replication in yeast and plant cells and high PE level in yeast via modulation of phospholipid biogenesis genes also leads to enhanced TBSV replication [43,44].
To test the role of PE-rich membranes in the protection of the viral dsRNA from RNAi, we induced the RNAi machinery in cho2Δ yeast, which is partially defective in converting PE to PC, thus leading to elevated PE level and enhanced TBSV replication [43]. Since the VRCs formed in wt yeast could provide good level of protection against RNAi under our standard conditions (Figs 1–4), we applied constitutive expression of the RNAi machinery (AGO1 and DCR1) that could enhance the effectiveness of RNAi in these experiments. This is achieved by culturing yeast in media supplemented with galactose all the time (resulting in constitutive expression of both DCR1 and AGO1). Under these conditions, the VRCs assembled in wt yeast provided less protection (Fig 5A, lanes 4–6). On the contrary, the level of TBSV dsRNA protection was ~5-fold higher in cho2Δ yeast expressing AGO1 and DCR1 (Fig 5B, lanes 4–6), suggesting that increased PE levels provide more protective subcellular environment for VRC assembly.
To further test if the viral RNA is less accessible in cho2Δ yeast to the RNAi machinery than in wt yeast, we utilized an in vitro slicing assay based on purification of FLAG-tagged AGO1 from cho2Δ and wt yeasts replicating TBSV repRNA and co-expressing DCR1 and FLAG-AGO1. In these yeasts, AGO1 is expected to be loaded with vsiRNA, which then could activate AGO1 endonuclease activity specifically against TBSV RNA target in vitro (Fig 5C). By using comparable amounts of purified FLAG-AGO1 preloaded with vsiRNA, we tested the slicing activity on 32P-labeled TBSV RNA in vitro. The purified FLAG-AGO1 from wt yeast replicating the TBSV repRNA showed slicing activity against 32P-labeled TBSV gRNA (Fig 5D, lane 1), whereas the purified FLAG-AGO1 from wt yeast replicating the unrelated Flock House virus RNA1 did not have a slicing activity against 32P-labeled TBSV gRNA (Fig 5D, lane 3), thus confirming the presence of bona fide RNAi machinery in wt yeast co-expressing AGO1 and DCR1. On the contrary, the purified FLAG-AGO1 from cho2Δ yeast replicating TBSV repRNA showed negligible slicing activity against 32P-labeled TBSV gRNA (Fig 5D, lane 4), suggesting that either DCR1 or AGO1 had a limited access to the TBSV replication compartment.
To define how increased PE level provides better protection against ribonucleases, we performed CFE-based replication/protection assay using purified recombinant p33/p92 for de novo assembly of VRCs in CFEs prepared from cho2Δ versus wt yeasts (Fig 5E). This assay can test the speed of VRC assembly in vitro, based on measurement of the level of dsRNA protection provided by VRCs against MNase, which was added at various time points (Fig 5E). When CFE was prepared from wt yeast, then the in vitro assembled VRCs partially protected the viral dsRNA after 45 min, and ~60% after 60 min of incubation (Fig 5E, lanes 2 and 3 versus 1). On the other hand, CFE prepared from cho2Δ yeast showed some dsRNA protection as early as 30 min and high level of protection by 45 min and complete protection by 60 min of incubation prior to MNase treatment (Fig 5E, lanes 7–9 versus 6). Thus, we suggest that accelerated VRC assembly due to high PE level in cho2Δ yeast might decrease the time available for ribonucleases to associate with VRCs during their assembly process.
The overall phospholipid content in yeast could be increased by deleting OPI1, which is a repressor of expression of many phospholipid synthesis genes [63]. The higher level of phospholipids favors TBSV repRNA accumulation by providing easy access to membranes for VRC assembly [46,64]. TBSV repRNA accumulation reached ~2-fold higher level in opi1Δ yeast than in WT yeast expressing AGO1 and DCR1 constitutively (Fig 6B, lanes 4–6), suggesting that increased phospholipid levels facilitate VRC assembly in a more protective subcellular environment.
The CFE prepared from opi1Δ yeast supported VRC assembly that provided high level of dsRNA protection by 45 min and complete protection by 60 min of incubation prior to MNase treatment (Fig 6C and 6D, lanes 6–7 versus 10). Whereas, CFE prepared from wt yeast supported the in vitro VRC assembly at a slower pace as the viral dsRNA was only partially protected after 45 min, and at a ~60% level after 60 min of incubation (Fig 6D, lanes 1 and 2 versus 5). These data suggest that the VRC assembly was enhanced due to increased level of phospholipids in opi1Δ yeast, possibly leading to reduced time available for ribonucleases to associate with VRCs during their assembly process.
On the contrary, decreasing cellular phospholipid levels via deletion of INO2 transcription factor required for expression of many phospholipid synthesis genes [63], resulted in ~3-fold more reduction in repRNA level when AGO1 and DCR1 were expressed in ino2Δ yeast in comparison with wt yeast (supplement S3 Fig). In addition, the CFE preparation obtained from ino2Δ yeast provided poor dsRNA protection in vitro against RNase III (S3 Fig, panel C, lanes 7–8 versus 5–6). Thus, these data indicate that inhibition of phospholipid synthesis likely reduces the formation of RNAi-resistant replication compartment.
One of the remarkable features of TBSV replication is that it can switch to the ER membranes when the peroxisomes are absent in yeast due to deletion of either PEX3 or PEX19 peroxisome membrane biogenesis genes [65]. To test if the VRCs formed by usurping ER membranes are RNAi insensitive, we co-expressed AGO1 and DCR1 in pex3Δ and wt yeasts, respectively, replicating TBSV repRNA. The accumulation of repRNA was comparable in these yeasts (Fig 7A). In addition, the CFE preparation obtained from pex3Δ yeast provided comparable level of dsRNA protection against RNase III to the CFE prepared from wt yeast (Fig 7B, lanes 3 versus 7).
In the second assay to test the level of dsRNA protection provided by VRCs formed using the ER membranes, we performed in vitro replication using CFEs prepared from wt and pex3Δ yeast with pre-expressed viral proteins (Fig 7C). These CFEs were programmed with (+)repRNA transcripts, followed by viral RNA replication in the presence of MNase to destroy the unprotected viral RNAs. The MNase was added at different time points (as shown) for 15 min and then the MNase was inactivated by EGTA, followed by TBSV repRNA replication based on the protected TBSV repRNAs up to 3 hours (Fig 7C). The level of protection of the viral dsRNA provided by the VRCs was comparable after 60 min of incubation in wt and pex3Δ CFEs (Fig 7C, lanes 5 versus 10). Also, the kinetics of VRC assembly, based on the level of protection of the viral dsRNA provided by the VRCs, was comparable at four different time points in wt and pex3Δ CFEs (Fig 7C).
To further confirm these findings, we performed a third assay, in which the CFEs prepared from wt and pex3Δ yeasts (lacking viral components) were programmed with purified recombinant p33 and p92 replication proteins and (+)repRNA transcripts as depicted in Fig 7D. The level of protection of the viral dsRNA provided by the VRCs against MNase treatment was comparable in all four time points tested in wt and pex3Δ CFEs (Fig 7D). These results indicate that the subverted ER membranes in pex3Δ could provide as good protection for TBSV dsRNA against RNAi or ribonucleases as the peroxisomal membranes in wt yeast.
In this paper, we have used a reconstituted antiviral defense pathway in a model host system to further our understanding of virus-host interactions in general, and specifically the role of the membranous VRC in protection of the viral dsRNA replication intermediates against ribonucleases. This is based on the RNAi machinery of S. castellii, which only consists of the two-component DCR1 and AGO1 genes [56]. This simple RNAi machinery is known to be effective against the yeast L-A dsRNA virus and its satellite RNA in yeast [66], and TBSV (this work). Similar to the RNAi machinery of higher eukaryotes [9], the reconstituted RNAi machinery of S. castellii also requires both DCR1 and AGO1 proteins to be effective against TBSV. DCR1 produces 23 bp vsiRNA [66], as we also observed in case of TBSV. We found that this reconstituted antiviral defense pathway in surrogate host yeast is useful as an intracellular probe to further our understanding of virus-host interactions and the role of co-opted host factors in formation of membrane-bound viral replicase complexes in protection of the viral RNA against ribonucleases.
We have obtained evidence that a group of pro-viral cellular factors involved in VRC formation is essential for the generation of the protective membranous subcellular environment for TBSV in cells. Previous works with TBSV have shown that tombusviruses co-opt several host factors to build replication compartment required for replication. The replication compartment consists of mostly peroxisomal membranes and includes many virus-induced spherules, which harbor the VRCs [67]. Spherule formation requires the viral replication proteins, the viral (+)RNA and co-opted ESCRT proteins [57,58] that are involved in bending the boundary membranes of peroxisomes or ER membranes towards the organellar lumen [67]. While spherule formation is inhibited when Vps23 ESCRT-I component is missing in yeast, incomplete spherule-like structures with wide openings are formed when ESCRT-III or Vps4 AAA ATPase are deleted in yeast [57,58].
One major proposed function of spherule formation is the protection of the viral dsRNA replication intermediate formed during TBSV replication (Fig 8A) [50]. How much extent is the viral dsRNA, which is present within the membrane-bound VRC, accessible to DCR1 and the RNAi machinery? We found that in comparison with wt yeast expressing the full set of ESCRT factors, deletions of Vps23 ESCRT-I and Bro1 accessory ESCRT protein, which play partial overlapping roles in TBSV replication [58], have rendered the viral dsRNA replication intermediate highly sensitive to the RNAi machinery in yeast and to nucleases in vitro. Similarly, deletions of SNF7 or VPS20 ESCRT-III factors made TBSV replication more sensitive to AGO1 and DCR1 expression in yeast. Moreover, the lack of Snf7p or Vps20p in combination with Snf7p in CFEs used for assembly of the TBSV replicase in vitro resulted in destruction of dsRNA replication intermediate by MNase at the 60 min time point when the CFE from wt yeast has provided good protection for dsRNA (Fig 2). Altogether, these results unambiguously demonstrate that co-opted protein factors, namely the ESCRT factors, are exploited by tombusviruses not only for pro-viral functions to facilitate replication [38,57,58,68], but the ESCRT factors are also recruited to build protective subcellular environment against the RNAi machinery and other ribonucleases. Without the co-opted ESCRT factors, the tombusvirus VRCs seem to have permanent assembly deficiency, rendering the dsRNAs harbored within VRCs continuously exposed to the RNAi machinery, other host ribonucleases, and possibly cellular dsRNA sensors (Fig 8B).
The TBSV-induced spherules harboring VRCs are relatively stable membranous structures that likely synthesize viral progeny RNAs in cells for several hours, which occurs even in vitro when membranous VRCs are isolated [69]. However, the ESCRT proteins are known to act rapidly and temporarily in membrane budding events [70,71]. Therefore, TBSV likely involves additional cellular components to stabilize the vesicle-like structures in infected cells. Co-opted lipids are likely candidates for this function due to their known involvement in shaping membrane characteristics [72–74]. Accordingly, we have tested the role of sterols and phospholipids, which are critical for TBSV replication [43–47], in providing protection of viral RNA against RNAi. Indeed, deletion of ORPs or Scs2p VAP, which are known to affect MCS formation and co-opted by tombusviruses to enrich sterols at replication sites [45], resulted in reduced protection of the viral RNAs against RNAi in yeast or ribonucleases in vitro. Similarly, deletion of ERG4 involved in ergosterol synthesis (the major cholesterol-like lipid in yeast) also sensitized TBSV against RNAi. Based on these data, we suggest that sterols likely facilitate the formation of more stable and durable spherules/VRCs. Similar to the co-opted ESCRT factors, enrichment of sterols in the replication compartment, and likely within individual spherules, seems to be required to assemble tombusvirus VRCs that are not continuously exposed to the RNAi machinery (Fig 8C). Interestingly, sterols are also thought to make the plasma membrane less permeable and wider [73].
In addition to sterols, we find PE and phospholipid levels also have a critical role in the formation of RNAi-insensitive replication compartment. Accordingly, the high PE level in cho2Δ yeast made the dsRNA replication intermediate less sensitive to RNAi in yeast and ribonuclease in vitro, suggesting more rapid and efficient VRC assembly when PE is abundant in membranes. The subverted PE molecules, due to their conical molecular structures, might facilitate the formation and stability of spherule structures by introducing negative curvature into lipid bilayers [73]. This model is further supported by an in vitro slicing assay, which demonstrated that purified FLAG-AGO1 from wt yeast showed slicing activity against TBSV RNA, whereas the purified FLAG-AGO1 from cho2Δ yeast showed negligible slicing activity, suggesting that DCR1 might have a limited access to TBSV dsRNA intermediates in cho2Δ yeast with high PE level.
Based on in vitro observations, we suggest that nucleases, including DCR1 and AGO1, can likely enter the VRC during the assembly process that makes the dsRNA sensitive to RNAi. Thus, rapid VRC assembly due to high PE level in cho2Δ yeast or high level of phospholipids in opi1Δ yeast decreases the time available for ribonucleases to associate with VRCs during their assembly process, thus leading to the enhanced protection of TBSV against RNAi in cho2Δ and opi1Δ yeasts. While in wt yeast the TBSV-induced local PE enrichment within the replication compartment takes up more time than the already PE-rich membranes in cho2Δ yeast. Thus the speed or efficiency of VRC assembly could be a key factor affecting the chance for the RNAi machinery to interact with viral RNAs harbored within VRCs.
The slower pace of VRC assembly process (Fig 8D) observed in wt versus cho2Δ or opi1Δ yeast CFEs (which could facilitate the entry of DCR1 and AGO1 into the forming VRCs) might also explain that continuous expression of DCR1 and AGO1 more efficiently inhibited TBSV repRNA accumulation in wt yeast than suppression of DCR1 and AGO1 expression during pre-growth (i.e., prior to induction of viral replication) in wt yeast (compare Figs 1 and 2A versus Figs 5 and 6A and S1A). The somewhat variable level of reduction in repRNA accumulation by the RNAi machinery in wt yeast background is likely due to differences among yeast colonies in their abilities to induce the expression of AGO1 and DCR1 from the pre-repressed GAL1 promoter after the addition of galactose to the culture media.
The development of the reconstituted RNAi as a cellular probe also allowed us to demonstrate that the tombusvirus VRCs act as protective structures when assembled in the ER membranes (in the absence of peroxisomal membranes) similar to those assembled in the peroxisomal membranes. This observation indicates that TBSV could usurp the ER membrane and efficiently co-opt pro-viral host proteins and enrich lipids in this new subcellular location, giving high flexibility for this virus without sacrificing the protective quality and functionality of VRCs formed.
Overall, the reconstituted RNAi machinery of S. castellii in yeast, which supports the replication of TBSV RNA in membranous compartments, is useful intracellular probe to study the direct interaction between the RNAi machinery and the viral replicase complex, and the roles of subverted host factors in protecting the viral dsRNA replication intermediate from RNAi-based degradation. The interpretation of data in this system is simplified since S. cerevisiae does not code for an RNA-dependent RNA polymerase, which is important component of the RNAi machinery in plants and animals by producing dsRNA templates for amplifying the silencing signals [7,9]. Nevertheless, this work has demonstrated the role of co-opted cellular proteins and lipids in generation of membranous subcellular environment protected from RNAi to support TBSV replication. Other (+)RNA viruses also co-opt cellular proteins, including ESCRT factors, and subvert lipids for generation of membranous replication organelles [1,5,75–80], thus, it seems highly likely that our findings will also be applicable for wide-range of viruses.
Summary: By using the reconstituted RNAi in yeast, we have developed a simple intracellular probe to characterize membranous VRCs and viral replication compartments formed with the help of co-opted host factors in cells replicating TBSV. Moreover, we have compared the cellular data with in vitro replication results to gain deeper insights into the level of protection provided by the membranous VRCs against ribonucleases. These approaches have helped us uncover that the RNAi machinery and ribonucleases could harm viruses the most efficiently when one of two aspects of VRC assembly goes wrong. First, when the VRC assembly is permanently hindered by a missing co-opted host factor (such as the ESCRT proteins) or in the absence of local sterol enrichment in the replication compartment, then the dsRNA within the VRC is continuously exposed to RNAi or ribonucleases. Second, when the VRC assembly is slow due to a limiting host factor (such as PE accessibility), which possibly allows the components of RNAi machinery or ribonucleases to enter the VRCs prior to the completion of VRC assembly. This deficiency in excluding ribonucleases from VRCs due to slow assembly then leads to lower level of viral RNA accumulation.
Yeast (Saccharomyces cerevisiae) strain BY4741 (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0) and the YKO library were obtained from Open Biosystems (Huntsville, AL, USA). Double deletion yeast strains ΔBro1::kanMX4,ΔVps23::hphNT1 and ΔVps20::kanMX4,ΔSnf7::hphNT1 were described previously [58]. SEY6210 (MATa ura3-52 his3Δ200 lys2-801 leu2-3, trp1Δ901 suc2Δ9), and JRY6266-his3 (SEY6210 osh3Δ::LYS2 osh5Δ::LEU2 osh6Δ::LEU2 osh7Δ::kan-MX4) were provided by Dr. Christopher T. Beh (Simon Fraser University) [81].
Yeast expression plasmids pESC-Ura-Gal10-HisDcr1 and pESC-Ura-Gal1-HisAgo1-Gal10-HisDcr1 were obtained as follows: First, DCR1 and AGO1 genes from S. castellii were PCR-amplified using primers #6015 (CCAGGAATTCATGGGTCATCATCATCATCATCATATGAATAGAGAAAAAAGCGCCGATC), #6016 (CCAGACTAGTTCACAGATTGTTGCAATGCCTC) and pRS315-Dcr1 [56] as a template for DCR1 and primers #6013 (CCAGGTCGACATGGGTCATCATCATCATCATCATATGTCATCCAATTCGGAGGAG), #6014 (CCAGAAGCTTTCATATGTAGTACATGATGTCAGTG) and pRS314-Ago1 [56] as a template for AGO1. The obtained PCR product of DCR1 was digested with EcoRI/SpeI and inserted into pESC-Ura plasmid (Stratagene), which was digested with EcoRI and SpeI to generate pESC-Ura-Gal10-HisDcr1. The PCR product of AGO1 was digested with SalI/HindIII and inserted into pESC-Ura-Gal10-HisDcr1, which was digested with SalI and HindIII.
To study the effect of DCR1 and AGO1 co-expression on TBSV replication in yeast strains with ESCRT gene deletions, we co-transformed yeast strains BY4741 (parental, control), vps20Δ snf7Δ, vps20Δ and bro1Δvps23Δ with three plasmids: pGBK-HIS-Cup-Flag33-Gal-DI-72, pGAD-Cup-Flag92 and one of the following: pESC-Ura (used as a control), pESC-Ura-Gal10-HisDcr1 or pESC-Ura-Gal1-HisAgo1-Gal10-HisDcr1. Transformed yeast cells were selected on SC-ULH- plates and pre-grown in 1 ml SC-ULH- media supplemented with 2% glucose and 100 μM BCS for 24 h at 29°C. Yeast cells were then centrifuged at 2,000 rpm for 3 min, washed with SC-ULH- media supplemented with 2% galactose and resuspended in 2 ml SC-ULH- media with 2% galactose and 100 μM BCS followed by culturing at 29°C for 24 h to express AGO1 and DCR1 prior to initiation of replication. Then, yeast cells were centrifuged at 2,000 rpm for 3 min, washed with SC-ULH- media supplemented with 2% galactose and resuspended in 3 ml SC-ULH- media with 2% galactose and 50 μM CuSO4, followed by culturing yeast cells at 23°C for 16 h (Fig 2) or 24 h (Fig 1), and then processed for total RNA and protein extractions. Northern blotting and Western blotting were done as previously published [53].
Yeast strains BY4741, ino2Δ, scs2Δ, and erg4Δ were transformed with plasmids pGAD-CUP1-His-p92 (Leu2 selection), pGBK-CUP1-His-p33-ADH1-DI72 (His3 selection), and pESC-Ura-Gal1-HisAgo1-Gal10-HisDcr1 (Ura3 selection), pESC-Ura-Gal10-HisDcr1 or pESC empty. Transformed yeast were pre-grown at 23°C for 24 h in SC-ULH- media supplemented with 2% glucose and 100 μM BCS. After cell harvest and a washing step, the yeasts were grown at 23°C for 16 h (for 8 h in case of pex3Δ yeast), in SC-ULH- media supplemented with 2% galactose and 100 μM BCS. Then, yeast cells were centrifuged at 2,000 rpm for 3 min, washed with SC-ULH- media supplemented with 2% galactose and resuspended in SC-ULH- media with 2% galactose and 50 μM CuSO4, followed by culturing yeast cells at 23°C for 16 or 24 h, followed by RNA and protein analysis as described [53].
BY4741, cho2Δ and opi1Δ yeasts were transformed with plasmids pGAD-CUP1-His-p92 (Leu2 selection), pGBK-CUP1-His-p33-Gal1-DI72 (His3 selection), and pESC-Ura-Gal1-HisAgo1-Gal10-HisDcr1 (Ura3 selection), pESC-Ura-Gal10-HisDcr1 or pESC-empty. Transformed yeasts were pre-grown at 23°C for 24 h in SC-ULH- media supplemented with 2% galactose and 100 μM BCS. Then, yeast cells were centrifuged at 2,000 rpm for 3 min, washed with SC-ULH- media supplemented with 2% galactose and resuspended in SC-ULH- media with 2% galactose and 50 μM CuSO4, followed by culturing yeast cells at 23°C for 16 h, followed by RNA and protein analysis as described [53].
SEY6210 and JRY6266-his3 yeast strains were transformed with plasmids pGAD-CUP1-His-p92 (Trp1 selection), pGBK-CUP1-His-p33-Gal1-DI72 (His3 selection), and pESC-Ura-Gal1-HisAgo1-Gal10-HisDcr1 (Ura3 selection), pESC-Ura-Gal10-HisDcr1 or pESC-empty. Transformed yeasts were pre-grown at 29°C overnight in SC-UTH- media supplemented with 2% galactose and 100 μM BCS. After cell harvest and a washing step, the yeasts were grown at 29°C for 16 h in SC-UTH- media with 2% galactose and 50 μM CuSO4, followed by RNA and protein analysis as described [53].
To support in vitro TBSV replication, cell-free extracts (CFE) were prepared from untransformed BY4741, snf7Δ, vps20Δ, snf7Δvps20Δ, bro1Δ, vps23Δ and bro1Δvps23Δ yeast strains as described earlier [42], whereas CFEs were obtained from cho2Δ, opi1Δ, and BY4741 (control) as described in [43]. Reaction mixture for the in vitro TBSV replication contained 2 μl of CFE, 0.15 μg (+) DI-72 RNA, 400 ng affinity-purified MBP-p33, 400 ng affinity-purified MBP-p92pol in 20 μl total volume. The reactions were performed for 3 h at 25°C.
To support in vitro TBSV replication, CFEs were prepared from BY4741, ino2Δ, scs2Δ, erg4Δ, pex3Δ, SEY6210 and JRY6266 yeast strains, which were transformed with pESC-CUP1-p92 (Ura3 selection) and pGBK-CUP1-p33 (His3 selection). The CFE-based reaction mixtures were programmed with 0.5 μg DI-72(+) RNA transcripts as described [42,69]. The CFE-based replication mixtures were incubated at 25°C for 3 h.
Treatments of the RNA products from the CFE-based TBSV replication reactions (for Fig 1B) with RNases were done as follows: After 1 h incubation at 25°C, the TBSV replication products were treated with both ssRNA-specific RNase A (VWR) and dsRNA-specific ribonuclease RNase III (NEB). After incubation at 37°C for either 15 or 20 min, the RNA samples were extracted with phenol-chloroform and precipitated.
Treatments of in vitro assembled VRCs in CFEs with the micrococcal nuclease (Amersham) were performed as follows: MNase (final concentration of 0.1 or 0.05 U/μl in the presence of 1 mM CaCl2) was added at different time points to CFE mixtures as shown in Figures. The reaction mixtures were incubated for 15 or 20 min at room temperature and, then, 2.5 mM EGTA was added to the samples to inactivate the MNase. The CFE reactions were further incubated for a total of 3 h (counted from the start of the replication assay) at 25°C before the products were extracted with phenol-chloroform and precipitated. The obtained 32P-labeled RNA products were separated by electrophoresis in 5% semi-denaturing polyacrylamide gel containing 8 M urea with 0.5x Tris-borate-EDTA buffer [50,58].
The CFE-based TBSV replication assay #2 was performed in the presence of 0.5 U RNase III (NEB) during the entire incubation (3 hours at 25°C). Then, the RNA samples were extracted with phenol-chloroform and precipitated. The obtained 32P-labeled RNA products without heat treatment were analyzed in 5% acrylamide/ 8M Urea gels to detect dsRNA level [50].
|
10.1371/journal.pntd.0002361 | Tracing the Tiger: Population Genetics Provides Valuable Insights into the Aedes (Stegomyia) albopictus Invasion of the Australasian Region | The range of the Asian tiger mosquito Aedes albopictus is expanding globally, raising the threat of emerging and re-emerging arbovirus transmission risks including dengue and chikungunya. Its detection in Papua New Guinea's (PNG) southern Fly River coastal region in 1988 and 1992 placed it 150 km from mainland Australia. However, it was not until 12 years later that it appeared on the Torres Strait Islands. We hypothesized that the extant PNG population expanded into the Torres Straits as an indirect effect of drought-proofing the southern Fly River coastal villages in response to El Nino-driven climate variability in the region (via the rollout of rainwater tanks and water storage containers).
Examination of the mosquito's mitochondrial DNA cytochrome oxidase I (COI) sequences and 13 novel nuclear microsatellites revealed evidence of substantial intermixing between PNG's southern Fly region and Torres Strait Island populations essentially compromising any island eradication attempts due to potential of reintroduction. However, two genetically distinct populations were identified in this region comprising the historically extant PNG populations and the exotic introduced population. Both COI sequence data and microsatellites showed the introduced population to have genetic affinities to populations from Timor Leste and Jakarta in the Indonesian region.
The Ae. albopictus invasion into the Australian region was not a range expansion out of PNG as suspected, but founded by other, genetically distinct population(s), with strong genetic affinities to populations sampled from the Indonesian region. We now suspect that the introduction of Ae. albopictus into the Australian region was driven by widespread illegal fishing activity originating from the Indonesian region during this period. Human sea traffic is apparently shuttling this mosquito between islands in the Torres Strait and the southern PNG mainland and this extensive movement may well compromise Ae. albopictus eradication attempts in this region.
| The range of the Asian tiger mosquito Aedes albopictus is expanding globally, raising the threat of emerging and re-emerging arbovirus transmission risks, including chikungunya and dengue. Detection of Ae. albopictus in southern Papua New Guinea (PNG) in 1988 and 1992 placed it 150 km from mainland Australia. In 2004–05 it reached the islands in Australia's Torres Strait that separate the mainland from PNG. Suspecting a range expansion from PNG driven by human adaptation to climate variability, we employed population genetics methodologies to investigate possible origins of the introduced population, as well as population structure of previously existing populations from New Guinea. Mitochondrial cytochrome oxidase I sequences and 13 novel microsatellite markers revealed a clear genetic distinction between regional populations in PNG and the newly introduced population in the Torres Strait and Fly Region. The closest genetic relative to the introduced population was found in the Indonesian region to the west and it is now suspected that this species may have been brought into the Torres Strait by sea vessels involved in extensive illegal fishing activities during this period.
| The Asian tiger mosquito Aedes (Stegomyia) albopictus, originally described by Skuse from Calcutta, India, in 1894, is considered native to the Southeast Asian region where the larvae are often found in forest tree holes – a characteristic that assists its current global expansion via rapid adaptation to human-made container habitats [1], [2], [3]. This global expansion is also driven by human behavior, often facilitated by the transport of used tyres that contain desiccation-resistant eggs or, in some cases, by the movement of the containers themselves [4], [5]. Prior to the 1980s, Ae. albopictus had spread to several islands in the Indian Ocean, as well as to the Hawaiian Islands in the Pacific [6]. It was discovered in Albania in Europe in 1979 [7], and has also established in both North [8] and South America [9], in Africa in 1992 [10], and in southern Europe [11]. It is currently expanding into over 20 European countries [2].
Alongside this species' global expansion, its status as a vector of human pathogens is also of increasing concern. As a laboratory vector of over 25 arboviruses, its role in arbovirus transmission cycles has mostly been secondary to other incriminated vectors [12], [13], [14]. In the absence of the primary dengue vector, Aedes (Stegomyia) aegypti, Ae. albopictus has been the epidemic vector of dengue viruses in Hawaii, Macao and China [14], [15], [16], [17]. In 2005, it was implicated as the epidemic vector during a resurgence of chikungunya (CHIKV), an alpha virus clinically similar to dengue, in the Indian Ocean and in Italy [17], [18], [19]. Subsequent studies revealed that Ae. albopictus is highly susceptible to the CHIKV, with the species not only responsible for these outbreaks but also able to transmit the virus after only two days [17], [20], [21].
In the Australasian region, Ae. albopictus was first detected in 1963 in Jayapura on the West Papua Province of Indonesia (see Fig. 1A). Subsequent surveys during the early 1970s confirmed its presence in northern Papua New Guinea (PNG) near Madang [22], [23]. By 1980 it had arrived in southern PNG's Port Moresby (PNG's capital), and moved eastwards into Bougainville Province and the Solomon Islands [24], [25]. Its detection in southern PNG's Western Province southern Fly River coastal fringe in 1988 (see Fig. 1B), combined with surveys in 1992 revealing it on Daru Island in the northern Torres Strait region and in Kiunga Port over 700 km up the Fly River, established beyond doubt that the species was extant just 150 km from mainland Australia's Cape York [24], [26]. Despite there being at least 28 collections of Ae. albopictus at six Australian seaports, this species has not yet established on Australia's mainland [27].
In 2005, Ae. albopictus was detected on Masig Island in the central Torres Strait Islands and molecular identification of previously collected Ae. albopictus larvae [28] (which discriminated it from local Ae. scutellaris species), dated its arrival to 2004. Subsequent surveys in the Torres Strait revealed its presence on 10 of the 17 inhabited islands [29]. Considering the potential of both the human health and societal (nuisance) impacts of Ae. albopictus establishing on mainland Australia, the obvious question was why Ae. albopictus had only expanded into the Torres Strait Islands in 2004–05 when it was known to have been extant on Daru Island (northern Torres Strait) and in Kiunga in 1992 – 12 years earlier? A number of potential sociological and ecological factors may have contributed to the mosquitoes' proliferation and led to its dramatic expansion into the Torres Strait islands in the mid-2000s. For example, the increase in human-made water storage containers and sundry smaller discarded disposable containers may have served as potential larval habitats, leading to a population expansion.
The discovery of Ae. albopictus in the Torres Strait in the mid 2000s led to the question of whether recent adaptation to climatic variability had played a role in its expansion – as we had suggested in earlier work on other container inhabiting Aedes species in this region [30]. The 1997–98 El Nino conditions contributed to the worst drought in PNG for 100 years: traditional groundwater supplies were greatly affected, either drying up or becoming contaminated [31]. As local springs and streams dried up, it became necessary for villages to store water in large containers including 220 L (44 gal) drums and rainwater tanks – both of which provide highly productive larval sites for container-inhabiting mosquito species [32]. As part of an international aid response, AusAID funded and transported 9,000 L polypropylene rainwater tanks and 200 L water containers to the southern Fly River region villages immediately adjacent to the Torres Strait in a project completed in 2002 [33], [34]. This human adaptation to climate variability may have provided abundant productive larval sources for the population of Ae. albopictus already present in PNG, leading to its rapid population expansion and a subsequent spillover into the Torres Strait Islands. Once Ae. albopictus was established on the islands, the continuous human ocean traffic would have rapidly shuttled mosquitoes through the region. Thus our working hypothesis was that climate variability driven by the 1997–98 El Nino resulted in water storage management changes in PNG's southern Fly River coastal villages and was indirectly responsible for the invasion of a local Ae. albopictus population through the Torres Strait Islands.
In this study, we use extensive regional mosquito collections and population genetics methodologies to investigate the origins and dynamics of the introduction of Ae. albopictus into and through the Torres Strait islands, as well as the population structure of this species throughout PNG. The maternally inherited mitochondrial DNA (mtDNA) cytochrome oxidase I (COI) is used as both a population genetics marker and a proxy for female movement between islands and southern PNG villages. The rationale here is that the DNA sequence of each female is (barring mutation) identical to that of her offspring, providing insights into the dynamics and diversity of the females' contribution to each population. This proxy would ultimately be an underestimate of movement as different females of the same sequence cannot be distinguished. Additionally, we developed and ran 13 microsatellites markers that permitted the evaluation of the nuclear background of these mosquitoes.
Container-inhabiting mosquitoes were collected from throughout the Torres Strait and PNG's southern Fly River region villages by Queensland Health between 2004 and 2010 (Tables 1 and 2, and Figure 1). Populations of Ae. albopictus collected in 1992 from Daru Island (northeastern Torres Straits) and the Kiunga Port area in Western Province were provided by the Australian Defence Force. When samples were collected from private residences, permission was granted prior to entry. In most cases larvae were sampled from different containers at each location and preserved in 70% ethanol, and in some cases adults were collected. In many cases only a few individuals were collected at each location in order to reduce the chance of sampling siblings (as larvae in the same container). Larvae were initially identified as Ae. albopictus using the morphological keys of [35] and then by either real-time PCR assays [36] or by a PCR-restriction digest procedure [28] to distinguish them from endemic members of the Aedes (Stegomyia) scutellaris taxonomic group.
Specimens identified as Ae. albopictus had genomic DNA extracted using a salt extraction method [37]. For PCR amplification of a 445 bp (final edited product size) region of the mtDNA cytochrome oxidase 1 (COI), the forward primer 5′ CAY CCT GGT ATA TTT ATT GG ′3 and reverse primer 5′AAT TAA AAT ATA AAC TTC TGG were modified from [38]. The reaction was carried out in 0.2 ml well PCR plates (Astral Scientific) using 25 µl final volume and oil overlay (single drop). Final PCR mixture contained 16.6 mM [NH4]2SO4, 67 mM Tris-HCl pH 8.8 (at 25°C), 0.45% Triton X-100, 0.2 mg/ml gelatin, 1.5 mM MgCl, 0.2 mM of each dNTP, 0.4 µM of each primer. One unit of Taq polymerase (Bioline) and 2–10 ng of purified genomic DNA (1 µl of gDNA) were used per reaction. Cycling (MJ research PTC200 or a BioRad C-1000 thermal cycler) was 94°C for 3 min followed by 30 cycles of 94°C for 1 min, 40°C for 1 min, and 72°C for 1 min using minimum transition times between steps. The PCR products were visualized on a 1% agarose gel containing 0.5 µg/ml ethidium bromide and visualized at 312 nm. PCR product purification was via QIAGEN (QIAquick) PCR purification columns using manufactures recommendations.
All sequences were edited and aligned using the Geneious software [39]. To examine phylogeographic relationships, we constructed maximum parsimony haplotype networks in TCS 1.21 [40] under a 95% connection limit. Pair-wise FST values were estimated in Arlequin version 3.5 (distance method) [41] to assess levels of differentiation between the regions for the COI locus: regions were designated Torres Strait Islands (excluding Daru Island), southern Fly region, Daru Island, Kiunga, Port Moresby, Madang/Lae Region, Timor Leste and Jakarta. The significance levels of FST comparisons were assessed using permutation tests (1,023 permutations per comparison), also implemented in Arlequin. DnaSP 5 [42] was used to estimate haplotype diversity and nucleotide diversity within regions. We performed Tajima's D and Fu's Fs tests of neutrality for the COI data per population in the program Arlequin.
Candidate microsatellite markers were isolated from Roche GS FLX 454 sequencing data (1/16 plate - 25,000 reads at ∼400 bp length) generated from genomic DNA of Ae. albopictus and performed by Macrogen (Korea). To design primers for microsatellite loci, we ran the resultant data through the program msatcommander [43]. We used this program to find primers for dinucleotide, trinucleotide and tetranucleotide repeats, and allowed the program to design primers with a melting temperature in the range of 50–62°C with a GC content between 30 and 70 percent. Long polynucleotide repeats (>5 bp) within sequences to be amplified were avoided and duplicate markers (i.e. primers designed for sequence analogues) were excluded. Screening of candidate markers involved PCR amplification of a subset of samples using standard primers and visualization of products on 1% ethidium bromide stained agarose gels. Positive product primer sets were re-amplified with M13 labeled forward primers and dyes (VIC, FAM, PET and NED) and standard reverse primers. The final PCR mixture contained 1× Mytaq buffer (Bioline)(containing pre-optimized concentrations of MgCl and dNTPs), 0.4 µM of each primer, 0.5–1.0 unit of MyTaq polymerase (Bioline) and 5.0–10.0 ng of extracted genomic DNA (1 µl of extraction). The cycling involved an initial denaturation of 95°C for 3 min, then 13 cycles of 95°C for 30 s, 56°C for 40 s with a gradient decrease of 0.5°C/cycle, and 72°C for 30 s, followed by 25 cycles of 95°C for 30 s, 50°C for 40 s and 72°C for 30 s, and a final 72°C for 5 min using minimum transition times. M13 labeled products for 13 microsatellite markers (see Table 3 for details) that generated clean peaks and that amplified consistently were purified using ExoSap (Antarctic phosphatase and Exonuclease I-New England Biolab) and were sent to Macrogen (Macrogen, Geumchun-gu, Seoul, Korea) for genotyping. We attempted to genotype 199 individuals sampled from the Torres Strait Islands, New Guinea, Timor Leste and Jakarta (see Table 2 for sampling information).
Alleles for each marker were scored manually in the program GeneMarker [44]. We checked for the possible presence of null alleles for each marker at a population level (based on the regions: Torres Strait, Fly Region, Daru, Kiunga, Madang, Port Moresby, Timor Leste and Jakarta) using the program MICRO-CHECKER [45]. Using these same population definitions, we checked for HWE, as well as calculating observed (Ho) and expected (He) heterozygosity in the program GenAlEx, v6 [46] and the program GenoDive [47] was used to calculated Fis for each population (Table 4 contains details of Null Alleles, HWE, etc).
The Bayesian program STRUCTURE, v. 2.3.2 [48], [49], was used to infer the most likely number of genetically distinct groups (K) in the region sampled, based on the microsatellite data. STRUCTURE was run using the admixture model, and using sampling locations as priors. Including information on sampling locations in STRUCTURE analyses has been shown to be useful for detecting subtle genetic structure, without detecting structure that is not present [48]. Due to the potential presence of null alleles at a number of markers in some populations, we used a dominant marker model in STRUCTURE (as recommended in the user manual). STRUCTURE was run for five iterations of K = 2 to K = 8, for a total of 1 million generations per iteration with a burn-in of 200 000 iterations. STRUCTUREHARVESTER, a program that implements the Evanno et al. Delta K method [50], [51], was then used to infer the most likely value of K and CLUMPP v. 1.1.2 [52] was used to average the results of the replicates for K = 5 (the most likely value based on the Delta K method). We used the Greedy algorithm in CLUMPP with 1000 repeats. The output from CLUMPP was run through DISTRUCT [53], which provides more flexibility in generating figures than STRUCTURE. Additionally, the program GENETIX v. 4.05 [54] was used to perform a Factorial Correspondence Analysis (FCA). FSTAT v2.9.3 [55] was used to test for linkage disequilibrium between loci, and finally Arlequin v.3.5 [41] was used to estimate pair-wise FST values between the eight populations defined above.
A total of 16 mtDNA COI haplotypes were identified from 377 individuals (haplotype diversity Hd = 0.769) collected throughout the southern Fly River villages, the Torres Strait islands, north and south PNG, Timor Leste and Jakarta – 10 of these haplotypes were present in the Torres Strait Islands and southern Fly River region (see Table 1 and Fig. 1 for details on collections and mtDNA haplotype occurrence). All DNA sequences are available through GenBank (KC572145 - KC572496, KF042861-KF042885) and all tests of neutrality (Tajima's D and Fu's Fs) were non-significant. Haplotype diversity varied for each region with the Torres Strait Islands (122 individuals, 10 haplotypes, Hd = 0.801) and southern Fly River region (60 individuals, 6 haplotypes, Hd = 0.649) having substantially higher haplotype diversities than the PNG populations from Daru Island (35 individuals, 6 haplotypes, Hd = 316), Kiunga (38 individuals, 3 haplotypes, Hd = 0.198), Port Moresby (58 individuals, 5 haplotypes, Hd = 0.462) and Madang/Lae Region (39 individuals, 2 haplotypes, Hd = 0.391) as well as those from Timor Leste (17 individuals, 2 haplotypes, Hd = 0.118) and Jakarta (8 individuals, 2 haplotypes, Hd = 0.250).
The COI haplotype network (Figure 2) suggests that there is some mitochondrial genetic structure between regions, with one of the most common haplotypes (H1) sampled found predominantly in the populations extant in PNG (the Madang/Lae region, Port Moresby, Kiunga and Daru Island – with Daru collections from both 1988 and 2008), and only being sampled once (in one individual) in the Torres Strait Islands. The other common PNG haplotype (H6) was sampled at relatively high frequency in the Torres Strait. Two other haplotypes (H11 and H12) that were sampled relatively commonly in both the Torres Strait and in the Fly Region of PNG were also sampled in Daru. Haplotype 11 was also the most predominant haplotype sampled in both of the Indonesian populations (Jakarta and Timor Leste), with Timor Leste also sharing H12 and Jakarta possessing one other private haplotype (H16). This suggests that there is a close affinity between Indonesian populations and those found in the Fly Region of PNG as well as those in the Torres Strait. Six private haplotypes were sampled in the Fly Region/Torres Strait (H7, H9, H10, H13, H14, H15), three of which were only found singly in Torres Strait (H7, H9, H13) and there were 5 private haplotypes sampled in PNG; H2, H4 and H5 found only in Daru; with H3 and H8 found only in Port Moresby.
Mitochondrial COI pair-wise FST relationships and significance comparisons supported the presence of structure between populations (see Table 5). Again the PNG Southern Fly Region and Torres Strait populations appeared highly distinct from the PNG populations with high and significant FST values between the populations (roughly between 0.4 to 0.5). The FST value between Torres Strait and Fly Region populations is significant but small (0.043), and most comparisons between PNG populations are non-significant (all FST<0.1). Indonesian populations are not significantly differentiated from each other but have significant FST values in all other comparisons, with the Torres Strait populations being the most closely related to them, followed by the Fly Region and then by PNG populations.
A total of 199 individuals were genotyped for the 13 microsatellites (see Table 2 for mosquito sampling). Putative null alleles were found at some loci in some populations and tests for Hardy Weinberg equilibrium revealed that some loci did not meet the expectations of this model (see Table 4 for detailed information), but no evidence of linkage disequilibrium between loci was found. The overall number of alleles per locus ranged from 8 to 17 (Table 3). The inbreeding coefficients (FIS) of the majority of loci were positive, and observed heterozygosity was less than expected heterozygosity in most cases, indicating an excess of homozygote genotypes at most loci (see Table 4). Although FST values are generally smaller for the microsatellite data than for the mitochondrial data (Tables 5 & 6), all microsatellite based pair-wise FST values between populations were significant, with the exception of the Torres Strait – Fly Region comparison (FST = 0.00421, Table 6). Low FST values were found between PNG populations, as well as between Torres Strait/Fly Region populations and those from Indonesia, providing further evidence of close affinities between these populations.
The most likely number of genetic clusters (K) inferred by STRUCTURE HARVESTER was K = 5 [50], [51]. The bar plot generated in STRUCTURE (Fig. 3) suggests five populations (Figure 3) with three historically extant populations within PNG that may have experienced various levels of admixture, and one distinct population encompassing the Torres Strait Islands and the southern Fly River region (purple). An additional population was found in Indonesia (pink), and bar-plots suggest some similarity of these populations to some individuals in the Torres Strait and Fly Region. The populations from Daru Island (collected in 1992 and 2008), which sits geographically adjacent to the southern Fly River coastal region, are clearly differentiated from the introduced populations, with all individuals from these regions being assigned with high probability to a single population (green). Samples from Kiunga are assigned with high probability to a distinct cluster (red) to which individuals from Port Moresby are also partially assigned, although these (Port Moresby) individuals are also assigned to another cluster (yellow) with higher probability. Individuals from Madang in northern PNG are assigned with highest probability to the green cluster (mostly found in Daru) and with a lower probability to the yellow cluster (mostly found in Port Moresby).
The factorial correspondence analysis performed in GENETIX (Figure 4) supports the results of the STRUCTURE analysis. It clearly shows the close relationship between individuals from Daru Island and Madang, as well as between individuals from Port Moresby and Kiunga. Additionally, the introduced populations from the Torres Strait Islands and the southern Fly region are closely associated. The population with the greatest genetic affinities to the introduced population based on the FCA is Timor Leste, suggesting that the source of the introduction was more likely from the Indonesian region (where Ae. albopictus is common) than from the extant PNG populations, as had been previously hypothesized. The Jakarta population is relatively isolated on the FCA plot.
Detection of Ae. albopictus in southern Papua New Guinea (PNG) in 1988 and 1992 placed it only 150 km across the Torres Strait from mainland Australia. In 2004–05 it appeared on the Torres Strait Islands and we initially suspected a range expansion from PNG potentially driven by human adaptation to climate variability. The AusAID funded drought-proofing expansion of rainwater tanks and 200 L water containers into the southern Fly River region villages immediately adjacent to the Torres Strait was completed in 2002 as a response to climate variability in the region [33], [34]. Thus it was reasonable to hypothesize that the population historically extant in PNG had undergone a range expansion, initially into the southern Fly River region and subsequently into the Torres Strait Islands. The introduction of the species into the Torres Strait Islands was traced back to 2004 [29], at a date that appeared to correlate with the change in water management which had occurred a few years earlier.
Initially the mtCOI suggested that two genetically distinct populations were present in this region, providing the first piece of evidence that the invading population may not have originated from the population previously extant in PNG. Shared haplotypes between the southern Fly Region, the Torres Strait Islands and Indonesian populations provided the first clue as to where the invading population may have originated. Despite haplotype diversity being biased by the larval sampling method (which favors the collection of siblings of the same haplotype), the Torres Strait Island populations revealed four more haplotypes (a total of 10, Hd = 0.801) than the southern Fly River (6 haplotypes, Hd = 0.649). This difference in haplotype diversity may suggest that the initial introduction into the region started in the Torres Strait Islands from the Indonesian region, however more COI sequencing from the southern Fly region may be needed to clarify whether or not this is the case. Interestingly however, genetic diversity appears higher in the Torres Strait and Fly Region than in the Indonesian populations, possibly suggesting multiple introductions of closely related populations from different parts of Indonesia, or that the founding population was more genetically diverse than those sampled from the Indonesian region.
The mitochondrial DNA was informative at another level with the discovery that multiple Ae. albopictus mtDNA haplotypes (representing different females contributing to the population) were moving between islands. This suggests that attempts to eradicate the species from individual islands would likely be unsuccessful given the high potential for re-introductions. Indeed, this information has assisted Queensland Health – the regional state health authority – in its decision to move from the island eradication program implemented in 2006 to a cordon sanitaire in 2008, whereby surveillance and control was focused on the inner Torres Strait islands of Waiben, Muralug and Ngurupai (Thursday, Prince of Wales and Horn islands) adjacent to mainland Australia. In particular, Muralug and Ngurupai act as the major regional transport hubs and are thus the most likely staging point for the species' introduction onto the Australian mainland. This cordon sanitaire was breached in 2009 and Ae. albopictus now exists on Waiben and Ngurupia, less than 30 km from Australia's Cape York Peninsula. In 2010, Aedes albopictus larvae were collected from New Marpoon on mainland Australia's Cape York although no other individuals have been collected in this area since.
The 13 microsatellite loci reaffirmed the findings of the mitochondrial data that the introduced population was genetically distinct to the populations already present in PNG. As microsatellites evolve more rapidly than mitochondrial sequence data, they were more informative, resolving five genetically distinct populations in total with three historically extant populations in PNG that have experienced various levels of admixture and one distinct population encompassing the Torres Strait Islands and the southern Fly River region – the introduced population. Samples from Madang in northern PNG, as well as from Port Moresby on the southern Papuan Peninsula, appear to be admixtures, with individuals from Port Moresby being more similar to samples from Kiunga, and individuals from Madang more similar to the Daru Island population (Figures 3 and 4). Although Daru Island is proximal to the introduced population, material collected there on two separate occasions (1992 and 2008) was assigned with high probability to a separate population with close genetic affinities to the Madang material. Interestingly, the material collected from the Indonesian region (Timor Leste and Jakarta) forms a distinct population with apparent genetic affinities to the introduced population in the Torres Strait and southern Fly Region. Timor Leste revealed the highest genetic affinities in the FCA analysis and there are records of Ae. albopictus being present in Timor Leste dating to the 1920s [23]. Thus, our combined data strongly suggests that the introduction of Ae. albopictus into the Torres Strait and southern Fly River region came from the Indonesian region.
Since it now appears highly unlikely that the introduced population originated from PNG but rather came from the Indonesian region (west of the Torres Strait), it is conceivable that the introduction was driven by foreign fishing vessels that travelled from the Indonesian region, harboring Ae. albopictus specimens which then infested the islands of the Torres Strait and/or the southern Fly region. Indonesian/Macassan visitations to the coastline of northern Australia have a long history predating European settlement, and Indonesian fishermen have been known to illegally enter Australian waters more recently to fish for shark fin (Walker, J. Pers. Com.). This activity is reported to have peaked between 2005–07, during which time several hundred landings occurred where fuel, water, shark fin, fishing nets and lines were often cached. Many of these landings were associated with well-established camps that received multiple visits. Uninhabited islands in the Torres Strait were one focus for these activities and evidence gathered from apprehended vessels indicates that most shark boats (Type III - highly mobile, motorized vessels) carried large open-topped water drums of which a significant proportion harboured Ae. aegypti and Ae. albopictus larvae. There have been documented collections of various Aedes (Stegomyia) species, including Ae. albopictus, from illegal fishing vessels that were intercepted at the port of Darwin in the Northern Territory that clearly indicate that the mosquitoes' survival in smaller vessels is possible [56], [57].
With regard to the expansion and movement of Ae. albopictus in this region, populations on Daru Island (which adjoins the Torres Strait) appear to have been unaffected by this exotic invasion up until 2008 and follow-up collections are now warranted to determine if the introduced population has established there since. Despite Daru Island sitting geographically adjacent to the southern Fly River coastal region, its Ae. albopictus population seems to be clearly isolated from the introduced population. The occurrence of two genetically distinct populations adjacent to each other in the northern Torres Strait/PNG region can be best explained by their different jurisdictions. While the Torres Strait operates as part of Australia, Daru Island is politically part of PNG and its function as the international customs clearance station into Western Province means that it encounters different incoming and outgoing transport movements.
The movement of Ae. albopictus appears to be extensive in the Australasian region, particularly where the Torres Strait Islands' junction separates New Guinea and mainland Australia. With the maximum reported flight range of Ae. albopictus being 1 km [58], movement between the Torres Strait's islands has most likely taken place via human-mediated transport. Australia has an oceanic border with PNG and the Torres Straits, but unlike most other international countries, it has no clearly marked frontier with border policing or customs controls. Thus relatively free movement occurs between PNG and the Torres Straits with approximately 5,000 international shipping movements per year [59], and countless domestic movements that would likely shuttle Ae. albopictus between the islands and back and forth from the southern Fly region's villages. This movement is sanctioned under The Torres Strait Treaty (Miscellaneous Amendments) Act 1984 that allows for cross-border movement for trade, fishing, and family gatherings or for seeking medical attention without the need for customs protocols [60] –all of which may compromise quarantine in the region. The primary mode of transport between PNG villages and the Torres Strait Islands is via open-topped, outboard-motor-powered boats. In addition to sheltered harborage sites for adult mosquitoes, these vessels contain fresh water – both within drums for human consumption, and where rainwater has collected – which in turn provides oviposition sites and larval habitats for container-inhabiting species such as Ae. albopictus.
Given its current location, its mobility and its phenotypic fondness for containers, Ae. albopictus is more than likely to arrive in a town or city on the Australian mainland via human transport. Due to the intrinsic ecological plasticity in both larval habitat (both natural and human-made) and host feeding patterns [3], it will likely be able to move between urban and sylvan habitat, and control will be extremely challenging once it enters the latter. Its effect on native virus vector systems in Australia represents an unknown risk both to humans and to native and domestic animals. However, its cool climate tolerant biology and plasticity will certainly present new risks for dengue and chikungunya transmission in summer throughout most of the Australian region [27], [30].
Aedes albopictus's particular biology permits its container inhabiting ecological niche to once again facilitate its global expansion. It is important to consider it's potential to rapidly exploit the outcomes of any socio-economic or policy-driven interventions, such as the recent and dramatic expansion of domestic rainwater tanks throughout Australian urban regions as a drought-proofing adaptation to observed and forecasted climate change [30]. Although this adaptation did not ultimately explain the range expansion of the species on Australia's northern doorstep, it will nonetheless provide a valuable niche in the landscape, which may augment this vector's existence across Australia's urban regions. In considering its eventual arrival, the public health risks associated with arboviruses meet the possibility of substantial daytime nuisance biting that will also negatively impact Australia's urban alfresco lifestyle.
|
10.1371/journal.pntd.0003678 | Frequent House Invasion of Trypanosoma cruzi-Infected Triatomines in a Suburban Area of Brazil | The demographic transition of populations from rural areas to large urban centers often results in a disordered occupation of forest remnants and increased economic pressure to develop high-income buildings in these areas. Ecological and socioeconomic factors associated with these urban transitions create conditions for the potential transmission of infectious diseases, which was demonstrated for Chagas disease.
We analyzed 930 triatomines, mainly Triatoma tibiamaculata, collected in artificial and sylvatic environments (forests near houses) of a suburban area of the city of Salvador, Bahia State, Brazil between 2007 and 2011. Most triatomines were captured at peridomiciles. Adult bugs predominated in all studied environments, and nymphs were scarce inside houses. Molecular analyses of a randomly selected sub-sample (n=212) of triatomines showed Trypanosoma cruzi infection rates of 65%, 50% and 56% in intradomestic, peridomestic and sylvatic environments, respectively. We detected the T. cruzi lineages I and II and mixed infections. We also showed that T. tibiamaculata fed on blood from birds (50%), marsupials (38%), ruminants (7%) and rodents (5%). The probability of T. cruzi infection was higher in triatomines that fed on marsupial blood (odds ratio (OR) = 1.95, 95% confidence interval (CI) = 1.22-3.11). Moreover, we observed a protective effect against infection in bugs that fed on bird blood (OR = 0.43, 95% CI = 0.30-0.73).
The frequent invasion of houses by infected triatomines indicates a potential risk of T. cruzi transmission to inhabitants in this area. Our results reinforce that continuous epidemiological surveillance should be performed in areas where domestic transmission is controlled but enzootic transmission persists.
| Most triatomines are found in forested environments. However, environmental imbalances can influence the decline of sylvatic vertebrate populations and promote food shortages for triatomines. As a survival strategy, these bugs tend to disperse to houses near forest remnants. We studied the spatial distribution, natural infections by Trypanosoma cruzi and feeding patterns of triatomines in a suburban area of Salvador, Bahia State, Brazil between 2007 and 2011. Triatoma tibiamaculata was collected in sylvatic, peridomestic and domestic habitats, and a high T. cruzi infection rate was established for triatomines that fed on blood from birds, marsupials, ruminants and rodents. The frequent invasion of dwellings by infected triatomines can present a risk for oral and vectorial transmission of T. cruzi to humans. We recommend that residents of triatomine-prone areas perform the following procedures: (i) avoid leaving artificial lights on overnight; (ii) periodically clean palm trees near their homes; and (iii) install net screens on windows. Moreover, education campaigns should promote community awareness of the risks of T. cruzi infection and provide alternatives to prevent human contact with vectors. Finally, we highlight the importance of continuous epidemiological surveillance in areas where domestic transmission is controlled but enzootic transmission persists.
| The demographic transition of populations from rural areas to large urban centers often results in the development of slums, disordered occupation of forest remnants, parks, and protected areas, and increased economic pressure to develop high-income buildings in these same areas. Ecological and socioeconomic factors associated with these urban transitions create conditions for the potential transmission of infectious diseases, which has been demonstrated for Chagas disease [1,2]. Vector-borne Trypanosoma cruzi transmission occurs in urban areas of South [2–7], Central [8] and North America [9]. Moreover, several urban outbreaks of acute Chagas disease related to the ingestion of contaminated food were recently described [10–14].
Deforestation, forest fragmentation and human occupation can cause the local extinction of mammals in Atlantic forest remnants [15], thus diminishing food sources for hematophagous insects, such as triatomine bugs, which are vectors of Chagas disease. Furthermore, anthropogenic landscape disturbances can increase triatomine abundance and their rates of infection with T. cruzi, which indicates that forest remnants may be sources for vector populations within disturbed areas [16,17]. These processes, coupled with socioeconomic factors and triatomine attraction to artificial lights, may influence the invasion of Chagas disease vectors in environments near human populations [18–20].
Recently, the urban growth of Salvador City, Bahia, Brazil [21] has caused a strong and increasing pressure on the remaining forested areas of the town and promoted the conditions required for the invasion of human dwellings by Chagas disease vectors, which have previously been reported in suburban and slums areas in Salvador City [18]. This invasion is concerning because the occurrence of infected triatomines inside homes can contribute to oral outbreaks and vectorial transmission of Chagas disease [22].
The first notification of triatomine bugs in Bahia was provided by Brumpt & Silva [23]. Chagas disease became a serious public health problem in the city of Salvador in the 1950’s, with the registration of numerous cases of autochthonous chagasic cardiopathy, which was associated with vectorial transmission by Panstrongylus megistus and Triatoma rubrofasciata. Triatomine populations were subsequently controlled with large-scale insecticide campaigns [24]. However, triatomines continue to invade homes in Salvador, which underscores the need to analyze the distribution of T. cruzi infection among vectors and associated risk factors for Chagas disease re-emergence in this city. This paper studied the spatial distribution, natural infection by T. cruzi and blood meal sources of triatomines captured between 2007 and 2011 in a suburban area of Salvador City.
This research was conducted in Salvador City from July 2007 to December 2011. The city of Salvador is the capital of Bahia state, which is in the northeast of Brazil (Fig 1). This city was founded within the Atlantic Forest, which is of the richest and most endangered ecoregions in Brazil. Salvador City has an annual average temperature of approximately 25°C with small variation. The climate is tropical, hot and humid with no pronounced dry season and classified as Af according to the Koppen-Geiger classification. The suburban area where triatomines were collected was developed recently near Atlantic Forest remnants. The houses are located in an upscale area where mansions are interspersed with vegetation. Most peridomiciles have swimming pools, gardens, kennels and a barbecue area. Chicken coops, pigsties and corrals were not observed.
Part of the examined triatomines were captured by community households or health workers from the Zoonosis Control Center (ZCC) of Salvador following the recommendations of the Brazilian Ministry of Health [25,26]. Briefly, manual triatomine searches in the houses notified by ZCC were conducted by trained individuals equipped with gloves, flashlights and tweezers. Triatomines were categorized as intradomestic, peridomestic or sylvatic based on the environment where they were captured. Peridomestic triatomines were captured in the area surrounding homes, usually in a fenced compound. Sylvatic triatomines were captured by our field group and ZCC workers by manual searches in Attalea spp. palm trees [27]. We registered information for each sample using a standardized entomological survey form. The geographical coordinates of the positive houses were determined by using a handheld GPS. The triatomines were separated by sex and nymphal stage and morphologically identified using taxonomic keys [28]. We performed abdominal dissection and isolation of intestinal contents, which were eluted in phosphate buffered saline (PBS) and subsequently frozen at -70°C until analysis.
DNA extraction was performed using the DNAzol commercial kit (Invitrogen, California, USA). Trypanosoma cruzi identification and lineage typing followed previously described PCR protocols [29]. We used specific primers from previous publications [30,31] and new primers developed in this study to evaluate triatomine blood meals (Table 1). The list of animals investigated as probable food sources for triatomines are shown in Table 1. The following PCR conditions were employed: initial denaturation at 95°C for 5 min; 35 cycles of 95°C for 1 min, 57–65°C (see Table 1) for 1 min, and 72°C for 1 min; and final elongation at 72°C for 10 min.
The PCR products of blood meal samples were analyzed using electrophoresis on 1.5% agarose gels, and T. cruzi PCR products were separated in 3% agarose gels. The gels were stained with ethidium bromide, visualized under ultraviolet light and photographed in a photo documenter MultiDoc-It (UVP Imaging Systems, USA). A reference size standard (staircase) of 100-bp DNA (Invitrogen, California, USA) was used.
Triatomine distribution in the studied area was analyzed with a 100-m bandwidth Kernel density estimator and the spatial analyst tool from ArcGis 10.1 (Esri, California, USA). The graphical representation was performed using orthophotos at a scale of 1:2000. Moreover, we estimated the association between triatomine blood sources and T. cruzi infection using odds ratios (ORs) and 95% confidence intervals (CIs). Descriptive analyses, ORs, 95% CIs and linear regression analyses were performed using the statistical package Epi Info 2000 (CDC, Georgia, USA).
A total of 167 households (37%) reported the presence of triatomines in the study area, mainly in the Patamares neighborhood, between 2007 and 2011. Increased triatomine notifications coincided with higher temperatures and were frequent between October and March. We obtained 930 triatomines, and most were identified as Triatoma tibiamaculata (99%) in the Patamares neighborhood (95%), where a wide distribution of triatomines near forest remnants was observed (Fig 1). The six adult specimens of Panstrongylus geniculatus were captured at peridomiciles. Therefore, we have only showed the analyses for T. tibiamaculata.
Most of the triatomines (n = 471, 51%) were captured in peridomiciles, such as balconies, outdoor kitchens and service areas. Adult bugs predominated in all studied environments, and nymphs were scarce inside houses. However, all triatomine stages were observed in peridomestic and sylvatic environments (Fig 2).
Molecular analyses of a randomly selected sub-sample (n = 212) of triatomines showed T. cruzi infection rates of 65%, 50% and 56% in intradomestic, peridomestic and sylvatic environments, respectively. We detected the T. cruzi lineages I and II and mixed infections (Fig 3). Four blood types were detected in T. tibiamaculata, and bird and marsupial DNA were the most frequent (Fig 4). Mixed blood meals were detected in 12% of the analyzed samples (bird+marsupial, bird+ruminant, rodent+ruminant). A higher frequency of T. cruzi infection was observed in triatomine samples with marsupial DNA (Fig 5), and the probability of T. cruzi infection was higher in triatomines that fed on marsupial blood (OR = 1.95, 95% CI = 1.22–3.11). Moreover, a protective effect against infection was observed in triatomines that fed on bird blood (OR = 0.43, 95% CI = 0.30–0.73).
In the present study we analyzed the spatial distribution, natural infections by Trypanosoma cruzi and feeding patterns of triatomines in a suburban area of Salvador, Bahia State, Brazil. Triatoma tibiamaculata was collected in sylvatic, peridomestic and domestic habitats, and a high T. cruzi infection rate was established for triatomines that fed on blood from birds, marsupials, ruminants and rodents. The frequent occurrence of infected T. tibiamaculata specimens near and inside residences exposes the local population to the potential risk of oral and vector-borne transmission of Chagas disease in suburban areas of Salvador City. The high infection rates and observed association between T. tibiamaculata and marsupials indicate an intense enzootic transmission of T. cruzi in Atlantic Forest remnants near houses.
Despite the presence of triatomines in other locations around Salvador City, most of the collection sites were in the Patamares neighborhood. This is a moderate disturbed area, where the houses are interspersed with Atlantic Forest remnants. In this scenario human populations co-occur with triatomines which inhabit sylvatic ecotopes and are benefited by the presence of synanthropic mammals. Moreover, in urban areas with high anthropogenic disturbance (where most houses of Salvador City are located) the number of invasive triatomines should be low due to habitat loss and consequent local extinction of triatomines and sylvatic vertebrate hosts [15,16,17].
High T. cruzi infection rates (>50%) of triatomines in urban areas were also observed in Chile [31] and Bolivia [6]. The high prevalence of T. cruzi in T. tibiamaculata collected from houses in the present study may be caused by their association with marsupials, mainly Didelphis spp. These mammals show a synanthropic behavior that facilitates the connection between wild and peridomestic cycles of T. cruzi [32,33].
Molecular typing demonstrated the presence of T. cruzi lineages I and II circulating among T. tibiamaculata specimens. The same lineages were detected in T. tibiamaculata populations from a Chagas disease outbreak that occurred through oral transmission in southern Brazil [34], and the infection of T. tibiamaculata with T. cruzi and its association with marsupials has been previously demonstrated [35, 36].
Triatoma tibiamaculata is a species that is primarily associated with palm trees and marsupial and rodent nests in bromeliads [27,35,37]; however, specimens are also found in domestic environments in certain areas of Brazil [38]. The distribution of houses near forest remnants may facilitate contact with infected T. tibiamaculata. Gottdenker et al. [16,17] observed that anthropogenic disturbances can increase triatomine abundance and infection with T. cruzi in forest remnants. Consequently, triatomine bugs invade dwellings in search of blood meals or because of their attraction to artificial light sources, which exposes the human population to Chagas disease transmission [18,20].
Birds and marsupials were the most frequent blood meals of T. tibiamaculata in the studied area. Notably, 10% of the samples did not present an amplification in any of the tested animals, which indicates that primers should be developed with new targets such as bats, which may be important reservoirs for T. cruzi. Moreover, the specific identification of blood sources using quantitative PCR [39], and sequencing [40] could improve the detection method accuracy and the detection of mixed blood meals.
Most of triatomine data obtained here were from a convenience sample (positive houses reported by the local health service). The bias associated with a non-standardized sampling method is a limitation of the present study and could underestimate the actual occurrence of triatomines. Repeated surveys [41] and improvement of vector surveillance with community participation [42] may increase the detection sensitivity in the house environment and also allow the study of colonization tendencies, considering the finding of nymphs of T. tibiamaculata inside houses of Patamares neighborhood. Moreover, the spatial distribution of T. tibiamaculata in forest remnants and houses can be improved from a standard sample in different urban scenarios allowing testing the hypothesis that the occurrence of infected triatomines is dependent on the house distance to forest remnants.
It is important to note that we did not quantify the risk for Chagas disease transmission for the inhabitants of this area and no acute cases of Chagas disease was reported in Salvador city during the last years. In addition, evidence of human blood in the gut of the triatomines was not presented, a serological survey in the human population was not performed, and infection in potential mammal reservoirs was not evaluated. These studies may reveal the actual risk of exposure to T. cruzi for the inhabitants of this area.
In conclusion, the results show the frequent invasion of infected triatomines in houses, which suggests the potential risk of T. cruzi transmission to inhabitants in this area. Our results reinforce the need for continuous epidemiological surveillance in areas where domestic transmission is controlled but enzootic transmission persists. We recommend that residents (i) avoid leaving artificial lights on overnight, (ii) perform periodic cleaning of palm trees near their homes, and (iii) install net screens on windows. Moreover, education campaigns should promote community awareness of the risk of T. cruzi infection and provide alternatives to prevent human contact with vectors.
|
10.1371/journal.pntd.0001912 | Transcriptional Profiling of the Bladder in Urogenital Schistosomiasis Reveals Pathways of Inflammatory Fibrosis and Urothelial Compromise | Urogenital schistosomiasis, chronic infection by Schistosoma haematobium, affects 112 million people worldwide. S. haematobium worm oviposition in the bladder wall leads to granulomatous inflammation, fibrosis, and egg expulsion into the urine. Despite the global impact of urogenital schistosomiasis, basic understanding of the associated pathologic mechanisms has been incomplete due to the lack of suitable animal models. We leveraged our recently developed mouse model of urogenital schistosomiasis to perform the first-ever profiling of the early molecular events that occur in the bladder in response to the introduction of S. haematobium eggs. Microarray analysis of bladders revealed rapid, differential transcription of large numbers of genes, peaking three weeks post-egg administration. Many differentially transcribed genes were related to the canonical Type 2 anti-schistosomal immune response, as reflected by the development of egg-based bladder granulomata. Numerous collagen and metalloproteinase genes were differentially transcribed over time, revealing complex remodeling and fibrosis of the bladder that was confirmed by Masson's Trichrome staining. Multiple genes implicated in carcinogenesis pathways, including vascular endothelial growth factor-, oncogene-, and mammary tumor-related genes, were differentially transcribed in egg-injected bladders. Surprisingly, junctional adhesion molecule, claudin and uroplakin genes, key components for maintaining the urothelial barrier, were globally suppressed after bladder exposure to eggs. This occurred in the setting of urothelial hyperplasia and egg shedding in urine. Thus, S. haematobium egg expulsion is associated with intricate modulation of the urothelial barrier on the cellular and molecular level. Taken together, our findings have important implications for understanding host-parasite interactions and carcinogenesis in urogenital schistosomiasis, and may provide clues for novel therapeutic strategies.
| Parasitic Schistosoma haematobium worms cause urogenital schistosomiasis in 112 million people worldwide. These worms lay eggs in the bladder wall, resulting in inflammation, fibrosis (internal scarring), bladder cancer, and passage of eggs into the urine. Indeed, the International Agency for Research on Cancer within the World Health Organization has classified S. haematobium as a “Class I” agent (“Carcinogenic to humans”). Moreover, S. haematobium-induced fibrosis and resulting obstructive kidney failure leads to 150,000 deaths annually. As a result, S. haematobium infection is one of the most important causes of worm-related death globally. In spite of this, research on this parasite is sparse due to a lack of suitable animal models. We have used our recently developed mouse model of urogenital schistosomiasis to understand the global bladder gene response to this infection. Large numbers of genes featured differential transcription after experimental infection, including specific immune response-, fibrosis-, cancer-, and bladder function-related genes. The relevance of these gene-based findings was verified through microscopic examination of egg-exposed bladders. Our data will improve our comprehension of urogenital schistosomiasis, and may help identify new targets for diagnosis and treatment of this disease, and possibly bladder cancer and bladder-based inflammatory disorders as well.
| Schistosomiasis, chronic infection with parasitic Schistosoma worms, affects at least 200 million people worldwide, and may rival malaria in terms of socioeconomic impact in endemic regions [1]. The two main forms of schistosomiasis are hepatoenteric and urogenital, caused primarily by Schistosoma mansoni and Schistosoma haematobium, respectively. In both forms of schistosomiasis, the tropism of adult worms for particular vascular beds (mesenteric or pelvic) determines where oviposition occurs. Deposited eggs trigger florid tissue inflammation that is believed to cause much of the morbidity of infection. For example, S. haematobium oviposition in the bladder and genital tract results in hematuria, urinary tract fibrosis (sometimes leading to obstructive renal failure), increased susceptibility to HIV infection, and enhanced risk of bladder cancer. Van der Werf et al. estimated that in a 2 week period in 2000, 70 and 32 million individuals in sub-Saharan Africa experienced hematuria and dysuria associated with S. haematobium infection, respectively [2]. Significant S. haematobium-triggered bladder wall pathology and severe hydronephrosis were predicted to afflict 18 and 10 million people, respectively. Urogenital schistosomiasis appears to predispose individuals to earlier onset and more aggressive bladder cancers [3], [4]. Moreover, an estimated 150,000 deaths are attributable annually to S. haematobium-induced obstructive renal failure alone. Consequently, urogenital schistosomiasis is one of the most important causes of helminth-related mortality worldwide.
Despite the importance of this infection, the mechanisms by which urogenital schistosomiasis leads to these sequelae are poorly defined, particularly in the early stages of infection. This deficiency in knowledge is due largely to the lack of suitable animal models for S. haematobium infection [5]. Natural transdermal infection of mice or other rodents (e.g., hamsters) with S. haematobium cercariae typically results in hepatoenteric rather than urogenital disease [6]–[9]. Non-human primates recapitulate human disease but are expensive and controversial to use as animal models [10]. Both hamster and primate models suffer from having few species-specific tools.
To address the dearth of good animal models for urogenital schistosomiasis we recently developed a novel mouse model of S. haematobium egg-induced immunopathology [11]. In this model, a single direct injection of S. haematobium eggs into the bladder walls of mice recapitulates multiple aspects of human disease, including a regional and systemic Type 2 immune response, development of bladder granulomata, hematuria, bladder fibrosis, egg shedding, and urothelial hyperplasia. We sought to leverage this model to determine the early molecular events in the bladder occurring after introduction of S. haematobium eggs, as well as the temporal evolution of these processes. Microarray analysis demonstrated time-dependent, differential transcription of large numbers of genes. Many differentially transcribed genes were related to the canonical Type 2 anti-schistosomal immune response, collagen and metalloproteinase activity, urothelial barrier functions, and carcinogenesis pathways. Taken together, our findings have important implications for understanding host-parasite interactions and carcinogenesis in urogenital schistosomiasis, and may provide clues for novel therapeutic strategies for this disease and perhaps bladder cancer and bladder inflammatory disorders in general.
All animal work was conducted according to relevant U.S. and international guidelines. Specifically, all experimental procedures were carried out in accordance with the Administrative Panel on Laboratory Animal Care (APLAC) protocol and the institutional guidelines set by the Veterinary Service Center at Stanford University (Animal Welfare Assurance A3213-01 and USDA License 93-4R-00). Stanford APLAC and institutional guidelines are in compliance with the U.S. Public Health Service Policy on Humane Care and Use of Laboratory Animals. The Stanford APLAC approved the animal protocol associated with the work described in this publication.
S. haematobium-infected LVG hamsters were obtained from the National Institute of Allergy and Infectious Diseases Schistosomiasis Resource Center of the National Institutes of Health. Eggs were isolated from hamsters as previously described [11]. In brief, hamsters were sacrificed at 18 weeks post-infection, at which time livers and intestines were minced, homogenized in a Waring blender, resuspended in 1.2% NaCl containing antibiotic-antimycotic solution (100 units Penicillin, 100 µg/mL Streptomycin and 0.25 µg/mL Amphotericin B, Sigma-Aldrich), passed through a series of stainless steel sieves with sequentially decreasing pore sizes (450 µm, 180 µm, and 100 µm), and finally retained on a 45 µm sieve.
Egg injections of the mouse bladder wall were performed as previously described [11], [12]. Specifically, 7 to 8 week-old female BALB/c mice (Jackson Laboratories) were anesthetized with isoflurane, a midline lower abdominal incision was made, and the bladder exteriorized. Freshly prepared S. haematobium eggs (3,000 eggs in 50 µl of phosphate-buffered saline, experimental group) or uninfected hamster liver and intestinal extract (in 50 µl of phosphate-buffered saline, control group) was injected submucosally into the anterior aspect of the bladder dome. Abdominal incisions were then closed with 4-0 Vicryl suture, and the surgical site was treated once with topical antibiotic ointment. Mice were sacrificed at 1, 3, and 5 weeks post-injection (n = 3 for each time point/treatment group [egg vs. vehicle])
Mouse bladder micro-ultrasonography was performed as previously described [11]. At various time points after bladder wall injection, mice were anesthetized using vaporized isoflurane and their abdominal walls were depilated. Transabdominal images of the bladder were then obtained using a VisualSonics Vevo 770 high-resolution ultrasound micro-imaging system with an RMV 704 scanhead [40 MHz] (Small Animal Imaging Facility, Stanford Center for Innovation in In-Vivo Imaging).
Mice were sacrificed after bladder wall injection, and bladders processed for routine histology. Five µm sections were stained with hematoxylin and eosin or Masson's Trichrome-stained sections.
Bladders were removed from mice and placed in 10% phosphate-buffered formalin for 18 hours before dehydration and embedding in paraffin. Serial 5 µm sections were cut using a microtome, placed on positively charged glass slides and dried overnight at 56°C. Sections were heated to 100°C in citric acid buffer (pH 6.0) for 15 minutes. Endogenous peroxidase activity was blocked using 0.3% hydrogen peroxidase in methanol. An avidin-biotin complex immunoperoxidase protocol was employed, including staining of sections with anti-arginase-1 antibody (Clone 19/Arginase I, BD Biosciences) at a dilution of 1∶1000, biotinylated anti-mouse IgG1 at 1∶1000(BioLegend). Next, sections were incubated with streptavidin labeled horseradish peroxidase (Biocare Medical, Concord, CA) followed by use of a DAB chromogen substrate kit (Biocare Medical, Concord, CA) and counterstaining with hematoxylin.
Mice were sacrificed and bladders were immediately collected and preserved in RNA Later (Ambion) at −80°C. RNA was extracted for each individual bladder by Trizol Reagent (Ambion) and was reverse transcribed to cDNA. RNA yields were measured using a QuBit 2.0 Fluorimeter (Life Technologies, Grand Island, NY, USA) and quality was assessed using an Agilent Bioanalyzer and RNA 6000 Nano Labchips (Agilent Technologies, Foster City, USA). Next, individual bladder cDNA was in vitro transcribed to synthesize cRNA using Illumina TotalPrep RNA Amplification Kits (Ambion, Applied Biosystems, Foster City, CA), and hybridized using standard Illumina protocols on the MouseWG-6 v2.0 chip (Stanford Functional Genomics Facility, Stanford, CA). Illumina Beadstation-generated scanned array files were filtered and normalized (quantile method) in GeneSpring GX version 11. Genes in egg- versus control vehicle-injected bladders were considered differentially transcribed at various time points if they were transcribed ≥2-fold and p<0.05 by unpaired T-test. Functional annotation of differentially transcribed genes was performed using DAVID (DAVID Bioinformatics Resources 6.7, National Institute of Allergy and Infectious Diseases [NIAID], NIH) in combination with the Biocarta and KEGG pathway databases [13], [14].
Mice were sacrificed after bladder wall injection, and bladder RNA preserved and isolated as above. RNA yields and quality were measured as above. cDNA was synthesized from the RNA of individual mouse bladders. Primer sequences for genes of interest were obtained from PrimerBank (http://pga.mgh.harvard.edu/primerbank/). (Table S3) GAPDH was used as a housekeeping gene. Real-time PCR was performed using SYBR Green and an Mx3005p thermal cycler (Stratagene). Cycle thresholds (Ct) were calculated for each reaction. Using the comparative Ct method relative gene transcription was calculated as 2−ΔΔCt, where ΔCt = Ct (gene of interest) - ΔCt (normalizer = β-actin). ΔΔCt was calculated as ΔCt (egg-injected) - ΔCt (calibrator). Correlations between microarray and real-time PCR results were assessed using Spearman's Rho measure of correlation in Microsoft Excel 2010 for Windows.
Microarray-based comparisons of S. haematobium egg- versus control vehicle-injected bladders demonstrated differential gene transcription over time (Figure 1 and Table S1). At one week post-egg injection, 279 and 22 genes featured significantly more and less transcription (≥2-fold and p<0.05), respectively. By three weeks post-injection, more genes were differentially transcribed, with 1001 and 570 genes demonstrating more and less transcription, respectively. At five weeks post-injection, fewer genes demonstrated altered transcription, with 794 and 308 genes exhibiting more and less transcription, respectively. Functional annotation clustering of genes featuring ≥2-fold differential transcription indicated that each of these gene clusters was associated with a discrete gene ontology (Table S2). Many of these clusters were related to immune responses. Illumina probe IDs and Entrez Gene IDs of selected genes are provided (Table S1).
Validation of a subset of the microarray data was performed using real-time PCR (Figure 2). Specifically, genes related to inflammation (IL4, CCL2, CCL11, inducible nitric oxide synthase [iNOS], arginase-1, and CD68), urothelial function (uroplakins 1A, 1B, 2, 3A, and 3B and claudin-8), and collagen (collagen type 3α1, 4α5, and 17 α1) were assayed. Overall, correlations between microarray and PCR findings were tight, with Spearman's correlation r = 0.87, 0.94, and 0.84 at 1, 3, and 5 weeks post-injection, respectively.
Bladder wall injection with S. haematobium eggs induced increased transcription of numerous genes related to granulomatous inflammation and type 2 immunity (selected examples in Table 1). Specific genes which exhibited increased transcription included cytokine and cytokine-related genes such as IL-4, IL-4-induced 1, IL-1β, IL-6, interferon gamma-inducible proteins (IFI30 and IFI47), TGF-β, IL-13 receptor alpha 2, IL-10 receptor alpha, and cytokine inducible SH2-containing protein (CISH); markers of macrophages including macrophage-expressed gene 1 (MPEG1, general macrophage marker), and arginase, Ym1 (CHI3L3), and mannose receptor C type 1 (all alternatively activated macrophage markers); and chemokines such as CCL4 (MIP-1 beta), CCL5 (RANTES), CCL11 (eotaxin) and CXCL1 (KC). In fact, KEGG pathways analysis confirmed that one of the biological clusters featuring the greatest number of differentially transcribed genes at one and three weeks post-egg injection was “cytokine-cytokine receptor pathways” (1 week: 15 genes, 3 weeks: 41 genes, Tables 2 and 3). By 5 weeks post-egg injection, the biological cluster with the greatest number of differentially transcribed genes (26) was “B cell receptor signaling pathway” (Table 4). Accordingly, numerous IgG and IgE Fc receptor genes featured increased transcription. Other immune response-related genes exhibiting differential transcription included those encoding for eosinophil ribonucleases, C1q, arachidonate 15-lipoxygenase (ALOX15), and platelet thromboxane A synthase 1 (TBXAS1).
Granuloma formation was confirmed in vivo by transabdominal bladder microultrasonography (Figure 3a–b) and histologically by hematoxylin and eosin staining of egg-injected bladders (Figure 3c–f). Finally, type 2 inflammation was verified through immunohistochemical staining for arginase-1, an enzyme strongly associated with type 2 immunity-mediated alternative activation of macrophages (Figure 3k).
Delivery of S. haematobium eggs to the bladder prompted global decreased transcription of all uroplakin genes, in addition to several tight junction-related genes, at three weeks post-egg injection (claudins and junctional adhesion molecule-4, Table 5). Interestingly, this occurred in the context of egg shedding in urine (Figure 4) and profound urothelial hyperplasia (Figure 3g–j), a precursor lesion for bladder cancer. Indeed, pathways analysis of microarray data implicated involvement of various carcinogenesis-related signaling pathways at 5 weeks post-injection, including vascular endothelial growth factor-, oncogene-, and mammary tumor-related genes (Tables 6, 7, 8).
The presence of S. haematobium eggs in the bladder prompted complex patterns of differential transcription of multiple collagen and metalloproteinase genes over time (Table 9). We confirmed through Masson's Trichrome staining that transcription of these extracellular matrix-associated genes was temporally associated with bladder tissue remodeling and fibrosis (Figure 5).
Although S. haematobium infection is one of the most important causes of helminth-related mortality worldwide, research on this important parasite has suffered due to a lack of high fidelity animal models. We recently demonstrated that direct injection of S. haematobium eggs into the bladder walls of mice recapitulates many features of human urogenital schistosomiasis, including granulomatous inflammation, urothelial hyperplasia, egg shedding, and bladder fibrosis [11]. Since oviposition is induced at a precisely known time point, our synchronous granuloma model is ideal for dissecting the initial biological responses that occur in the bladder after egg exposure. We applied gene microarray approaches to our mouse model in order to interrogate the early molecular events associated with the bladder sequelae of urogenital schistosomiasis. This first-ever microarray analysis revealed complex modulation of multiple genes, with a peak occurring 3 weeks after egg exposure. As expected, type 2 inflammation- and macrophage function-associated gene transcription was increased. Extracellular matrix remodeling-related gene transcription was differentially modulated over time. Pathways analysis pointed to differential transcription of multiple genes implicated in carcinogenesis. Surprisingly, microarray analysis uncovered decreased transcription of certain tight junction and all uroplakin genes, which occurred despite urothelial hyperplasia.
Clearly, S. haematobium eggs induced a complex bladder gene response that waxed and waned during the time course examined. Few genes were differentially transcribed one week after bladder injection with S. haematobium eggs. By three weeks after egg injection of the bladder, the numbers of differentially transcribed genes had peaked. For example, by five weeks post-egg injection, the pool of differentially transcribed genes was already contracting. This suggests that the chronic bladder changes seen in urogenital schistosomiasis cannot be sustained by a single set of eggs; rather, it is driven by continuous oviposition by adult worms. In this model, successive waves of oviposition, rather than any lone egg bolus (such as that featured in our model), would sustain a long-term bladder response. This is consistent with observations that early stage schistosomal urinary tract pathology eventually resolves after praziquantel therapy-induced worm death (which leads to cessation of oviposition) [15]. Regardless, our single bolus model of synchronous granuloma formation proved valuable for characterizing the earliest molecular events in the bladder that occur in response to exposure to S. haematobium eggs.
As expected, many of the early molecular events in the bladder were related to granulomatous type 2 inflammation. Schistosome eggs are potent inducers of granuloma formation in various tissues, including the intestines, liver, lung, and bladder. These granulomata feature various leukocyte subsets, including eosinophils, neutrophils, lymphocytes, macrophages, and epithelioid cells (activated macrophages). Pathways analysis suggested a role for B cells, given that a number of relevant genes were differentially transcribed. We suspect that many of these genes promote immunoglobulin functions, including those involved with IgE, the isotype most commonly associated with helminth infection (reviewed by Erb [16]). Accordingly, our mouse model features IgE production [11]. Granuloma formation in mouse models of S. mansoni and Schistosoma japonicum infection (i.e., hepatoenteric schistosomiasis) is associated with a local and systemic type 2 inflammatory response [17], [18]. This immune polarization typically features elevated levels of IL-4 and IL-13 which trigger alternative activation of macrophages. These macrophages selectively metabolize arginine through arginase-1, whereas their classically activated counterparts convert arginine to nitric oxide through nitric oxide synthase [19]–[21]. In our mouse model of urogenital schistosomiasis we have demonstrated that bladder granuloma formation is likewise associated with mixed leukocyte infiltration and regional and systemic production of type 2 cytokines [11]. Herein we have extended these findings by confirming granulomatous inflammation at additional time points and enhanced gene transcription for several chemokines, IL-4, the alternatively activated macrophage markers arginase-1, mannose receptor, and Ym-1/CHI3L3, and other indicators of type 2 inflammation. Although IL-13 gene transcription was unchanged, we have previously documented elevated protein levels of this cytokine from 1 through 4 weeks post-egg injection [11]. IL-13 has been implicated as a major mediator of fibrosis associated with S. mansoni egg exposure [22]–[29]. The pro-fibrogenic role of IL-13 is kept in check by the decoy receptor IL-13 receptor alpha 2 [28], [30]–[33]. Indeed, transcription of the IL-13 receptor alpha 2 gene was increased in our model, suggesting a possible role in negative feedback on IL-13-mediated fibrosis. Other genes with dampening influences over S. mansoni-induced inflammation and fibrosis include RELM-alpha/Fizz1 34,35, IL-10 [32], [36]–[38], and arginase [39]. Of these three mediators, only arginase featured increased transcription (although IL-10 receptor alpha, but not IL-10, also demonstrated enhanced transcription). Thus, we hypothesize that arginase may have a role in resolving bladder fibrosis. Additional studies will be necessary to clarify this issue. Finally, the observed protein expression of arginase-1 and lack of increased iNOS transcription is consistent with selective polarization of macrophages along the alternative activation program.
Another relevant issue is whether the immune and fibrosis responses to S. haematobium eggs in the bladder differ from those directed against S. mansoni eggs in other tissues. Perhaps the most appropriate comparisons can be made between our data and microarray analyses that have employed the S. mansoni egg-induced, synchronous lung granuloma model [40], [41]. These studies are methodologically analogous to this study's microarray analysis of our synchronous egg injection model. Numerous immune and fibrosis response genes feature increased transcription in both the S. mansoni and S. haematobium models. These genes include those encoding for CCL4 (MIP-1β), IL-4 induced 1, IL-6, cytokine inducible SH2-containing protein (CISH), C1q, IgG Fc receptors, eosinophil ribonucleases, arginase, arachidonate 15-lipoxygenase (ALOX15), platelet thromboxane A synthase 1 (TBXAS1), MMP9 and MMP13. Nonetheless, several notable genes showing elevated transcription in the S. mansoni studies did not feature differential transcription in our datasets, namely MCP1 (CCL2), CCR9, MCP2 (CCL8), and MMP12. The transcription patterns for MMP9, MMP12, and MMP13 may reflect distinct collagen-remodeling pathways in S. mansoni lung versus S. haematobium bladder fibrosis. Surprisingly, some genes with increased transcription in Th1-skewed mouse strains from the S. mansoni studies were also increased in our microarray analysis, which was based on the Th2-skewed BALB/c strain. These genes include IL-1β, interferon gamma-inducible proteins, CCL5 (RANTES), and macrophage-expressed gene 1 (MPEG1). Hence, our model seems to feature increased transcription of a greater mix of type 1 and 2 immune response-associated genes as compared to the S. mansoni studies. We conclude that although the S. haematobium egg-directed immune and fibrotic response in the bladder shares many similarities to the S. mansoni egg-triggered lung response, there are a number of potentially important disparities. There is a precedent in the literature for schistosome- and tissue-specific immune and fibrotic responses. Liver- and lung-associated, S. mansoni egg granulomata develop in a highly organ-specific fashion [42]. S. japonicum granulomata also evolve in a tissue-specific manner in the liver, lung, and intestinal tract [43]. These reports highlight the critical need to develop in vivo models which properly match schistosome species with their tropism for specific host organs.
Besides validating our prior immunologic- and fibrosis-related observations, microarray analysis also identified complex modulation of genes integral to urothelial function. Uroplakins are key structural proteins that form organized plaques on the surface of urothelial cells. The coordinated expression of the various uroplakin genes is believed to confer both impermeability and flexibility to the urothelium. These two functions are critical to the bladder's ability to safely sequester accumulating urine and expel it during micturition. Accordingly, mice deficient for various uroplakins exhibit bladder dysfunction and increased urothelial permeability [44], [45]. We were intrigued by the finding that the transcription of all uroplakin genes was dampened at three weeks after S. haematobium egg introduction to the bladder. Other workers have reported less uroplakin gene transcription and urothelial hyperplasia after administration of disparate noxious stimuli to the bladder, including Bacillus Calmette-Guerín (BCG) [46], cyclophosphamide [47]–[49], and an in vitro model of culture media-induced urothelial hyperplasia [50]. Urothelial hyperplasia in this setting is a secondary response to reseal urothelial defects that have resulted from desquamation of dead urothelial cells [51], [52]. Strikingly, this urothelial response appears to be conserved across species, given that these studies have encompassed mice, rats, and pigs. To our knowledge, we report here the first characterization of uroplakin changes triggered by urogenital schistosomiasis. In our past work we have identified the onset of urothelial hyperplasia after S. haematobium egg injection [11], and herein have corroborated this at other time points. Taken together, this suggests that decreased uroplakin gene transcription and urothelial hyperplasia are coupled processes that are part of a conserved bladder response to a range of forms of urothelial injury. In our model, decreased transcription of uroplakin genes and urothelial hyperplasia occurs in the setting of egg shedding in urine. We speculate that S. haematobium eggs induce, or at least exploit, the compromised urothelial barrier (i.e., decreased uroplakins) to pass into urine. These possibilities highlight the complex dynamics of urothelial biology in urogenital schistosomiasis. Our findings also underscore the importance of high fidelity urogenital schistosomiasis models. Namely, extrapolation of S. haematobium urogenital disease mechanisms from S. mansoni hepatoenteric disease models is not possible, given the tissue-specific expression of uroplakins.
Loss of uroplakin expression has been noted in some urothelial cancers [53], [54]. Although it is tempting to causally link these observations to our findings and schistosomal bladder cancer, the lack of reports of development of bladder cancer in uroplakin-deficient mice suggests that loss of uroplakin expression may not be carcinogenic. Conversely, we identified differential transcription of genes implicated in multiple carcinogenesis pathways, including vascular endothelial growth factor (VEGF)-, oncogene-, and mammary tumor-related genes. Tissue, plasma, and urine levels of VEGF have been reported to be elevated in patients with schistosomal bladder cancer [55]. Our past work is consistent with these findings in that bladder VEGF increases after egg injection [11]. We speculate that elevated VEGF in the bladder may promote cancer progression by stimulating tumor vasculogenesis. In addition, we also conjecture that VEGF mediates growth of abnormal, friable blood vessels which rupture and leak blood through the uroplakin-poor, compromised urothelium, ultimately resulting in the hematuria associated with urogenital schistosomiasis.
Additional evidence for a compromised urothelial barrier was identified in the form of dampened transcription of tight junction-related genes. While uroplakins contribute to the urothelium's water and urea permeability barrier, tight junctions have a complementary role. These structures confer transmembrane epithelial resistance to the urothelium [56]. We found that gene transcription of claudin-8 and junctional adhesion molecule-4, components of tight junctions in the bladder urothelium [50], [57]–[59], was lessened after egg injection. Bladder urothelial tight junction expression of claudins and junctional adhesion molecules is found in rats, mice, rabbits, pigs, and humans, which highlights the tight conservation of these genes across mammalian species and hints at their biological importance.
Bladder fibrosis, disorganized deposition of extracellular matrix in the bladder wall, is another critical biological process associated with urogenital schistosomiasis. We have previously reported induction of bladder fibrosis after S. haematobium egg injection that resembles human disease [11]. Here, we also histologically confirmed fibrosis at additional time points that correlated with fibrosis-related gene transcription. Specifically, we identified augmented transcription of the collagen genes COL7A1 and COL17A1, metalloproteinases-3, -9, -10, and -13, and tissue inhibitor of metalloproteinase 1 (TIMP1). Interestingly, transcription of other collagen genes, i.e., COL3A1, COL4A5, COL6A3, and COL8A2, were decreased after egg injection. Normal levels of COL3A1 have been shown through transgenic mouse studies to be important for proper bladder function [60]. In addition, MMP9, MMP13, and TIMP1 are expressed in many bladder cancers and may mediate tumor invasiveness through extracellular matrix regulation [61], [62]. The differential transcription of numerous genes linked to extracellular matrix remodeling underscores the intricate fibrosis- and cancer-promoting pathways associated with schistosomal granuloma formation.
Although our findings are highly informative, the employed mouse model features limitations. Since Mus musculus and S. haematobium are both eukaryotes, theoretically speaking these species may feature homologous genes. However, we believe that the probability of orthologs sharing significant nucleotide homology is quite low. Codon usage in mice and trematodes such as S. haematobium is different, as exemplified by the need for codon optimization to maximize expression of Schistosoma genes by mammalian cells [63], [64]. Moreover, we only injected 3000 eggs per mouse bladder, and eggs are shed in the urine over time. Hence, S. haematobium RNA is only a small fraction of the total RNA in egg-injected mouse bladder tissue, making it unlikely that S. haematobium cRNA hybridization to microarray chips (if any) significantly affected our analysis. Another limitation is that our egg injection model does not reproduce the entire S. haematobium life cycle of the human host. Cercariae are found in the skin and subcutaneous tissues, schistosomula circulate in the systemic and portal circulation, and adult worms reside in the pelvic venous plexus. However, for the purposes of this bladder-focused study, eggs are sufficient since it is the only S. haematobium life stage present in bladder tissue. It is also possible that S. haematobium eggs transferred from hamsters (the source of eggs in our model) to a final mouse host may be metabolically and immunologically distinct from eggs laid in situ in a single mouse host. This has been postulated to occur with S. japonicum eggs [43]. We have controlled for this in part by using control injections of hamster liver and intestine-derived “vehicle”. However, studies are underway to compare in vitro laid eggs to hamster-derived eggs in our mouse model. The single, large egg bolus utilized in our model is unlikely to reflect the kinetics of continuous oviposition in the human host. On the other hand, our model features highly focal, coalescing multiple egg-based granulomata that are reminiscent of those observed in the human bladder [65]. Regardless of these issues, our model mitigates the relative scarcity of early disease stage bladder tissue available for research. Bladder tissue from schistosomiasis patients is typically only available from those undergoing cystectomy or bladder reconstructive surgery for advanced bladder cancer or fibrosis, respectively. Thus, our model fills an important niche in basic research on urogenital schistosomiasis.
In conclusion, we have defined for the first time the initial molecular underpinnings of the bladder response to S. haematobium eggs in an experimental model of urogenital schistosomiasis. Although this response is manifold, it is discrete, involves known inflammatory, fibrosis, epithelial, and cancer-related pathways, and expands and contracts over time. This work may direct future efforts to develop diagnostic and therapeutic tools for the bladder sequelae of urogenital schistosomiasis and potentially cancers and inflammatory disorders of the bladder in general.
|
10.1371/journal.ppat.1004275 | Motility and Chemotaxis Mediate the Preferential Colonization of Gastric Injury Sites by Helicobacter pylori | Helicobacter pylori (H. pylori) is a pathogen contributing to peptic inflammation, ulceration, and cancer. A crucial step in the pathogenic sequence is when the bacterium first interacts with gastric tissue, an event that is poorly understood in vivo. We have shown that the luminal space adjacent to gastric epithelial damage is a microenvironment, and we hypothesized that this microenvironment might enhance H. pylori colonization. Inoculation with 106 H. pylori (wild-type Sydney Strain 1, SS1) significantly delayed healing of acetic-acid induced ulcers at Day 1, 7 and 30 post-inoculation, and wild-type SS1 preferentially colonized the ulcerated area compared to uninjured gastric tissue in the same animal at all time points. Gastric resident Lactobacillus spp. did not preferentially colonize ulcerated tissue. To determine whether bacterial motility and chemotaxis are important to ulcer healing and colonization, we analyzed isogenic H. pylori mutants defective in motility (ΔmotB) or chemotaxis (ΔcheY). ΔmotB (106) failed to colonize ulcerated or healthy stomach tissue. ΔcheY (106) colonized both tissues, but without preferential colonization of ulcerated tissue. However, ΔcheY did modestly delay ulcer healing, suggesting that chemotaxis is not required for this process. We used two-photon microscopy to induce microscopic epithelial lesions in vivo, and evaluated accumulation of fluorescently labeled H. pylori at gastric damage sites in the time frame of minutes instead of days. By 5 min after inducing damage, H. pylori SS1 preferentially accumulated at the site of damage and inhibited gastric epithelial restitution. H. pylori ΔcheY modestly accumulated at the gastric surface and inhibited restitution, but did not preferentially accumulate at the injury site. H. pylori ΔmotB neither accumulated at the surface nor inhibited restitution. We conclude that bacterial chemosensing and motility rapidly promote H. pylori colonization of injury sites, and thereby biases the injured tissue towards sustained gastric damage.
| H. pylori is a disease-causing bacterium that commonly infects the human stomach in both developed and underdeveloped countries. Infected individuals can develop digestive diseases, including stomach inflammation, peptic ulcer, and cancer. There has been only limited investigation into the events when H. pylori first interacts with stomach tissue. Using anesthetized mice in which we have induced microscopic damage to the stomach surface, we find that H. pylori is able to rapidly detect and navigate towards this damage site. Within minutes, bacterial accumulation slows repair of the damage. This is the earliest event of H. pylori pathogenesis that has been reported in vivo. We further define that this pathology is due to the bacterial accumulation at damage sites and that this also occurs in a model of larger stomach damage (ulceration). The broader implications of our work are that even sub-clinical insults to the stomach that occur in daily life (damage from grinding of food, ingestion of alcohol, taking an aspirin) can potentially attract H. pylori and not only slow repair of any existing damage, but maybe also provide an initiation site that can start the pathogenic sequence of stomach disease caused by H. pylori.
| H. pylori infection promotes gastritis, gastric ulceration, and gastric cancer [1]. The mechanisms of early H. pylori interaction with gastric tissue have not been explored, but are essential for successful colonization and ultimately the disease consequences of H. pylori infection. H. pylori is believed to migrate through the gastric environment by chemotaxis, using its multiple flagella [2], [3]. Indeed, motility and chemotaxis have been shown to lower the dose of H. pylori needed to establish an infection [2], [3]. In vitro, H. pylori motility responds to various conditions, including CO2, urea/ammonium, arginine, bacterial energy status and low pH [4]–[9]. Additionally, H. pylori has been shown to use gradients that are sensitive to pH to localize close to the gastric epithelium in vivo [7]. We and others have shown that damage to the gastric epithelium (commonly produced by alcohol, salt intake, smoking, and non-steroidal anti-inflammatory drugs usage) can affect the luminal microenvironment adjacent to tissue [10]–[13]. We were thus curious whether H. pylori can sense and respond to these damage-induced microenvironments using chemotaxis, and whether these regions might serve as preferential colonization sites.
H. pylori uses largely canonical chemotaxis signal transduction to promote chemotaxis [4]. H. pylori senses its environment through four chemoreceptors TlpA, TlpB, TlpC, and TlpD that detect conditions including pH, autoinducer 2, CO2, arginine, urea/NH4 and molecules that lead to bacterial energy generation [4]–[6], [8], [9], [14], [15]. Ligand interactions with chemoreceptors control the direction of flagellar motor rotation via the Che family of proteins [4]. The most downstream member of the Che protein family, CheY, directly interacts with the flagellar motor to control flagellar rotational direction [4], [16], [17]. Generation of torque for flagellar rotation is regulated by members of the Mot family of proteins, MotA and MotB [4]. Isogenic mutant strains lacking the flagellar motor protein MotB (ΔmotB) retain wild-type flagellar structure (Fla+), but the Mot− flagella are non-functional [2]. H. pylori lacking the motB gene have no motility and a substantially reduced ability to colonize the stomach initially and to multiply to full bacterial loads [2], [4]. Mutants lacking CheY (ΔcheY) retain functional flagella (Mot+) but cannot change the direction of flagellar rotation, and thus are non-chemotactic (Che−) [3], [4], [18]. ΔcheY swim with fewer direction changes as compared to wild-type H. pylori [19], [20].
In vivo, chemotaxis promotes multiple aspects of H. pylori gastric colonization. Che− H. pylori similarly show a reduced ability to colonize the stomach, particularly at early time points [3], [18], [21], [22]. In addition, they lose the ability to locate close to the epithelial surface and deep in the antral gastric glands [3], [21]. Thus, evidence supports that chemotaxis aids colonization, but whether it is directed toward sites of damage is not known.
In the present study, we test the role of H. pylori chemosensing and motility in response to damaged tissue in vivo, using acetic-acid ulcer induction as a long term injury/repair model, and then extend results to short term injury/repair using two-photon induced gastric epithelial damage.
We first confirmed the motility of wild-type H. pylori (Sydney Strain 1: SS1) used in this study by tracking individual fluorescently labeled H. pylori using a fast-scan confocal microscope (Movie S1). As shown in Figures S1 and S2, fluorescently labeled wild-type H. pylori (n = 6 experiments) had motility properties (swimming in straight or arced lines at mean velocity 25.5±0.7 µm/sec, number of stopping: 0.13±0.05 per sec, frequent changes in speed, slowed motility at pH 5 and pH 4) consistent with previous findings [19], [20], [23]. Results from fluorescently labeled H. pylori were indistinguishable from unlabeled H. pylori (velocity: 28.5±1.1 µm/sec, number of stopping: 0.15±0.06 per sec, n = 6).
Acetic acid-induced ulcer models are well established and resemble human ulcers in terms of both pathological features and healing process [24]. Murine gastric ulcers were induced by serosal topical application of acetic acid to the exterior of the intact stomach corpus, then two days later, when ulcer size is maximal, wild-type SS1 H. pylori was gavaged into the ulcerated stomach. We evaluated both ulcer size and H. pylori colonization at days 3, 9 and 32 after ulcer induction (which are respectively, Day 1, 7, and 30 after inoculation; as shown in the Figure 1A diagram). Wild-type H. pylori delayed ulcer healing 7 Days after inoculation, in a dose dependent manner, as compared to almost complete ulcer healing in the absence of H. pylori over the same time interval (Figure 1B and 1D). Inoculation with ≥106 H. pylori resulted in reproducible detection of colony forming units (CFU) from the small ulcerated area (Figure 1C) as well as significantly impaired ulcer healing, so was used as the routine inoculum. Looking at a more extensive time course, ulcer size diminished in the uninfected control group approximately 6-fold in the first 9 days after damage, and no macroscopic ulcer was observed 32 days after damage (Figure 2A). H. pylori had no effect on initial damage size at day 3 (1 Day after inoculum), however H. pylori significantly delayed ulcer healing such that at day 32 the ulcers were still prominent (Figure 2A).
In the same experiments, we compared the time course of bacterial colonization in the ulcerated tissue versus the contralateral region of undamaged tissue on the opposite side of the same stomach, using small tissue regions collected as shown in Figure 1D. We used two independent methods to measure H. pylori colonization of tissue: bacterial culture to obtain CFU (Figure 2B) and PCR of the H. pylori ssa gene (Figure S3) [25]. Control experiments confirmed the accuracy of the ssa PCR method versus CFU determination (Figure S3). Both methods identified significantly higher H. pylori colonization at the ulcer site versus the undamaged area of gastric tissue, at 1, 7 and 30 Days after inoculation (Figure 2B and S3). Both methods also reported undetectable levels of H. pylori in control (uninfected) tissues, confirming the specificity of our techniques. Additional control experiments examined conditions of no surgery and sham surgery (in the latter case applying PBS instead of acetic acid) to confirm that H. pylori colonization was not affected by these factors at either Day 1 or Day 7 after inoculum (Figure S4). Figure S4 also confirms that bacterial growth between Day 1–7 is not affected by surgery.
We also measured whether ulceration changed the total gastric bacterial load, and separately quantified gastric Lactobacilli. In mice not treated with H. pylori, PCR of 16S rRNA reported no difference in total bacterial load between the ulcers and healthy regions, suggesting that ulceration did not generally increase bacterial colonization (Figure 2C). As predicted, gavage with H. pylori increased total bacterial load versus the control condition in absence of both ulceration and H. pylori (Figure 2C). PCR for the genus Lactobacillus, a documented stomach commensal, confirmed the presence of the bacteria in control and H. pylori infected stomachs, but Lactobacillus did not preferentially colonize ulcerated tissues in either the absence or presence of H. pylori (Figure 2D). As reported by others, we observed that Lactobacillus increased at 30 Days after H. pylori inoculation [26]. Since this was not observed at earlier time points, this H. pylori-induced gastric microbiota change is a more long time effect. Combined, these results suggest that the observed preferential colonization of damaged tissue is specific to H. pylori.
We examined two proteins by real-time PCR (Figure 3) and immunofluorescence (Figure 4–6) to interrogate tissue architecture and H. pylori tissue localization during the ulcer repair cycle. Trefoil factor 2 (TFF2) is a marker for gastric mucous cells, and a peptide that has been shown to promote repair of gastric damage [12]. H,K-ATPase mediates the gastric acid secretion that can oppose ulcer healing [24] and is a classic marker of parietal cells. Three days after ulcer induction in the absence of H. pylori, H&E staining (Figure 4A) showed many necrotic or apoptotic cells present at the ulcerated area (the ulcer crater recognized by macroscopic imaging in Figure 1D), and both PCR of ulcerated tissue (Figure 3A) and immunostaining (Figure 4A) showed the same area depleted of parietal cells. At this time point, H. pylori inoculation (for 24 hrs; Day 1) did not change these features recognized by macroscopic imaging (Figure 1D), PCR (Figure 3A) or histology and immunostaining (Figure 4A). H. pylori did, however, increase the presence of TFF2 recognized by both PCR and immunostaining (Figures 3B and 4A). By immunostaining, H. pylori bacteria were identified at the tissue surface and within gastric gland lumens (Figure 4B), while we failed to detect H. pylori in non-H. pylori infected stomachs (Figure S5). In infected stomachs, H. pylori appeared most abundantly in the ulcerated tissue and at ulcer margins (Figure 4B,b–c), but were also identified in undamaged tissue (Figure 4Ba).
In tissue at Day 7 (Figure 5) and Day 30 (Figure 6) after H. pylori inoculation (or 9 and 32 days post-ulcer induction), we observed the gastric epithelium still displayed altered distributions of H,K-ATPase and TFF2. In the uninfected tissues at both of these later time points the ulcer was healed and the epithelium had fully regenerated, but both immunostaining and PCR reported a sustained decrease in parietal cells (H,K-ATPase) and increase of mucous cells (TFF2) in the area that had been damaged. A more extensive comparison of the stomach distribution of these two proteins across the corpus at day 9 (Figure S6) confirms that these alterations are restricted to the area of gastric damage. In the H. pylori infected tissues at these later time points, no ulcer crater was visible macroscopically 32 days after ulceration (Day 30 after H. pylori inoculum), but histologically the tissue showed partial epithelialization and inflamed tissue at ulcer margins (Figure 5A & 6A). In this condition, both PCR and immunostaining report that infected ulcerated tissue has reduced parietal cells (H,K-ATPase) compared to infected tissue from non-ulcerated areas. In contrast, PCR reports an elevation in TFF2 in both intact and ulcerated areas of the infected tissue suggesting a general effect of H. pylori infection on TFF2 mRNA abundance (Figure 3B). In contrast, although TFF2 immunostaining is generally elevated in the damaged and ulcer margin areas at these later time points, we could not discern a consistent increase in TFF2 protein levels in general non-ulcerated tissue (Figure 5 and 6). Differences between mRNA and protein levels are commonly observed, and in this case discrepancies may be compounded in the PCR analysis by the need to collect tissues with variable amounts of ulcerated tissue over this repair time course. H. pylori bacteria were identified at both tissue surface and within gastric gland lumens at Day 7 after inoculation, but only within glands at Day 30 (Figure 5B and 6B). Qualitatively, H. pylori density appeared highest at ulcer margins. These results suggest that H. pylori preferentially colonizes the ulcerated tissue at the early time point, followed by migration/expansion to surrounding viable epithelial cells as colonization is established.
To test the importance of H. pylori motility and chemotaxis in the preferential colonization of ulcerated tissue at early time points (1–7 Days after inoculation), we used H. pylori ΔmotB or ΔcheY mutants. Prior to use, we confirmed the previously described motility phenotypes of these strains [2], [3]. As expected, ΔmotB mutants are not motile, ΔcheY mutants are motile but do not change direction, and wild-type strains display normal motility and pH sensitivity (Figure S1 and S2). After inoculation with 106 ΔmotB, we failed to detect H. pylori colonization in the stomach (Figure 7A). However, ΔmotB could colonize the stomach if we infected with a higher dose of 108 (Figure 7A). Under this condition, we recovered ∼100 fold lower CFU/g of this mutant as compared to wild-type (Figure 7A). These findings are consistent with the documented elevated infectious dose and poor colonization properties previously reported for strains lacking motility [2]. Although there was moderate colonization by the ΔmotB strain, it did not inhibit ulcer healing (Figure 7B) as compared to the wild-type. In addition, H. pylori ΔmotB displayed no difference in the colonization of undamaged and ulcerated areas of stomach at either time point (Figure 7A). In contrast, 106 ΔcheY colonized the stomach initially at low levels that increased over time (Figure 7A), and ΔcheY inhibited ulcer healing (Figure 7B). Results suggest ΔcheY grows reasonably in gastric tissue, consistent with previous findings [27]. In contrast to what we observed with wild-type H. pylori, ΔcheY colonized undamaged and ulcerated areas of stomach equally at either time point (Figure 7A). These results suggest that motility and chemotaxis facilitate the ability of H. pylori to colonize damaged gastric tissue to elevated levels.
In Figure 8, we plotted all data of wild-type and mutant H. pylori CFU/g in ulcerated regions versus the ulcer size from the same animal. As expected, 24 hr after inoculation, there is no correlation between ulcer size and H. pylori CFU (Figure 8A). In contrast, at Day 7 after inoculation (9 days after ulceration), we observed a strong correlation between bacterial load of all H. pylori genotypes and ulcer size (Figure 8B). This correlation suggests that bacterial load leads to compromised ulcer healing independent of H. pylori genotype. Once colonization is established, other virulence factors independent of chemotaxis and motility may slow ulcer repair. These results also suggest that chemosensing and motility help H. pylori target regions of tissue damage, and increase the efficiency of colonization.
Our observations above show that H. pylori preferentially colonizes damaged tissue within one day after inoculation, but it was not clear when within this 24 hour period H. pylori localized with the damaged tissue, so we examined H. pylori behavior at earlier time points. Pilot experiments applied fluorescently tagged wild-type H. pylori directly to the surface of the surgically exposed gastric mucosa of anesthetized animals, whose stomachs were previously ulcerated by acetic acid. As shown in Figure S7, intravital confocal microscopy reported wild-type H. pylori were found more abundantly in ulcerated areas than in intact tissue within 1 hr of bacterial addition. In contrast, ΔcheY was less abundant in both intact and ulcerated tissue (Figure S7). This model was not pursued further since acetic acid ulcers were too large and heterogeneous to test in the same microscopic field if H. pylori selectively accumulates at a site of injury.
Therefore we used established two-photon microscopy methods to induce microscopic epithelial lesions in surgically exposed mouse gastric mucosa and then tracked repair in real-time over 15 min [11], [12], [28]. In initial experiments, we added fluorescent beads (1.0 µm diameter) to the gastric luminal superfusate, then induced photodamage of 3–5 gastric surface epithelial cells by two-photon laser. We then measured the rheological properties of the in vivo injury site environment. As shown in Figure 9B, S8 and Movie S2, fluorescent beads moved away from gastric tissue after damage, suggesting that the injury creates fluid flow away from the tissue into the lumen. Fluorescently labeled H. pylori ΔmotB showed a similar passive flow away from the injury site in response to gastric damage (Figure 9B). In contrast, wild-type H. pylori rapidly accumulated at the two photon-induced damage site (Figure 9B) and this tissue accumulation was not observed in the absence of imposed damage (Figure 9A). H. pylori ΔcheY gradually accumulated at the gastric surface, but ΔcheY did not show an obvious affinity for the site of gastric damage. Results from such time course experiments were quantified by comparing surface intensity of H. pylori fluorescence at the site of damage versus surface regions (>50 µm) distant from damage (Figure 10C–E). Results from multiple experiments are compiled in Figure 10F, showing that ΔcheY had more limited surface accumulation than wild-type but greater than the strain lacking motility, and that only wild-type responded to damage with a preferential accumulation at damage sites.
In the same experiments evaluating bacterial surface accumulation in real time, we measured speed of gastric repair. Figure 10A shows the time course of repair, and Figure 10B shows the value of repair rate compiled from multiple experiments. Compared to the absence of bacteria (vehicle control), both wild-type and ΔcheY inhibited repair of damage, but ΔmotB did not.
H. pylori chemotaxis genes have been identified as acid-induced genes and virulence factors that promote colonization initially and throughout infection, as well as promote H. pylori disease [4], [27], [29]. In H. pylori, however, the in vivo host-derived chemotactic factors regulating interactions of H. pylori with gastric tissue have been difficult to study. In this report, we used mouse models of ulceration and superficial epithelial damage to study the mechanisms employed by H. pylori to colonize the gastric niche. We demonstrate that H. pylori preferentially colonizes at an injury site compared to healthy regions of the stomach, with the highest efficiency of colonization requiring contributions of the chemotaxis system and flagellar motor activity in sensing the location of damaged tissue, and migrating towards the damage.
In our experiments, H. pylori can colonize gastric tissue in the absence of chemotaxis or even bacterial motility. Consistent with previous results [2], [3], our results suggest that high bacterial inocula can overcome the need for chemotaxis and motility. This finding is likely due to the fact that at supra-physiologic concentrations, there is increased probability that H. pylori will locate the correct niche by chance. Thus Che− and Mot− bacteria can establish colonization by luckily landing in favorable growth conditions and then engaging the mechanisms for attachment and other virulence factors normally engaged downstream of the chemotaxis functions.
Our study makes the striking observations that not only can H. pylori slow the repair of macroscopic ulcers over a period of 30 Days, but H. pylori in the gastric lumen can find damaged tissue and slow repair in a matter of minutes. Many investigators have confirmed that H. pylori inhibits ulcer healing induced by acetic acid [30]–[32], and one interesting study showed that gastric ulcer relapse occurs rapidly following H. pylori infection [30]. Using Mongolian gerbils, these investigators demonstrated that 8 weeks post-ulcer induction, ulcers were healed. However inoculation with H. pylori at this time point lead to re-expression of gastric ulcer more rapidly compared to previously uninjured animals [30]. Although the H. pylori abundance at the ulcerated area was not specifically determined in this prior study, it can be speculated that H. pylori were able to sense damaged tissue and preferentially colonize this site for rapid re-ulceration. Results from our micro-lesion model do not yet allow us to determine what H. pylori factors slow gastric repair. However, in both the ulcer and micro-lesion models, the ability of the ΔcheY mutant to slow repair makes it clear that the chemosensing system is not necessary to compromise epithelial repair.
We observed a striking decrease in H,K-ATPase expression during gastric ulcer healing, consistent with previous reports [33], [34]. We extended these findings by demonstrating that H,K-ATPase expression remains suppressed in the regenerating epithelium even by 32 days after ulceration, similar to observations by Blom's group that gastric wound repair resulted in full re-epithelialization by one month but parietal cells only re-appeared at 3 months and the number of parietal cells in the healed gastric mucosa never reached pre-ulceration levels [33]. Since the regenerated epithelium is slow to restore a normal cell census of parietal and TFF2-containing mucous cells, it is clear that ulcer healing results in an altered gastric epithelium that will alter both local gastric function and sensitivity to further insult. Results suggest that H. pylori may further delay the restoration of a normal epithelium in part by further delaying the epithelial repair process.
It is not yet known what signals drive H. pylori toward the damaged tissue. Changes in luminal pH in the microenvironment near the gastric surface may be a signal used by H. pylori chemotaxis to colonize this niche. We have shown that the pH increases in the luminal juxtamucosal space near the microlesions induced by two-photon laser damage [11], [12]. Others have used pH electrodes to show pH increases over sites of pervasive superficial gastric damage [13]. In preliminary studies with a pH 3 luminal perfusate, we observed using fluorescent pH indicators that luminal pH adjacent to the ulcer tissue was significantly elevated compared with intact tissue (Figure S9). H. pylori has a well-established pH-tactic response. It is reported that most H. pylori are free-swimming in the mucus gel layer within ∼30 µm of the epithelial surface [7], [35], [36]. H. pylori furthermore are spatially oriented within the mucus layer based on pH gradients [7]. Disruption of these gradients, by equalizing luminal and surface pH, causes H. pylori to lose their direction of motion, and spread over the entire mucus layer [37]. Specifically, H. pylori are repelled by acidic regions in vitro, and this behavior is eliminated by deletion of one of the chemotaxis receptor genes, tlpB, as well as the general chemotaxis pathway [8], [15]. In this study, we report that ΔcheY behavior is also different from wild-type H. pylori in a low pH environment. Thus one might predict that H. pylori's response to migrate away from acidic conditions would drive it toward the more neutral conditions overlying the damaged tissue.
Hydrogen ion concentration is only one of numerous gradients that could be act as chemorepellents or chemoattractants after gastric damage. Our 2-photon experiments determined that H. pylori responds to injury within minutes, and therefore it is unlikely that any signals result from inflammatory products. Instead, they are likely to arise by release promptly from epithelial damage. There are many potential candidates for H. pylori chemoattractants, including pH (CO2, HCO3−, H+), serum factors, urea, and ions (Ca, Mg, Fe, Zn etc) [4], although their presence at the injury site is not yet known. Several of these are reported chemoattractants for H. pylori in vitro, but not yet established in vivo [4], [20]. Based on the ability to rapidly evaluate H. pylori accumulation at the site of injury, further studies can now determine the chemoattractant(s) in the injury environment. Since this accumulation also slows gastric repair within minutes, future work also offers an opportunity to explore the mechanism and genes allowing H. pylori to initiate its earliest pathophysiological effects.
Several pathogens including H. pylori require chemotaxis for colonization, but to our knowledge, this report is the first to show bacterial chemotaxis aiding colonization of damaged tissue. However, Amieva's group has shown that in the intestine, Listeria monocytogenes preferentially adheres at sites of epithelial cell shedding [38]; and this may be considered a physiologic correlate of the microscopic epithelial discontinuity that we have produced experimentally. Supporting this conjecture, the same group has also shown that H. pylori can form microcolonies over intercellular junctions of cultured non-gastric cells [39]. Combined, the evidence suggests that compromises to the epithelial barrier can be an invitation to bacteria. There are several examples of chemotaxis directing pathogens to specific niches in vivo. For example, H. pylori requires chemotaxis for optimal initial colonization of the corpus and antrum, but after months of infection chemotaxis is only required to promote stable colonization and bacterial proliferation in the gastric glands of the antrum [3], [27]. In contrast, Vibrio cholerae uses chemotaxis to restrict its colonization to the jejunum and ileum. Che− V. cholerae are found in elevated numbers compared to wild type throughout the gastrointestinal tract [40]. As documented in this work, bacteria are exquisite biosensors of their environment. As we continue to study the role of chemotaxis in vivo, we will learn about the variety and importance of these niches for promoting bacterial colonization and disease pathogenesis.
All mice experiments were conducted according to both Animal Welfare Act Regulations and Public Health Service Policy on Humane Care and Use of Laboratory Animals. Mice were maintained in an AAALAC approved facility and all animal studies followed protocol 04-03-08-01 that was approved by the Institutional Animal Care and Use Committee of the University of Cincinnati (Cincinnati, OH).
Experiments used C57BL/6J mice (Jackson lab, Bar Harbor, ME). Animals were used for experiments at 3–6 month of age, were fed a standard rodent chow diet, and had free access to water.
Gastric ulcers were produced by acetic acid, according to a previously described method with slight modification [24]. In brief, under isoflurane anesthesia, the abdomen was incised and the intact stomach exposed. A microcapillary tube (0.7 mm in diameter: Drummond Scientific Co. Broomall, PA) filled with acetic acid (99%) was placed in contact with the exterior surface of the stomach corpus region and left in place for 25 sec. Buprenorphine hydrochloride (0.75 mg/kg i.p., Buprenex, Rechkitt Benckiser Pharmaceuticals Inc., Richmond, VA) was given as preemptive analgesia. After the acid was removed, the treated exterior of the stomach was wiped with gauze, the abdomen was closed, and the animals were routinely maintained with food and tap water. Using this procedure, no incision was made to the stomach as part of ulcer induction.
To examine ulcer healing, animals were sacrificed on day 3, 9, or 32 after ulceration, and the stomach was removed and opened along the greater curvature. The area (mm2) of ulceration was measured by digital caliper (Mitutoyo, Kanagawa, Japan). Since deep, well-defined ulcers were consistently observed 2 days following acid application [24], [41], H. pylori were gavaged at day 2 after the ulceration. Control uninfected animals received Brucella broth vehicle.
The surgical preparation of animals for in vivo gastric microscopic damage study has been described [11], [12], [42]. Briefly, mice were anesthetized with inactin (10 mg/kg i.p., Sigma, St. Louis, MO) and ketamine (50 mg/kg i.p., Phoenix, St. Joseph, MO, USA), then the stomachs of anesthetized mice were exteriorized and everted to expose the gastric mucosa. The mouse was placed in prone position such that a portion of the exposed mucosa protruded into a perfusion chamber on the stage of an inverted confocal/two-photon microscope (Zeiss LSM 510 NLO; Carl Zeiss, Jena, Germany), with the microscope stage enclosed and heated to keep the animal's body temperature at ∼37°C. The mucosal surface was exposed to fluorescently labeled H. pylori.
H. pylori strain SS1 and isogenic mutants [2]–[4] were grown on Columbia blood agar plates (Remel, Lenexa, KS, USA) containing 5% defibrinated horse blood (Colorado Serum, Denver, CO, USA), 0.2% β-cyclodextrin (Sigma), 50 µg/ml cyclohexamide (Sigma), 5 µg/ml vancomycin (Sigma), 10 µg/ml trimethoprim (Sigma) and 5 µg/ml chloramphenicol (only for ΔcheY mutant, Sigma) for 4 days. Colonies from these plates tested positive for urease (BD Diagnostic Systems, Franklin Lakes, NJ, USA), catalase (using 3% H2O2) and oxidase (DrySlide, BD Diagnostic Systems). Bacteria harvested from the plate were grown in Brucella broth (BD Diagnostic Systems) supplemented with 10% fetal bovine serum and CO2 gas pack (BD Diagnostic Systems) in a humidified microaerophilic chamber (BBL Gas System, with CampyPak Plus packs, BD Microbiology, Sparks, MD, USA) in an incubator at 37°C for 16 to 18 hr without shaking. Bacteria were collected by centrifuge at 2000 rpm for 5 min, and resuspended in Brucella broth without serum. H. pylori were diluted in 50% glycerol (1∶100 dilution) and counted in haemocytometer. Each mouse received 200 µl Brucella broth containing 103 to 108 H. pylori bacteria. Control mice (uninfected group) received 200 µl of Brucella broth alone.
For in vivo gastric microscopic damage study, H. pylori was labeled with one of the following fluorescent dyes (Invitrogen, Carlsbad, CA, USA): cell tracker green CMFDA, cell tracker orange CMRA or 5-(and -6)carboxyfluorescein diacetate, succinimidyl ester (5(6)- CFDA, SE: CFSE)-mixed isomers. Briefly, fluorescent dye (5 µM) was added into Brucella broth containing H. pylori for 15 min at 37°C, then washed with Brucella broth twice using 0.22 µm filter. H. pylori were adjusted to 108/ml for usage.
The wet weight was measured of small regions of mouse gastric tissue collected from intact (right side of gastric corpus in Figure 1D) and ulcerated (left side of gastric corpus in Figure 1D) tissue. Tissue was homogenized by Tissue Tearor (Model: 985370-395, BioSpec Products) in 1 ml saline and 1–10 dilutions were spread on a Columbia blood agar plate containing 5% defibrinated horse blood, 50 µg/ml cyclohexamide, 5 µg/ml vancomycin, 10 µg/ml trimethoprim (and 5 µg/ml chloramphenicol only for CheY mutant). Plates were incubated for 5–7 days at 37°C in a humidified microaerophilic chamber with CO2 gas pack. Colonies were counted and data normalized using the tissue weight and expressed in CFU/g tissue.
Quantitative analysis of bacterial load was performed by qRT-PCR using specific primers targeting different gut bacterial groups. Total RNA was isolated from either the intact or ulcerated area of stomach using TRI Reagent (Molecular Research Center) and single-stranded cDNA was synthesized by iScript cDNA synthesis kit (Bio-Rad, Hercules, CA, USA). Quantitative real-time PCR was performed using SYBR Green PCR Master Mix (Applied Biosystems) with following specific primers: 16S rRNA (Forward: 5′-ACTCCTACGGGAGGCAGCAGT-3′ and Reverse: 5′- ATTACCGCGGCTGCTGGC-3′) [43], Lactobacillus (Forward: 5′-GGAAACAGATGCTAATACCG-3′ and Reverse: 5′-CACCGCTACACATGGAG-3′) [44], SSA specific to H. pylori (Forward: 5′-TGGCGTGCTCTATTGACAGCGAGC-3′ and Reverse: 5′-CCTGCTGGGCATACTTCACCATG-3′), H,K-ATPase (Forward: 5′-AGATGGTGTTTGCTCGAACC-3′ and Reverse: 5′-TCCAGCAAGATCATGTCAGC-3′), TFF2 (Forward: 5′-GCAGTGCTTTGATCTTGGATGC-3′ and Reverse: 5′-TCAGGTTGGAAAAGCAGCAGTT-3′), or GAPDH (Forward: 5′-AACGACCCCTTCATTGAC-3′ and Reverse: 5′-TCCACGACATACTCAGCAC-3′). Each PCR amplification was performed in duplicate wells in a StepOnePlus Real-Time PCR System (Applied Biosystems).
Mouse intact or ulcerated stomach tissue was embedded in OCT and frozen and stored at −80°C. Serial cryosections (10 µm) were prepared by cryostat. The section was stained with hematoxylin & eosin to localize and evaluate damage severity. The adjacent sections including damaged area were then stained using anitbodies against H. pylori, TFF2 or H+/K+-ATPase. For immunostaining, sections were fixed with 4% paraformaldehyde followed by incubated with goat serum (4%) for 60 min. Sections were then incubated with primary antibodies indicated below for 60 min in room temperature. Primary antibodies to H. pylori (Rabbit polyclonal 1∶ 1000, Abcam) and H+/K+-ATPase alpha subunit (Mouse monoclonal 1∶ 1000, Thermo Science) or TFF2 (Rabbit polyclonal, 1∶200) were used. Secondary antibodies (Alexa 633-labeled goat anti-mouse IgGγ1: 1∶1000, and Alexa 488-labeled goat anti-rabbit IgG, 1∶1000 for H. pylori, 1∶ 400 for TFF2, Invitrogen, Carlsbad, CA, USA) were incubated for 60 min in room temperature. Nuclear DNA staining was performed by incubation with Hoechst 33342 (Invitrogen) at 1 µg/ml for 1 min.
H. pylori labeled with CMFDA or CFDA-SE (Invitrogen) fluorescence was imaged at 500–530 nm in response to 488 nm excitation, while CMRA was imaged at 565–615 nm in response to 560 nm excitation. H. pylori was loaded into uncoated μ-Slide chamber (ibidi) and monitored on the inverted fast scan confocal microscope (Zeiss LSM 7 LIVE; Carl Zeiss), with the microscope stage enclosed and heated at ∼37°C. To analyze bacterial swimming, the program Image J with manual tracking plugin was used. The number of stopping was counted from 4 sec motility traces and data showed as number of stopping per second.
The method of inducing, tracking, and quantifying microscopic photodamage in the gastric surface epithelium has been described previously [11], [12], [28], [42]. Two-photon imaging of tissue NAD(P)H autofluorescence (Ti-Sa excitation 730 nm, emission 435–485 nm) was collected simultaneously with a confocal reflectance image (reflecting 730 nm light to show cell/tissue structure). In some experiments, CMFDA-labeled H. pylori was applied to chamber and monitored in parallel (Ex: 488 nm, Em: 500–550 nm). After collecting a set of control images using minimal laser power, a small rectangle region (≈200 µm2) of gastric surface epithelium in the corpus was repetitively scanned at high Ti-Sa laser power (350 mW average) for 150 iterations (requiring 5–15 sec). Photobleaching of endogenous NAD(P)H induced epithelial cell death and extrusion followed normally by epithelial restitution via epithelial cell migration in ∼15 min, as recorded by two-photon/confocal time course imaging.
Damage (and repair) was quantified from the time course of images as described [12], [28] using Metamorph software (ver. 6.3, Molecular Devices, Downington, PA, USA). Briefly, the damaged area was measured as the region of cellular loss of autofluorescence NAD(P)H, using confocal reflectance images to positively identify the location of cellular structures. In each experiment, we determined the time point displaying maximal damage area and estimated rates of epithelial restitution starting from this time with a single exponential curve fit to the changing size of damaged area over time [12]. Best fit values of the rate constant (κ) were used as estimates of the rate of restitution (estimate of the fractional recovery per time, in units of sec−1). In some experiments, intensity of H. pylori fluorescence at the gastric surface was measured. The damage-repair cycle was measured independently several times per animal in different locations of the corpus, and outcomes from at least 4 animals were compiled for each experimental protocol.
All values are reported from representative experiments as the mean ± standard error of the mean (SEM) from multiple experiments. The number of repetitions is reported as the number of independent damage-repair cycles (n) analyzed, and the number of different animals used is also indicated. All results were reproduced in at least 4 animals. Statistical significance was determined using unpaired Student's T-test, or one-way ANOVA with Dunnett's multiple comparison post-hoc test. A p value of <0.05 was considered significant.
|
10.1371/journal.pgen.1006463 | The Mitochondrial m-AAA Protease Prevents Demyelination and Hair Greying | The m-AAA protease preserves proteostasis of the inner mitochondrial membrane. It ensures a functional respiratory chain, by controlling the turnover of respiratory complex subunits and allowing mitochondrial translation, but other functions in mitochondria are conceivable. Mutations in genes encoding subunits of the m-AAA protease have been linked to various neurodegenerative diseases in humans, such as hereditary spastic paraplegia and spinocerebellar ataxia. While essential functions of the m-AAA protease for neuronal survival have been established, its role in adult glial cells remains enigmatic. Here, we show that deletion of the highly expressed subunit AFG3L2 in mature mouse oligodendrocytes provokes early-on mitochondrial fragmentation and swelling, as previously shown in neurons, but causes only late-onset motor defects and myelin abnormalities. In contrast, total ablation of the m-AAA protease, by deleting both Afg3l2 and its paralogue Afg3l1, triggers progressive motor dysfunction and demyelination, owing to rapid oligodendrocyte cell death. Surprisingly, the mice showed premature hair greying, caused by progressive loss of melanoblasts that share a common developmental origin with Schwann cells and are targeted in our experiments. Thus, while both neurons and glial cells are dependant on the m-AAA protease for survival in vivo, complete ablation of the complex is necessary to trigger death of oligodendrocytes, hinting to cell-autonomous thresholds of vulnerability to m-AAA protease deficiency.
| Oligodendrocytes are cells of the central nervous system that produce the myelin sheath. Myelin production is extremely costly from the energetic point of view, and oligodendrocytes that are synthesizing myelin are particularly susceptible to mitochondrial dysfunction. However, the function of mitochondria in mature oligodendrocytes, after myelination is completed, has been poorly explored using genetic models. Here, we have generated and characterized mouse models expressing different levels of the m-AAA protease, a proteolytic complex preserving proteostasis of the inner mitochondrial membrane and respiratory activity. We show that oligodendrocytes are capable to cope with reduced levels of the complex, but undergo rapid death upon complete ablation of the m-AAA protease. Thus, the m-AAA protease is essential for cell survival, but oligodendrocytes are less vulnerable than neurons to a deficiency of the complex.
| Oligodendrocytes are glial cells of the central nervous system (CNS) that produce myelin to enhance conduction velocity. Oligodendrocytes utilize high amount of energy to synthesize proteins and lipids to build up myelin [1] and depend on mitochondrial respiration heavily during differentiation and myelination, when they are dramatically susceptible to ischemia, energy deprivation, and oxidative stress [2, 3]. It has been hypothesized that post-myelination oligodendrocytes can undergo a metabolic switch to glycolysis, and provide metabolic support to axons, by supplying lactate as an energy source [2, 4]. In support of this hypothesis, oligodendrocyte-specific deletion of an essential assembly factor for complex IV, Cox10, did not lead to axonal degeneration or demyelination [2], strongly suggesting that these cells can survive a respiratory chain deficiency. Moreover, recent data suggest that oligodendrocyte mitochondria may be involved in specialized functions relevant for myelin maintenance, such as lipid synthesis, or fatty acid oxidation, rather than in ATP production [5].
The m-AAA protease is a large proteolytic complex in the inner mitochondrial membrane endowed with crucial and pleiotropic roles in mitochondria. It regulates the turnover of respiratory chain subunits [6–8], controls ribosome assembly and thereby mitochondrial translation [9, 10], and affects mitochondrial dynamics [11]. In humans, the m-AAA protease is composed of two subunits, paraplegin and AFG3L2, which form either homo-oligomeric (AFG3L2 alone) or hetero-oligomeric (AFG3L2 and paraplegin) hexameric functional complexes [12]. The mouse genome contains a third gene, encoding a functional m-AAA protease subunit, Afg3l1, which can form either homo-oligomers or hetero-oligomers with AFG3L2 or paraplegin [12].
The discovery that both paraplegin and AFG3L2 are implicated in human neurodegenerative diseases has sparked increasing interest in the m-AAA protease. Recessive mutations in SPG7, encoding paraplegin, lead to hereditary spastic paraplegia (HSP) [13], a neurodegenerative disease affecting the long corticospinal motor axons, while dominant mutations in AFG3L2 cause spinocerebellar ataxia type 28 (SCA28) [14], associated with atrophy of the cerebellum. Moreover, a severe phenotype combining features of spastic paraplegia and ataxia associated with myoclonic epilepsy (SPAX5) has been linked to a homozygous mutation in AFG3L2 [15].
A plethora of dysfunctional pathways have been unravelled in cells when the m-AAA protease is depleted, including reduced assembly of respiratory complexes [9, 16, 17], COX deficiency, impaired mitochondrial translation, fragmentation of the mitochondrial network [9], disturbance of mitochondrial anterograde transport [18], and calcium dysregulation [19, 20]. Neurons are extremely susceptible to decreased levels of the m-AAA protease, and cannot survive Afg3l2 deficiency [9]. The role of the m-AAA protease in glial cells is so far unknown.
Here, we used an inducible Plp1-CreERT transgenic mouse line to delete Afg3l2 in a wild-type or Afg3l1-null background in adult oligodendrocytes. We found that AFG3L2 deficiency was tolerated by oligodendrocytes for a long time, but ultimately led to late-onset myelin abnormalities and axonal degeneration in the spinal cord. In contrast, deletion of both Afg3l2 and Afg3l1, which completely abolishes the m-AAA protease, caused rapid cell death of targeted cells. Our study unravels a crucial role of the m-AAA protease in protection against cell death, independent from the metabolic profile of the cell, and demonstrates that different thresholds of m-AAA protease activity are required in neurons and glial cells.
AFG3L2 is highly expressed in the brain [12], however its abundance in neuronal versus glial cells is unknown. We investigated the expression of subunits of the murine m-AAA protease in lysates from enriched neuronal cultures, purified astrocytes, and purified late oligodendrocyte progenitors by immunoblotting. Notably, we did not observe remarkable differences in the levels of AFG3L2 or SPG7 (paraplegin) in neurons versus glial cells (Fig 1A). Minor differences in the expression levels correlated with the abundance of other mitochondrial markers such as SDHA (Fig 1A). Our results are consistent with published quantitative proteomic data obtained from acutely isolated cell population in the brain [21]. These data raise the question of the functional role of the m-AAA protease in glial cells.
To target oligodendrocytes, we utilized a well-established tamoxifen-inducible Plp1-CreERT transgenic line [22]. Sensitivity and specificity of the Cre expression was confirmed by crossing Plp1-CreERTwt/tg mice with a reporter transgenic line expressing a mitochondrially targeted YFP (mt-YFP) upon Cre recombination (ROSA26+/SmY)[23]. Tamoxifen was injected intraperitoneally for five consecutive days at P29, a time point used in a previous study to induce Cox10 deletion using the same promoter [2], and the corpus callosum analysed at P36 (S1A Fig). mt-YFP expression was largely restricted to cells expressing both or either of the oligodendrocyte markers APC and Olig2 (S1B and S1C Fig). We determined that approximately 50% of oligodendrocytes (positives for either or both APC and Olig2) in the corpus callosum were targeted (S1D Fig).
To induce Afg3l2 deletion in oligodendrocytes, we crossed Plp1-CreERT mice with mice carrying a floxed allele of Afg3l2 [9], and used the same protocol described above. We analysed Afg3l2fl/fl Plp1-Cretg/wt mice (referred to as L2KO) and compared them with Afg3l2fl/fl Plp1-Crewt/wt mice (L2fl/fl), similarly treated with tamoxifen. L2KO mice did not show any apparent phenotype in the cage or weight loss compared to controls up to 90 weeks of age (S2A and S2B Fig). However, at this old age, they displayed a mild but significant impairment in motor performance on an accelerating rotarod test (Fig 1B). We then carefully analysed the brain and the spinal cord of L2KO mice to detect any sign of late-onset pathology. No obvious demyelination was detected up to 90 weeks of age, when we only observed the appearance of few bigger APC+ cells in the corpus callosum (S2C Fig). Semithin sections of the spinal cord revealed no alterations at 56 weeks (S2D Fig), but abnormal myelin profiles, characterized by myelin thickening and infoldings, and myelin whorls, indicative of axonal degeneration were visible at 90 weeks (Fig 1C and 1D). Ultrastructural analysis of the spinal cord white matter disclosed a few axons characterized by thin myelin, and others showing adaxonal myelin detachment and vacuolization already at 56 weeks. These abnormalities were more prominent at 86 weeks, when degenerating axons surrounded by damaged myelin or containing accumulation of material were present (Fig 1E). Oligodendrocytes contained enlarged mitochondria with disrupted cristae (Fig 1E and S2E Fig), closely resembling those previously described in Afg3l2-deficient neurons [9, 17].
Fragmentation of the mitochondrial network occurs at early time points after deletion of Afg3l2 in neurons [9], and is caused by activation of the stress protease OMA1, which in turn cleaves the dynamin-like GTPase OPA1, leading to impaired mitochondrial fusion [11]. The functional role of this fragmentation in neurons is unclear, since they die shortly after showing this phenotype. To visualize mitochondrial morphology in targeted oligodendrocytes, we further crossed mice with the mt-YFP (ROSA26+/SmY) reporter line [23]. To exclude a toxic effect caused by mt-YFP and/or Cre expression, we used as controls mice haploinsufficient for Afg3l2. At 8 weeks of age, oligodendrocytes in the corpus callosum of Afg3l2fl/+ Plp1-Crewt/tg ROSA26+/SmY mice had a tubular mitochondrial network, while in absence of Afg3l2 (genotype: Afg3l2fl/fl Plp1-Cretg/wt ROSA26+/SmY) mitochondria appeared swollen and fragmented (Fig 1F and 1G), indicating that the residual m-AAA protease is not sufficient to prevent this stress response.
The mt-YFP reporter also allowed us to trace the fate of targeted oligodendrocytes. In the corpus callosum, the total number of targeted mt-YFP+ cells and of APC+ oligodendrocytes was not significantly changed in absence of AFG3L2 compared to control mice at 56 weeks (S2F Fig), in agreement with the lack of overt demyelination. However, while in control mice almost all mt-YFP+ cells were also APC+, L2KO mice displayed a significantly increased number of mt-YFP+APC- cells (S2G Fig).
Thus, even though lack of Afg3l2 triggered early-onset pronounced mitochondrial morphology defects, oligodendrocytes survived for long time, and myelin alterations occurred only at very old age.
The murine m-AAA protease subunit AFG3L1 is highly expressed in liver, kidney and heart, but is hardly detectable in the brain (Fig 2A), consistent with previous data [24]. To rule out a major role of Afg3l1 in the mouse nervous system, we generated a full body knock-out of Afg3l1, by deleting exons 2 and 3 (S3A Fig). Analysis of Afg3l1 transcript levels from liver of 5-week-old mice showed the presence of residual mRNAs after splicing from exon 1 to either exon 4 or exon 5 (S3B Fig). While exon 1–4 splicing gives rise to an out-of-frame transcript, splicing from exon 1 to 5 leads to an in-frame transcript potentially encoding a shorter protein that is devoid of large part of the mitochondrial targeting sequence. However, immunoblotting of liver mitochondria showed no mature AFG3L1 protein (Fig 2B). Most importantly, blue-native PAGE demonstrated lack of assembled AFG3L1 in high-molecular weight m-AAA complexes (Fig 2C), confirming that Afg3l1-/- mice are bona-fide knock-out. AFG3L2 and paraplegin abundance, as well as OPA1 processing, were not affected by lack of AFG3L1 (Fig 2B). Afg3l1-deficient mice were born at the expected Mendelian ratio, showed a comparable growth curve to control littermates (Fig 2D), were fertile, and did not show any evident phenotype up to 78 weeks of age. We carefully examined the brain and the spinal cord of Afg3l1+/- and Afg3l1-/- mice and detected neither obvious myelination defects nor axonal degeneration at least till 1 year of age (Fig 2F and 2G). Ultrastructural analysis of mitochondria in the spinal cord did not revealed morphological abnormalities (Fig 2G). Thus, in contrast to Afg3l2, mouse Afg3l1 is dispensable in the central nervous system both in neurons and oligodendrocytes.
Paraplegin cannot form homo-oligomeric functional complexes [12], however in absence of AFG3L2 it may assemble together with AFG3L1 and constitute a functional m-AAA protease. Therefore, total ablation of m-AAA complexes can be achieved in oligodendrocytes by Plp1 promoter-driven recombination of Afg3l2 in a null Afg3l1 background. Double knock-out animals (referred to as DKO; genotype Afg3l1-/- Afg3l2fl/fl Plp1-Cretg/wt) were compared to control Afg3l1-/- littermates (CTRL; genotype Afg3l1-/- Afg3l2fl/fl Plp1-Crewt/wt). Cre expression was induced by tamoxifen at 4 weeks, as previously described. Starting from 8 weeks of age, DKO mice failed to gain weight compared to CTRL mice (Fig 3A). This difference persisted even after putting food pellets directly in the cage (Fig 3B). Furthermore, DKO mice had decreased fat mass compared to CTRL mice (Fig 3C). At about 13 weeks, DKO mice started to show signs of motor dysfunction. In a rotarod test DKO mice of 11–13 weeks of age spent less time on the rotating rod compared to CTRL mice (Fig 3D). Moreover, the number of foot slips while walking on a 1 cm-wide beam was significantly increased at 13 weeks, and became dramatically higher at 28 weeks (Fig 3E, S1 and S2 Movies). Surprisingly, DKO mice developed a progressive pattern of hair greying starting ventrally close to the forelimbs at 10 weeks, resulting in a grey belly at 17 weeks, and finally extending to the dorsal skin at 28 weeks of age (Fig 3F).
Thus, concomitant loss of Afg3l1 in oligodendrocytes strongly exacerbates the phenotypes observed in absence of Afg3l2. The DKO mice therefore serve as valuable model to examine the significance of a complete m-AAA protease deficiency in myelinating cells.
To shed light on the neurological phenotype of DKO mice, we examined brains and spinal cords from DKO and CTRL mice at different time points. At 4 weeks, before tamoxifen injection, the degree of myelination in CTRL and DKO mice was comparable both in the lumbar spinal cord and in the brain (Fig 4A and S4A Fig). However, progressive demyelination was detected in the lumbar spinal cord of DKO, leading to the appearance of demyelinated and degenerating axons as well as dark cells at 28 weeks in the antero-lateral funiculus (Fig 4A–4C). Ultrastructural analysis of the white matter of the spinal cord confirmed progressive demyelination with some axons showing adaxonal detachment of the myelin at 13 weeks, and pronounced signs of demyelination already at 18 weeks (Fig 4D). Signs of secondary axonal degeneration, with accumulation of organelles and material in axons, were visible at 28 weeks (Fig 4D). We identified several axons surrounded by thin myelin, and by oligodendrocytes with dark cytoplasm containing heterogeneous membranous material, probably of lysosomal origin (Fig 4D). These cells have the characteristic of the dark oligodendrocytes, previously proposed to represent mature oligodendrocytes [25, 26]. The g ratio, expressing the ratio between the diameter of the inner axon and the total fiber diameter, was significantly increased at this age (Fig 4E). In agreement with these data, Gallyas’ silver staining of myelinated tracts in the brain showed prominent loss of white matter in the corpus callosum, the internal capsule, and the cerebellum at 28 weeks of age in the DKO mice (S4B Fig).
Progressive loss of myelin was confirmed by western blot analysis of myelin proteins in spinal cord and brain lysates at different time points (S4C and S4D Fig). Concomitantly, we observed upregulation of GFAP, indicating reactive astrogliosis (S4C, S4D and S5 Figs). At 28 weeks, activated microglia cells, which can be recognized by a change in morphology from small cells with slender processes to larger amoeboid-like cells with thick processes, were also detected in the corpus callosum, emphasizing the presence of a neuroinflammatory response (S5 Fig).
Since neuroinflammation is a very sensitive read-out of cell damage, we investigated whether cell demise underlies the phenotype. To this end, we crossed DKO mice with the mt-YFP reporter line to visualize targeted oligodendrocytes in vivo. Strikingly, mt-YFP+ cells were rapidly and progressively lost after 6 weeks, and only a few targeted cells remained at 28 weeks (Fig 5A and 5B). Initially, loss of targeted oligodendrocytes was paralleled by a decrease in APC+ cells that was especially evident at 10 weeks both in the corpus callosum and in the spinal cord (Fig 5A and 5C, S6A and S6B Fig). At 10 weeks, the percentage of mt-YFP+ cells in the corpus callosum that were also APC+ was significantly reduced in DKO mice in comparison with the control line carrying only the mt-YFP reporter, while the percentage of APC- Olig2- targeted cells was significantly increased (Fig 5D). This result is reminiscent of what observed in L2KO mice (S2G Fig).
Surprisingly, the number of mature oligodendrocytes was recovered at 28 weeks in the DKO. This might be explained by the compensatory proliferation and differentiation of untargeted oligodendrocytes that still express Afg3l2. Consistently, the size of APC+ cells in the DKO mice was increased at 28 weeks (Fig 5A, S6C and S6D Fig), and the enlarged APC+ cells did not colocalize with mt-YFP+ cells (Fig 5A). We found that the enlarged APC+ cells were in fact intensively stained for Olig2 (S6D Fig), a transcription factor expressed at higher levels in migrating and remyelinating oligodendrocytes [27–29]. However, quantification of total Olig2+ cells (both intensively and less intensively stained) indicated no statistical difference in the corpus callosum of the CTRL and DKO mice (S6E and S6F Fig).
When we monitored mitochondrial morphology, taking advantage of the expression of the mt-YFP reporter in targeted oligodendrocytes, we found abnormal swollen mitochondria already at 6 weeks of age in the DKO mice (Fig 6A). COX1 staining was preserved at this time, but was lost at 8 weeks in targeted oligodendrocytes of the DKO, indicating impairment of mitochondrial respiratory function (Fig 6A). Moreover, cytochrome c was undetectable in several swollen mitochondria in DKO oligodendrocytes at 8 weeks (Fig 6B). We found several oligodendrocytes showing features of dark cell death (Fig 6C), a caspase-independent form of death, characterized by strong cytoplasmic condensation, chromatin clumping, ruffling of the cell membrane, but no blebbing of the nucleus or plasma membrane [30]. Consistently, in situ TUNEL assay showed only a few apoptotic cells in DKO mice at 7 weeks of age (quantification in the corpus callosum: 4.13 ± 0.533 in the CTRL mice, 9.75 ± 0.85 in the DKO mice, n = 4 mice per genotype). In summary, these results suggest that the complete loss of m-AAA protease causes major mitochondrial dysfunction and death of mature oligodendrocytes followed by compensatory repopulation by untargeted oligodendrocytes.
One surprising finding was that the DKO mice showed progressive hair greying (Fig 3F). During embryonic development, the Plp1 promoter has been shown to target not only oligodendrocytes, but also Schwann cell (SC) precursors (SCPs), bipotential progenitors of both SCs and melanoblasts [31, 32]. The role of SCPs in the formation of new melanocytes in the adult, and during age-related hair greying remains unknown. Moreover, there is evidence for weak expression of the Plp1 promoter in embryonic melanocytes [33–35], but its activity in adult melanoblasts and melanocytes is unclear. The observed greying of DKO mice thus raised the question whether melanoblasts were targeted and SCPs were affected.
To this end, we performed fate-mapping experiments using the mt-YFP reporter line to establish which cells are targeted in our experiments. We administered tamoxifen at P29 for 5 days, and then collected the ventral or dorsal skin of wild-type mice at P36. This time corresponds to the growth phase called anagen of the second hair cycle in the mouse. During hair follicle (HF) growth, unpigmented melanoblasts differentiate from melanocyte stem cells residing in or close to the hair follicle bulge area and migrate within the outer root sheath of the HF towards the hair matrix where they differentiate into fully mature pigmented melanocytes [36]. We identified mt-YFP+ signal in SCs in the subcutaneous nerve plexus or in the nerves surrounding the HFs, in the bulge area containing melanocyte stem cells, in melanoblasts located in the outer root sheath of the HFs, in pigmented bulbar melanocytes (S7 Fig). Remarkably, targeting was more efficient in the ventral than in the dorsal skin, thus providing a potential explanation for the ventral to dorsal progression of hair greying (S7 Fig).
We then investigated the fate of targeted cells in the skin of DKO mice. At 10 weeks of age, we observed a strong reduction of mt-YFP+ melanoblasts in the outer root sheath in DKO mice (Fig 7A). Moreover, non-myelinating SCs in the subcutaneous nerve plexus also showed a significant decrease of mt-YFP signal (Fig 7A). At 28 weeks, although general skin structure was preserved in DKO mice, there was reduced fat deposited in the dermis (Fig 7B and 7C), consistent with the observed general reduction of fat mass in these mice (Fig 3C). Since HF cycling is largely non-synchronized in the mouse at 28 weeks, we shaved the back of the mice, selected pigmented areas of the skin (containing HFs in anagen) for biopsy, and stained sections with antibodies against c-KIT, a marker of melanocytes and melanoblasts. Strikingly, DKO mice showed a significant reduction of pigmented HFs (79.7% ± 2.9 in CTRL versus 28.0% ± 12.3 in DKO mice, n = 3, Student’s t-test, p < 0.05) and c-KIT-positive melanoblasts and melanocytes (Fig 7D). We conclude that hair greying is caused by progressive loss of melanocyte stem cells and melanoblasts that are targeted in our experiments.
Given the fact that DKO mice showed a pathological phenotype in unmyelinated cutaneous nerves, we further examined if peripheral nerves were affected. The Plp1 promoter is known to be expressed at low level in adult SCs [22]. Consistently, we observed scattered mt-YFP signal within the sciatic nerve of wild-type mt-YFPtg/wt mice at 10 weeks of age, and noticed a reduction of this signal in DKO mice (Fig 8A). Semithin sections of the sciatic nerve did not show a remarkable phenotype at 10 and 28 weeks (Fig 8B), probably because of the very low number of targeted cells. We therefore performed ultrastructural analysis and found clear signs of pathology affecting preferentially small calibre unmyelinated fibers (Fig 8C). These fibers are normally associated with non-myelinating SCs in the so-called Remak bundles that contain several axons wrapped by one individual SC. In a normal Remak bundle the cytoplasm of a SC separates individual axons. In the DKO several Remak bundles appeared affected with individual axons touching each other, and showing initial signs of axonal degeneration (Fig 8C). Some alterations were also observed in a few large calibre myelinated axons. In most cases, these were characterized by enlargement of the inner adaxonal tongue, which contained large vacuoles or other material. Similar changes have been previously observed in Cnp knock-out mice [37, 38]. These changes were more pronounced at 28 weeks, when also a few demyelinated axons were noted (Fig 8C).
Together, our data establish the vulnerability of both SCs and melanoblasts to loss of the m-AAA protease.
Although the m-AAA protease in the inner mitochondrial membrane is essential to preserve respiratory activity in neurons, nothing is known about cell autonomous requirements of this complex in glial cells in vivo. Here, we have generated mouse models expressing different levels of the m-AAA protease in adult myelinating cells. We found that these glial cells survive for long time with reduced levels of the m-AAA protease, but total absence of the m-AAA protease triggers rapid cell death.
A main conclusion of our study is that the threshold of m-AAA protease activity allowing survival of neurons and myelinating cells is remarkably different. This is in line with the fact that mutations in AFG3L2 or SPG7 lead to distinct neurodegenerative diseases, characterized by a pure neuronal and axonal phenotype, respectively. What underlies the different cellular vulnerability? We found no significant difference in the expression of AFG3L2 and paraplegin among astrocytes, oligodendrocytes, and neurons, excluding that different stoichiometry of the individual subunits of the m-AAA protease plays a crucial role. A possible explanation for our results is the different metabolic profile of adult oligodendrocytes and SCs that are able to survive using glycolysis alone when mitochondrial respiration is impaired. Consistently, oligodendrocyte-specific deletion of an essential assembly factor for complex IV, Cox10, using the same promoter and Cre induction paradigm as in this study, did not lead to axonal degeneration, demyelination, or cell death up to 14 months of age [2]. Moreover, deletion of Tfam in SCs caused conspicuous respiratory deficiency, but did not affect their survival [39].
At odds with the hypothesis that oligodendrocytes compensate metabolically for respiratory deficiencies, is the fact that the complete ablation of the m-AAA protease is incompatible with cell survival. All targeted cells in the DKO mice, oligodendrocytes, SCs, and melanoblasts, showed mitochondria with dramatically abnormal morphology, and died shortly after removal of the complex. Non-myelinating SCs were affected earlier and more prominently than myelinating SCs, in agreement with previous findings of a peculiar susceptibility of these cells to mitochondrial dysfunction [39]. Oligodendrocytes showed features of dark cell death, very similar to those observed in neurons haploinsufficient for Afg3l2 [40]. It is conceivable that the respiratory function of mitochondria in post-myelinating oligodendrocytes and SCs is more important than previously thought. Moreover, deficiencies of the m-AAA protease likely have more severe effects on the oxidative capacity of the organelles than deletion of Cox10 or Tfam. Indeed, Cox10 deletion results in isolated complex IV defect, and that depletion of mtDNA upon loss of Tfam occurs after a rather long time [41]. In contrast, it was sufficient to ablate Afg3l2 alone in oligodendrocytes to trigger mitochondrial fragmentation and swelling, a stress response to defective turnover of de novo synthesized inner membrane proteins [8, 42], which was not observed in Cox10-deficient oligodendrocytes [2]. Consistently, deletion of Afg3l2 in adult neurons provoked neuronal loss much earlier than observed when a similar strategy was applied to delete Cox10 or Tfam [39, 43], and deletion of Afg3l2 in oligodendrocytes leads to a late-onset myelin phenotype. In the future, it will be important to develop more genetic models lacking specific mitochondrial proteins involved in respiratory function to fully understand the relevance of oxidative phosphorylation for energy metabolism of adult oligodendrocytes.
An alternative, not mutually exclusive, explanation for the rapid cell demise induced by the lack of the m-AAA protease is the activation of a death pathway independent from energy deprivation. We recently found that loss of the m-AAA protease results in accumulation of constitutively active MCU-EMRE channels leading to mitochondrial Ca2+ overload, mitochondrial permeability transition pore opening and cell death [44]. Such a death pathway would be consistent with the swollen mitochondrial morphology, and the ultrastructural dark appearance of dying neurons and oligodendrocytes depleted of the m-AAA protease. Furthermore, the intrinsic vulnerability of neurons to Ca2+-dependent cell death may provide a rationale for their increased susceptibility to deficiency of the m-AAA protease compared to oligodendrocytes [45].
The fragmentation of the mitochondrial network triggered by Afg3l2 deletion in oligodendrocytes, as previously observed in neurons [9], raises the question whether this stress-mediated response has different outcomes in oxidative versus glycolytic cells. Emerging data suggest a possible relationship between mitochondrial network morphology and the metabolic capacity of cells [46, 47]. However, in oligodendrocytes, stress-induced mitochondrial fragmentation may be beneficial when transient, but become detrimental if persistent [42], thus contributing to the late-onset myelin abnormalities in mice carrying Afg3l2 deletion in oligodendrocytes.
In the central nervous system, our DKO model recapitulates features already observed in models of demyelination [48–50]. Loss of the myelin sheaths in DKO mice occurred a few weeks after oligodendrocyte cell death, consistent with previously described long-term stability of the myelin proteins and lipids [51–53], and was associated with a regenerative response of oligodendrocytes leading to some degree of myelin repair, further highlighting the reparative potential of adult oligodendrocytes. However, we cannot completely exclude that part of the phenotype is caused by impaired myelin formation, since myelination is not totally complete at P28 when we injected tamoxifen. A consistent finding, both in L2KO and DKO mice, was an increased percentage of targeted cells that were negative for both APC and Olig2. These cells may represent either oligodendrocytes that lose expression of these markers before dying, and/or oligodendrocyte precursors which fail to differentiate. In fact, experiments in wild-type mice using the mt-YFP reporter line showed that a small percentage of APC- Olig2- cells were targeted shortly after tamoxifen injection, but were hardly detectable at 10 weeks (compare S1C Fig and Fig 5D). Tailored experiments need to be performed in the future to address whether the m-AAA protease or a tubular mitochondrial network are required during oligodendrocyte differentiation.
Surprisingly, DKO mice lost weight and fat mass. Albeit we do not know at present the reason for this phenotype, we excluded that this is the result of neurological impairment, hampering to access food. It is conceivable that hypothalamic brain areas involved in feeding behavior may be affected by demyelination. Furthermore, we cannot rule out the possibility that the Plp1 promoter is expressed in other cell types than myelinating cells, contributing to this phenotype.
An unexpected phenotype in DKO mice was premature and progressive hair greying. Melanocytes and SCs both arise from the neural crest. A previous study has identified both in chick and mouse two distinct migratory pathways producing melanocyte stem cells during development. At E10-E11.5, melanoblasts delaminate from the murine neural tube and migrate dorsally between the dermomyotome and the epidermis to populate the skin. Later, at E12-E14 bipotential precursors of both melanoblasts and SC, the SCPs, leave the neural crest along a ventral migratory route along the nerves [31, 32]. According to these studies, at around E13, SCPs detaching from the nerve differentiate into melanoblasts and downregulate Plp1, while those that remain attached to the nerve acquire SC properties [31]. This second migratory wave would contribute to melanoblasts in the limbs, the belly, and the dorsal skin, and has been identified performing fate-mapping experiments using a Plp1-CreERT2 transgenic line [31]. In our hands, at 4 weeks the Plp1 promoter is active not only in the nerve, but also in cells that are located in the bulge area and in the outer root sheath of the HF, and therefore have lost contact with the nerve, suggesting that they are melanoblasts. Importantly even a few melanocytes were targeted, as others have observed at developmental stages [34]. We therefore conclude that the Plp1-CreERT line is not suitable to determine whether SCPs contribute to melanocytes in the adult mouse. Aging-associated hair greying has been linked to increased differentiation and/or loss of melanocyte stem cells [36]. As in our model stem cells are lost, without an initial increase in differentiated pigmented cells, the latter is the most likely mechanism, coupled with the concomitant death of targeted melanocytes. Premature hair greying has been previously reported in the mutator mouse, a genetic model of accelerated aging caused by expression of a proof-reading-defective mitochondrial Polg DNA polymerase [54]. The mechanism of hair greying in this model has not been investigated in detail, and more studies are needed to understand the role played by mitochondrial dysfunction or intrinsic pathway of apoptosis in aging-related hair greying.
Finally, we show here that constitutive deletion of Afg3l1 alone in the mouse does not lead to an evident neurological phenotype. This is consistent with the extremely low levels of expression in the brain. However, our study indicates that AFG3L1 can largely rescue deficiency of AFG3L2 in oligodendrocytes, by sustaining residual m-AAA activity likely in complex with the more abundant paraplegin subunit. Since Afg3l1 is a functional gene in the mouse but not in humans [55], examining the phenotypic consequences of mutations in Afg3l2 or Spg7 in an Afg3l1 null background now offers a more suitable model mimicking the human situation.
In summary, our data shed new light on functional requirements of the mitochondrial m-AAA protease in adult oligodendrocytes, help understanding cell-specificities in the context of the human pathologies, and provide insights in oligodendrocyte, SC, and melanoblast survival mechanisms.
All animal procedures were carried out in accordance with European (EU directive 86/609/EEC), national (TierSchG), and institutional guidelines and were approved by local authorities (Landesamt für Natur, Umwelt, und Verbraucherschutz Nordrhein-Westfalen, Germany; approval numbers 87–51.04.2010.A219 and 84–02.04.2015.A402). Plp1-CreERT mice [22] were purchased from Jackson Laboratory. Conditional Afg3l2fl/fl mice [9] and ROSA26+/SmY mice [23] were previously reported. Afg3l1-/- mice were commercially generated in C57BL/6N background at Taconic-Artemis. Plp1-CreERT mice were mated with ROSA26+/SmY mice to visualize mitochondria in oligodendrocytes. To obtain L2KO mice, Afg3l2fl/fl mice were crossed to Plp1-CreERT mice. As controls, we used Afg3l2fl/fl Cre-negative littermates. Mice of both genotypes were injected with tamoxifen as specified below. Afg3l1-/-Afg3l2fl/fl Plp1-Crewt/tg (DKO) mice were compared with Afg3l1-/-Afg3l2fl/flPlp1-Crewt/wt (CTRL) littermates injected with tamoxifen. To investigate mitochondrial morphology in L2KO mice or in DKO mice, when indicated, Afg3l2fl/flPlp1-Crewt/tg or Afg3l1-/-Afg3l2fl/flPlp1-Crewt/tg mice were crossed with ROSA26+/SmY mice. Tamoxifen (T5648, Sigma) was dissolved in a corn oil/ethanol (9:1) mixture at a final concentration of 10 mg/ml. 1 mg tamoxifen was administrated by intraperitoneal injection once a day for 5 consecutive days to 4-week-old (P28-30) mice. Unless specified, mice of either sex were used for experiments.
The rotarod apparatus (TSE systems) was used to test motor ability and coordination. Mice were placed on a rotating rod (accelerating model) and the latency time to fall was recorded for each mouse up to a maximum of 300 seconds. Three tests were performed for three consecutive days. Mice were allowed to rest for 15 minutes after each test. In the beam walking test, mice were trained to walk on a 90 cm long and 3 cm wide beam, elevated by 30 cm on a metal support, for three times for three consecutive days. The actual test was performed by allowing the mice to walk on a 1 cm wide beam on the third day. The performance was filmed and the number of foot slips was quantified. The lean and fat mass of mice was measured with Bruker Minispec Live Mice Analyzer (LF50H).
Mice were deeply anesthetized with xylazine/ketamine (10 mg/100 mg per kg of body weight) and perfused transcardially with PBS and 4% paraformaldehyde (PFA). Brain, spinal cord, and the peripheral nerves were then dissected and postfixed in 4% PFA for histology and immunofluorescence or in 2% glutaraldehyde in 0.12 M phosphate buffer for electron microscopy. The skin was collected after shaving the mice and immersed in 4% PFA for 2–4 h at 4°C. For RNA extraction, western blot analyses and TUNEL assay, mice were sacrificed by cervical dislocation.
Tissues were quickly collected and frozen in liquid nitrogen. RNA extraction was performed with TRIzol reagent (Life Technologies) according to the manufacturer specifications. cDNA was synthesized using SuperScript First-Strand Synthesis System (Life Technologies). The sequence of primers used for RT-PCR are available upon request.
Postfixed brain and spinal cord were embedded in 6% agar, and 30 μm sections were cut using a vibratome (VS1000, Leica). Gallyas’ staining was performed as previously described [56] and images were captured with slide scanner (SCN400, Leica). For immunofluorescence, free-floating sections were permeabilized and blocked in 0.4% Triton X-100 and 10% goat serum in TBS for one hour at RT. Primary antibodies were incubated overnight at 4°C, followed by incubation with secondary antibodies for 2 h at RT. Sections were mounted in FluorSave Reagent (Calbiochem). Skin specimens were embedded in paraffin and sectioned at 5 μm thickness using a microtome (RM2255, Leica). Skin sections were stained with Haematoxylin solution (MHS32, Sigma) and Eosin Y-solution 0.5% aqueous (1098441000, Millipore). Antigen retrieval was conducted by boiling sections in 0.1 M citrate buffer (pH 6) before immunofluorescence analysis. The following primary antibodies were used for immunofluorescence: APC (1:400, OP80, Calbiochem), COX1 (1:1000, 459600, Invitrogen), cytochrome c (1:1000, 556432, BD Pharmingen), Olig2 (1:500, AB9610, Millipore), GFAP (1:400, 3670, Cell Signaling), IBA1 (1:2000, 019–19741, Wako), GFP (1:1000, ab6556, Abcam), MBP (1:1000, SMI94, Covance), and c-KIT (1:1000, 553352, BD Pharmingen). All secondary antibodies were from Molecular Probes: anti-mouse Alexa Fluor 488 (A-11029), 546 (A-21143), anti-rabbit Alexa Fluor 488 (A-11034), 546 (A-11035), 594 (A-21207), and anti-rat Alexa fluor 488 (A-11001). All fluorescent images were acquired using an Axio-Imager M2 microscope equipped with Apotome 2 (Zeiss) or gSTED super-resolution and confocal microscope with HyD detector (TCS SP 8, Leica), as specified. When specified, Huygens Deconvolution software was employed.
Quantification of APC+, Olig2+, and mt-YFP+ cells was performed manually on single plane images of brain coronal vibratome sections (30 μm). Three sections of each brain cut at comparable levels (about -1.50 mm, -2.8 mm, -3.4 mm from the bregma) were stained with the indicated antibodies. Two to five images of non-overlapping fields in the corpus callosum of one hemisphere were taken for each section and the number of positive cells/area was manually counted. 3 independent mice per genotype and time point were used for quantification. Quantification of APC+ cell size was performed on images acquired using the same exposure time using the measure function of the Axiovision software (Zeiss).
To visualize mitochondrial morphology in targeted cells, mice were crossed with ROSA26+/SmY mice. To visualize endogenous mt-YFP signal, sciatic nerves were dissected out, embedded in O.C.T. (Tissue-Tek) and were cut longitudinally at a thickness of 7 μm using a cryostat (CM1850, Leica). For cryosections, the skin was cryoprotected in 15% sucrose for 2 h and then in 30% sucrose overnight, embedded in O.C.T. (Tissue-Tek), frozen on dry ice and sectioned with a cryostat (CM1850, Leica). 10 μm frozen sections were directly mounted for imaging. For all analyses performed in the brain, vibratome sections were incubated with an anti-GFP antibody to enhance the endogenous YFP signal. Fluorescent images were acquired using a gSTED super-resolution and confocal microscope with HyD detector (TCS SP 8, Leica), as specified. The circularity of mitochondria in targeted oligodendrocytes within the corpus callosum was measured using a macro of ImageJ. The circularity was calculated with the following formula: 4 *pi*(area/perimeter^2) [57].
To detect TUNEL+ cells, the In Situ Apoptosis assay (S7101, Millipore) was used on cryostat sections out following the manufacturer’s protocol. The number of TUNEL+ cells within the corpus callosum was quantified manually. 4 mice per genotype and 2–3 sections from each mouse were used for quantification.
The corpus callosum and the lumbar spinal cord were post-fixed in 2% glutaraldehyde (Sigma) in 0.12 M phosphate buffer and were treated with 1% osmium tetroxide (Sigma). After dehydration with ethanol and propylene oxide, tissues were embedded in Epon (Fluka). Tissue in Epon-block was further trimmed and cut using an ultramicrotome (EM UC7, Leica). 1 μm semithin sections were prepared and stained with 1% toluidine blue for light microscopy. For electron microscopy, 70 nm ultrathin sections were cut and stained with uranyl acetate (Plano GMBH) and lead citrate (Electron Microscopy Sciences). Images were taken by a transmission electron microscope (CM10, Phillips) equipped with Orius SC200W camera.
Quantification of dark cells and number of myelinated axons was performed on at least three semithin micrographs of the anterolateral funiculus of the lumbar spinal cord per mouse. Three independent mice per genotype were analyzed. The number of myelinated axons was quantified using ImageJ particle analyzer with the setting of size 50-infinity and circularity 0.3–1.0. The g ratio was determined by measuring the ratio between the diameter of the axon and the diameter of the myelinated fiber on electron micrographs from 3 mice per genotype.
Brains from P5 pups were removed and manually dissociated with the Neuronal Tissue Dissociation Kit (130-092-628, Miltenyi Biotec). To purify oligodendrocytes, the cell suspension was further incubated with anti-O4 magnetic beads (130-096-670, Miltenyi Biotec), washed and loaded onto 30 μm pre-separation filters fixed on top of MS MACS Columns (130-042-201, Miltenyi Biotec), which were placed in a magnetic field of the MACS separator (Miltenyi Biotec). The magnetic labeled O4+ cells were retained within the columns, while flow-through was collected. Finally, magnetically labeled cells were flushed out by firmly pushing the plunger into each column.
Astrocytes were isolated from the cerebellum of P0-P3 newborn pups. The cerebellum was dissected in dissection solution (60% EBSS; 4% Glucose; 30 mM HEPES; 30% FCS III). After removing the meninges, the tissue was mechanically meshed, washed twice with EBSS solution (10% HEPES; 90% EBSS), and placed in glia medium (90% DMEM/F12 Hams media with L-glutamine; 9% FCS III; 1% Pen/Strep). The single-cell suspension was obtained by mechanical dissociation. The cells were then passed through a 100 μm Nylon cell strainer and the strainer was then washed with 5 ml of glia medium. The cells were centrifuged at 800g for 5 minutes at 4°C. The supernatant was removed and the pellet was resuspended in 10 ml glia medium and plated in 75 cm2 flasks previously coated with poly-L-lysine (0.1mg/ml). For enriched neuronal cultures, the cortex and the hippocampus were dissected from E16.5 mouse embryos. The meninges were removed prior mechanical dissociation of the tissue. Chemical dissociation was obtained with Trypsin solution (0.025% in HBSS) for 15 minutes at 37°C. The tissue was then washed 3 times with HBSS for 5 minute at 37°C and triturated with Pasteur pipettes (0.5 mm opening size) for 7 times. The cells were cultured in in Neurobasal plating Media. One week following the seeding, the cells were collected for protein extraction.
Mitochondria from tissues were isolated in MOPS sucrose buffer (440 mM sucrose, 20 mM MOPS, 1 mM EDTA, 0.2 mM phenylmethylsulfonyl fluoride) by differential centrifugation at 10,000 g. 100 μg of mitochondria were solubilized in 1 M ε-amino n-caproic acid, 50 mM Tris (pH 7.0) and digitonin at a detergent to protein ratio of 4 g/g. BN-PAGE was performed as previous described [58].
Cells or tissues were lysed in RIPA buffer and immunoblot analysis was conducted as described [18]. The following primary antibodies were used: AFG3L1 (1:1000) [12], AFG3L2 (1:1000) [12], Paraplegin (1:500) [59], SDHA (1:4000, A11142, Molecular Probes), β-III tubulin (1:1000, T8660, Sigma), MBP (1:2000, SMI94, Covance), CNP (1:500, C5922, Sigma), GAPDH (1:2000, MAB374, Chemicon), CNX (1:4000, ADI-SPA-860, Enzo), GFAP (1:2000, Z0334, Dako).
All statistical analyses were performed using GraphPad Prism 6 software, presenting the data as mean ± standard deviation (SD) or as mean ± standard error of the mean (SEM). If not stated otherwise, p value was determined by two-tailed unpaired Student’s t test. Statistical significance was defined as *p < 0.05, **p < 0.01 and ***p < 0.001.
|
10.1371/journal.pntd.0007360 | Seasonal patterns of dengue fever in rural Ecuador: 2009-2016 | Season is a major determinant of infectious disease rates, including arboviruses spread by mosquitoes, such as dengue, chikungunya, and Zika. Seasonal patterns of disease are driven by a combination of climatic or environmental factors, such as temperature or rainfall, and human behavioral time trends, such as school year schedules, holidays, and weekday-weekend patterns. These factors affect both disease rates and healthcare-seeking behavior. Seasonality of dengue fever has been studied in the context of climatic factors, but short- and long-term time trends are less well-understood. With 2009–2016 medical record data from patients diagnosed with dengue fever at two hospitals in rural Ecuador, we used Poisson generalized linear modeling to determine short- and long-term seasonal patterns of dengue fever, as well as the effect of day of the week and public holidays. In a subset analysis, we determined the impact of school schedules on school-aged children. With a separate model, we examined the effect of climate on diagnosis patterns. In the first model, the most important predictors of dengue fever were annual sinusoidal fluctuations in disease, long-term trends (as represented by a spline for the full study duration), day of the week, and hospital. Seasonal trends showed single peaks in case diagnoses, during mid-March. Compared to the average of all days, cases were more likely to be diagnosed on Tuesdays (risk ratio (RR): 1.26, 95% confidence interval (CI) 1.05–1.51) and Thursdays (RR: 1.25, 95% CI 1.02–1.53), and less likely to be diagnosed on Saturdays (RR: 0.81, 95% CI 0.65–1.01) and Sundays (RR: 0.74, 95% CI 0.58–0.95). Public holidays were not significant predictors of dengue fever diagnoses, except for an increase in diagnoses on the day after Christmas (RR: 2.77, 95% CI 1.46–5.24). School schedules did not impact dengue diagnoses in school-aged children. In the climate model, important climate variables included the monthly total precipitation, an interaction between total precipitation and monthly absolute minimum temperature, an interaction between total precipitation and monthly precipitation days, and a three-way interaction between minimum temperature, total precipitation, and precipitation days. This is the first report of long-term dengue fever seasonality in Ecuador, one of few reports from rural patients, and one of very few studies utilizing daily disease reports. These results can inform local disease prevention efforts, public health planning, as well as global and regional models of dengue fever trends.
| Dengue fever exhibits a seasonal pattern in many parts of the world, much of which has been attributed to climate and weather. However, additional factors may contribute to dengue seasonality. With 2009–2016 medical record data from rural Ecuador, we studied the short- and long-term seasonal patterns of dengue fever, as well as the effect of school schedules and public holidays. We also examined the effect of climate on dengue. We found that dengue diagnoses peak once per year in mid-March, but that diagnoses are also affected by day of the week. Dengue was also impacted by regional climate and complex interactions between local weather variables. This is the first report of long-term dengue fever seasonality in Ecuador, one of few reports from rural patients, and one of very few studies utilizing daily disease reports. This is the first report on the impacts of school schedules, holidays, and weekday-weekend patterns on dengue diagnoses. These results suggest a potential impact of human behaviors on dengue exposure risk. More broadly, these results can inform local disease prevention efforts and public health planning, as well as global and regional models of dengue fever trends.
| Seasonality of infectious disease is a phenomenon commonly observed in the northern and southern hemispheres, with seasonality of influenza being the most well-known and well-studied infectious disease with a seasonal pattern [1–6]. Seasonality has also been observed with other infectious diseases, including malaria [7], dengue [8], tuberculosis [9, 10], acute respiratory infection [1, 11], and foodborne illness [12–15]. These relationships are often a combination of climatic and environmental factors and how these factors affect pathogen transmissibility [15, 16], vector abundance [8, 17–21], and human health, and drive human behaviors such as diet, crowding, travel patterns, and outdoor exposures [8, 14, 15, 19, 20].
Mosquito-borne viral infections include dengue fever, yellow fever, chikungunya, and Zika, among others [22]. These illnesses are common in tropical countries and are most often spread by mosquitoes in the Aedes genus. Dengue virus is the most common, and may present with fever, rash, and general pain; although an estimated 80% of dengue patients are asymptomatic [23], this infection can have serious health consequences, including death [24].
The diagnosis of dengue and other acute febrile illnesses can be extremely difficult, depending on the stage of the illness and the resources available at the point of care. Dengue cannot always be distinguished from other febrile illnesses, though diagnostic testing, including rapid tests, ELISA, and PCR-based assays are sometimes available and can aid with diagnosis [25], though the sensitivity and specificity of these tests are not perfect. Correct diagnosis of dengue additionally relies on the patient’s presenting signs and symptoms as well as the expertise of the clinician.
Seasonality affects dengue diagnosis rates through several mechanisms. Seasons drive human behavior: people may be more or less likely to spend time crowded indoors or spread outdoors depending on the time of year, which affects exposure rates. This can be the result of weather conditions or a result of seasonal holidays, which affect school and work schedules, and drive public gatherings (such as parades) or private family gatherings. There is also reason to believe that seasonality affects host immunity: in tropical countries, both cell-mediated and humoral immune responses are decreased during the rainy season [26]. This could be driven by seasonal variation in gene expression [27], levels of immune-modulators and blood cell composition [28], food availability, daylight exposure, and/or environmental exposures [26], though the causal direction of changes in the immune system, season, and seasonal disease is unclear. In addition, long-term or multi-annual disease trends are often a reflection of a buildup of disease-specific immunity in a population: for outbreaks to occur, there must be a sufficient number of susceptible individuals in the population. If all persons in the community were infected in the previous years and are therefore immune to circulating strains of virus, no outbreak occurs and the season will have a relatively low intensity, and the low intensity will continue until additional susceptibles are available from birth, migration, or introduction of a new dengue serotype.
Climate is a major component of seasonality and directly impacts the life history and behavior of the mosquito vector. Aedes aegypti, which is the principal vector of dengue in Ecuador, has been well-characterized in its relationship to temperature, which has been shown to impact development rates, lifespan, fecundity, survival, biting rates, transmission probability, infection probability, abundance and incubation rates in both field and laboratory studies [29–36]. Field studies of rainfall have found associations between larval or adult abundance and precipitation [37–39]. Because temperature and precipitation can affect mosquitoes throughout their life course, the temporal scale of climate-mosquito associations can vary, depending on the life stage of the mosquito. For example, lagged precipitation (one to two months prior) is linked to larval indices due to the impact of precipitation on larval breeding sites [37], while both lagged temperature (4 weeks) and unlagged [i.e. current] mean temperature have been associated with adult abundance [39, 40]. Adult abundance and biting patterns are critical to dengue risk; climate plays a major role in the activity levels of these vectors [33].
The climate of Ecuador is highly diverse; though small in area, it contains 11 different Köppen-Geiger climate classifications, with the coast being generally classified as hot and semi-arid or tropical savanna climates, the central Andean range as oceanic or warm-summer Mediterranean climates, and the eastern rainforest as tropical rainforest climates [41]. Ecuador is also impacted by the El Niño/Southern Oscillation (ENSO) phenomenon in which the surface temperature of the Pacific ocean leads to periodic changes in regional weather patterns [42]. Specifically, an El Niño year will be warmer and wetter than average in Ecuador, and a La Niña year will be drier and cooler than average [42].
Studies of disease seasonality in tropical regions are limited. For mosquito-borne disease, previous research has largely focused on climatic and environmental variables, which directly affect vector abundance. In Ecuador, this research has been limited to two studies of dengue cases in coastal regions; In one study, minimum weekly temperature and mean weekly precipitation were shown to be strongly linked to weekly number of dengue cases [19]. A second study in the same area found that minimum weekly temperature, precipitation, and El Niño events were positively associated with dengue risk [20]. These studies both occurred in a large city the southern coast of Ecuador; given the diversity of climates and communities in Ecuador and the need for relevant evidence to make policy decisions, it is important to determine if the causal relationships between seasonal factors, climates, and dengue cases are similar in other areas of Ecuador.
With the present study we determined the seasonality of dengue fever by decomposing seasonality into two components: non-climate seasonality and climate-driven seasonality, using data from patients clinically diagnosed with dengue fever at two hospitals in rural Ecuador with a subtropical climate. Non-climate trends included short- and long-term trends, and the effects of school sessions, public holidays, and weekdays on these diagnoses. Climate-driven trends included an examination of regional and local climate variable impacts on dengue fever diagnoses.
Hospital Pedro Vicente Maldonado (HPVM) is a 17-bed rural hospital located in Pedro Vicente Maldonado (PVM), Pichincha, Ecuador (Fig 1). It primarily serves patients from Cantons Pedro Vicente Maldonado, Puerto Quito, San Miguel de los Bancos, and Santo Domingo. Pedro Vicente Maldonado is located at 0°05'12.3"N, 79°03'08.0"W, and northwest of Quito, at approximately 600 meters altitude, with a projected 2016 population of 6,944. Hospital Saludesa (HS) is a 60-bed metropolitan hospital located in Santo Domingo de los Tsáchilas (SD), Santo Domingo de los Tsáchilas, Ecuador (Fig 1). It serves patients from Santo Domingo de los Tsáchilas Province. Santo Domingo de los Tsáchilas is located at 0°15'15"S, 79°10'19"W, and west of Quito, at approximately 550 meters altitude, with a population of 305,632 (2010 Census). Both hospitals have 24-hour, 7-days-a-week emergency rooms, with regular consultation available on Mondays—Saturdays. During holidays, only the emergency room services are available. Both hospitals have clinical laboratory services available, including the NS1 dengue antigen rapid and dengue IgG antibody rapid tests (Human, Wiesbaden, Germany); the NS1 dengue antigen rapid test is the diagnostic of choice. These cities have a tropical rainforest climate; average monthly temperatures run from 71.8° Fahrenheit (22.1° Celsius) in November to 74.8° Fahrenheit (23.8° Celsius) in March. Average total monthly precipitation runs from 110 millimeters (mm) in July to 671 mm in April. Both sites have ongoing mosquito control programs. Cities are fumigated approximately once per month with repellent, and residents are provided with temephos (Abate) treatment for water stored in large laundry tanks.
For this medical record review, we examined de-identified records with a primary diagnosis of arthropod-borne viral fevers and viral hemorrhagic fevers. These included International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10) codes A90—A99. Records from Hospital Pedro Vicente Maldonado included consult dates from August 1, 2009 through July 31, 2016. Records from Hospital Saludesa included consult dates from July 1, 2014 through July 31, 2016. The following variables were available for analysis: consult date, primary diagnosis, ICD-10 code, and patient demographics (age, sex, insurance status, county-level address, weight, and height). We set criteria to exclude patients missing more than 50% of these variables. Information regarding school schedules and holiday dates in each year was obtained from the Ecuadorian Ministry of Education, the Ministry of Tourism and local residents [43–46]. School sessions and holidays analyzed in this study are in Table 1. Data for monthly climate variables measured at the La Concordia station (0°01'29.0"N, 79°22'49.0"W, Fig 1) were obtained from the National Institute of Meteorology and Hydrology in Ecuador [47, 48]. Oceanic Niño Indices (ONI), a measure of ENSO effects, were obtained from the National Weather Service (http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ensoyears.shtml).
This research was certified as non-human subjects research by the Institutional Review Board of University of Wisconsin-Madison (#2017–0033).
Four observations (5% of total) for monthly absolute minimum temperature were missing. Multiple imputation was used to estimate these values using all available monthly climate variables, monthly case counts, and time. Ten imputations were performed with a fully conditional specification algorithm; parameters were pooled and used to obtain estimates.
Log-linked Poisson generalized linear models with generalized estimating equations (GEE) (autoregressive correlation structure) were used for all models. Models using GEE account for correlation in data, as is common in time series data. To account for temporal autocorrelation, cases were clustered by week of diagnosis. Model fit was assessed using quasi-likelihood under the independence model information criterion (QIC). Model 1 was used to evaluate the intra-annual and long-term seasonality of disease in a non-climate seasonal model; these seasonal components were included because there is evidence for intra-annual patterns in dengue diagnoses elsewhere in Ecuador [20] and our dataset had large year-to-year variations in diagnoses. Daily case counts were the outcome of interest and data from both hospitals were combined, with an indicator variable for hospital of origin. Long-term trends were estimated with a restricted cubic spline; number of knots was determined by best fit. For intra-annual effects, we compared sine and cosine waves with frequencies of once, twice, and/or three times in 365 days, using best fit to select the final fit. After selecting the best fit for the long-term and intra-annual effects, we added day-of-week and holidays as indicator variables, with the hypothesis that day-of-week may impact care-seeking decisions, and that patients may be less likely to seek care on a holiday (due to family obligations or travel). Holidays included official government-declared holidays and any weekends immediately before or after these holidays, as well as the day after Christmas.
Because children may differ in their exposure to dengue risk factors when school is in session, the effect of school schedules were examined using a subset analysis (Model 2). We restricted this analysis to school-aged children (ages 4–18) who sought care at Hospital Pedro Vicente Maldonado (n = 142). We used the best-fit long-term and intra-annual effects model from the first analysis and included an indicator variable for days where school was in session (including weekends during the school year).
To determine the impact of climate on disease seasonality, we built a log-linked Poisson generalized linear model (Model 3). Climate data were available as monthly averages, so daily case counts were aggregated to monthly counts. Temperature and precipitation variables were centered on their mean value; temperatures were scaled at 2° Celsius, number of days with precipitation were scaled at 5 days and total monthly precipitation was scaled at 10 mm. The effects of climate variables (all continuous or integer variables), including ONI, average monthly temperature, minimum monthly temperature, maximum monthly temperature, total monthly precipitation, and number of days per month with precipitation were evaluated. Because climate variables interact with each other in reality, we also examined interactions between the significant climate variables in the final model.
Data analysis and visualization was performed using SAS version 9.2 (SAS Institute, Cary, NC) including the macros DASPLINE, DSHIDE, and weekno [49, 50], and R version 3.2.2 (R Foundation for Statistical Computing, Vienna, Austria) including packages haven, raster, dismo, ggmap, OpenStreetMap, sp, geepack, and MASS [51–59].
Characteristics of the data used in this study are in Table 2, with patient demographics available in S1 Table. No cases met the exclusion criteria; all cases were included in analysis. The diagnoses in the dataset included dengue fever (A90), dengue hemorrhagic fever (A91), other mosquito-borne viral fevers (A92), and mosquito-borne viral encephalitis (A83). Dengue diagnoses comprised 98.7% of the patients in the study. On average, one case is diagnosed at Pedro Vicente Maldonado every 4.3 days, and one case is diagnosed at Saludesa every 25 days. Time series plots of aggregated monthly case data from both hospitals, and monthly climate data are in Fig 2.
The final model for non-climate seasonality (Model 1, parameters in S2 Table) included a sine and cosine wave with an annual cycle, long-term patterns, day-of-week effects, and indicator variables to designate holidays and hospitals. Fits metrics for the null model and each considered model are available in S3 Table. Model 1 predictions for daily diagnoses are presented in Fig 3 and exhibit an annual peak of disease in mid-March each year on average. Day-of-week effects are summarized in Fig 4. Compared to the average of all days, Tuesdays and Thursdays were more likely to have dengue fever diagnoses (Tuesday: relative risk (RR) = 1.26, 95% confidence interval (CI) 1.05–1.51, p = 0.012, Thursday: RR = 1.25, 95% CI 1.02–1.52, p = 0.033), while Saturdays and Sundays were less likely to have dengue fever diagnoses (Saturday: RR: 0.81, 95% CI 0.64–1.01, p = 0.062 Sunday: RR: 0.74, 95% CI 0.58–0.95, p = 0.016). Compared to non-holidays, dengue fever cases were much more likely to be diagnosed the day after Christmas (RR: 2.80, 95% CI 1.46–5.30, p = 0.002), after holding all other covariates constant. The subanalysis (Model 2) did not find an effect of school session on dengue diagnoses.
Most climate variables exhibited small but significant effects on risk of dengue fever diagnoses. Fit metrics for the null model and each considered climate model are available in S4 Table. Greater total monthly precipitation (RR: 2.14, 95% CI 1.26–3.64, p = 0.005) results in increases in dengue fever diagnoses, i.e. for every 10mm increase in monthly precipitation, there is an approximately two-fold increase in dengue fever diagnoses on average. In addition, there were significant interactions between total monthly precipitation, number of days with precipitation, and monthly absolute minimum temperature. Model 3 predictions of interaction variable effects are in Fig 5, wherein observed values for monthly absolute minimum temperature, total monthly precipitation, and days per month with precipitation were used to predict the number of dengue cases per month within a reasonable range of precipitation and minimum temperature values. At an absolute minimum temperature of 18–19° C, the predicted number of cases increased (5 to 15 cases per month) as total monthly precipitation increased (from 125 to 875 mm per month) and decreased as the number of days with precipitation increased (from 5 to 30 days per month), but as minimum temperatures warm, the direction of these relationships changes. When the absolute minimum temperature is 20° C, additional days with precipitation or increases in monthly amounts of precipitation have little effect on the number of diagnoses. For a monthly minimum temperature of 21–22° C, the effect of increased amounts of precipitation is weaker, but still positive, while the impact of number of days with precipitation at warmer temperatures leads to increases in the number of dengue diagnoses (from 2 to 10 cases per month).
Understanding the seasonality of infectious diseases can be crucial to the public health efforts to control these diseases. Seasonality is a major determinant of vaccination scheduling, timing of educational campaigns, and allocation of resources. In this paper, we examine non-climate (long-term trends, intra-annual patterns, day-of-week and holiday effects) and climate components of seasonality.
Our data exhibits annual peaks in dengue fever diagnoses, occurring in mid- March. Long-term studies of dengue seasonality in coastal Ecuador and other countries also exhibit annual peaks as well as inter-epidemic periods (high-intensity dengue fever seasons followed by a low-intensity season or seasons the following two years). Coastal Ecuador exhibits significant annual and two-year peaks in dengue incidence [20]; additional studies indicate that El Niño events, which occur in variable annual or multi-year patterns, may also influence dengue incidence patterns [19]. Our data exhibits peaks in 2012, and 2015; 2009 and 2015 were moderate and very strong were El Niño years, respectively [60]. Research from Peru suggests annual and three-year peaks in dengue incidence [61], while Colombia experiences two- to five-year cycles [62], with some parts of Colombia lacking annual disease peaks [63].
In this dataset, dengue fever diagnoses were likely affected by healthcare-seeking behavior. The decision and timing of seeking care for health problems can be affected by short-term time trends including day-of-week and holiday patterns. This type of research is scarce in South America. In the US and the UK, research on day-of-week effects has found that patients are less likely to visit the hospital on a weekend and that weekend hospital visits tend to be non-elective [64, 65], suggesting that patients may put off healthcare for less serious health conditions. Our findings agree with previous research, with Saturdays and Sundays being the least likely days for dengue fever diagnosis. However, we additionally found an increase in diagnoses on Tuesdays and Thursdays. We speculate that there may be some underlying pattern to diagnostic capabilities (e.g. staffing patterns, shipment days for lab supplies, or a backlog of patient samples from the weekend). We also examined holidays, with the reasoning that patients would also delay healthcare until after holidays. Previous studies suggest that holiday effects may be complex: research from Colombia has shown increases in dengue during periods immediately following holidays, from patients travelling to dengue-endemic areas during the holidays [66]. In our study, the individual holidays largely had no effect on dengue diagnoses, except for the day after Christmas (p = 0.0015), when patients were more likely to be diagnosed with dengue fever. Since the incubation period for dengue is 4–10 days, and since most patients are local, we feel this spike in diagnoses is from those who became ill over the holiday and delayed their care, rather than acquired their illness during holiday travel. However, other than decreased diagnoses on Saturdays, day-to-day patterns of infectious disease healthcare-seeking at these hospitals did not have the same fluctuations as the dengue patients (S5 Table), meaning these patterns may be the result of statistical noise and not general health-care seeking behaviors in the community.
In our seasonality assessment, we found that dengue fever diagnoses peaks during mid-March on average. This is the first assessment of dengue fever seasonality in rural Ecuador. Reports from nearby Colombia regarding dengue fever seasonality have not found an annual seasonal pattern for dengue incidence [63, 67], though these studies did not utilize sinusoidal variables, making it difficult to detect these patterns.
Climatic factors such as temperature or precipitation can affect the survival and distribution of mosquito vectors and the transmissibility of pathogens from these vectors [16–18]. In previous research in Colombia, studies have found average temperature, changes in average temperature, average relative humidity, total precipitation, and El Niño events to be major predictors of dengue incidence [63, 68]. Research in Ecuador has been limited to studies of dengue cases in coastal regions. In one study, minimum weekly temperature and weekly average precipitation were shown to be strongly linked to weekly number of dengue cases [20]. Minimum weekly temperature, precipitation, and El Niño events were also positively associated with dengue risk [19]. Our data illustrate a complex relationship between climate factors and dengue fever diagnoses. Temperature is a major factor; dengue transmission is sensitive to extremes of temperature as Aedes aegypti propagate and transmit dengue best between 18–32° C [63], but precipitation is also important. In isolation, total monthly precipitation and number of days with precipitation had opposite effects, suggesting that sufficient precipitation is necessary for dengue cases to occur, but that too many days with precipitation decrease risk. However, when we consider minimum monthly temperature, temperature modifies the effects of precipitation in a U-shaped pattern. All amounts of precipitation drive increases in dengue diagnoses but additional days with precipitation lead to decreases in dengue diagnoses while when temperatures are lowest, as in the months of July through November (mean minimum temperatures of 19.0–19.7° C). During these months precipitation amounts are all below 250 mm on average and durations are 11.7 to 15.7 days on average, resulting in a relatively low predicted number of dengue fever cases. At warmer temperatures, both number of days with and amount of precipitation have positive relationships with the number of dengue diagnoses. At the warmer part of the year–i.e. December through June (mean minimum temperatures of 19.8–21.2° C), precipitation quantity is higher (mean 271.6–635.4 mm per month) and occurs on more days (mean 21.6–28.3 days per month).
Our results likely reflect the effect of precipitation on mosquitoes: female Aedes aegypti mosquitoes tend to lay eggs just above the water surface in containers [69] until additional precipitation (i.e. flooding of the eggs) causes the eggs to hatch, but too much precipitation can wash eggs or larvae out of their containers [70], meaning some dry periods are necessary or even beneficial to Aedes aegypti abundance. Previous research has found that Aedes aegypti breeding site occupancy is increased at sites with longer dry periods [71]. Temperature levels affect evaporation rates and the durability of standing water (i.e. breeding and development sites); this may explain temperature’s modifying impact on the relationship between precipitation and dengue diagnoses.
Human hosts may also change their travel outside the home during consistently rainy periods, which may alter their exposure to dengue-infected mosquitoes (depending on where they are most exposed). Research in Australia found that virus acquisition was spatiotemporally linked to the case’s residence in 42% of dengue cases [72], though this proportion may differ in other geographic locations. Human movement and behavior is a major component of dengue fever risk [73]. Weather patterns affect human movements, with high movement variation on days with higher precipitation [74]. The patterns between dengue fever risk and climate variables observed in our data are likely a combination of the effect of climate on mosquito vectors and human behaviors.
Notably, the best-fit climate seasonality model included both long-term and annual sinusoidal variables, in addition to climate variables. If dengue seasonality were entirely driven by climate, we would expect that a model adjusting for the effects of climate to be sufficient with no long-term nor annual sinusoidal variables (i.e. all variation in the diagnosis rate would be explained by the climate variables). The importance of the long-term and annual sinusoidal variables in our climate model suggests that we are not completely adjusting for the effect of climate or that non-climate phenomena impact the seasonality of dengue diagnoses. Our ability to disentangle climate and non-climate seasonality is complicated by the introduction of chikungunya into a naïve population in 2015. These cases were treated as dengue diagnoses in our models, but it is impossible to know how many during this period were true dengue cases. Because chikungunya and dengue are spread by the same mosquito species, Aedes aegypti, we expect that many of the effects of climate will be the same for both chikungunya and dengue. The effects of chikungunya emergence on overall seasonality are important to consider. This introduction occurred outside of the typical dengue season (November) and had a high number of cases, meaning that the average annual peak of dengue is slightly earlier in our model than the true average annual peak of dengue in this population.
This dataset combines the patients from two hospitals. The patients at each hospital differ in their gender composition and insurance status. Both gender and insurance status likely affect healthcare-seeking behavior, meaning that the hospital populations may have different non-climate seasonal patterns of dengue diagnoses and some uncontrolled confounding may affect our results. However, we do control for the source hospital in our analysis which would control for some of these differences, and since the majority of cases (94.5%) are from one hospital, we do not expect this issue to substantially affect our results.
This dataset represents dengue fever diagnoses in the community and is only a proxy for dengue fever incidence rates. There are likely to be many more cases dengue fever in the community: 80% of dengue cases are estimated to be asymptomatic, some symptomatic patients may never seek care, and some symptomatic patients may have sought care at hospitals other than those included in this study. This could be a potential source of selection bias. However, our study hospitals are the major source of care in their communities and we are assessing seasonality and climate variables; we have no reason to believe that the effect of seasonality and climate is any different among symptomatic versus asymptomatic patients nor for the small number of persons who sought care at other clinics. The effect of selection bias on these data is likely minimal.
Dengue diagnosis can be difficult even for experienced clinicians, especially in a resource-limited setting such as Ecuador. Not all patients with a final dengue diagnosis were necessarily lab-confirmed; the use a laboratory confirmation likely varies by clinician, patient, and presenting symptoms, though the clinicians at the study hospitals are all experienced with dengue diagnosis. Based on observed hospital practices, we believe many of the cases in our dataset had positive dengue rapid tests, but that some were clinically diagnosed. Because not all cases were laboratory-confirmed, it is possible that some non-dengue cases were diagnosed as dengue, particularly when chikungunya was introduced to Ecuador (late 2015) and no diagnostic tools were available for chikungunya.
Our dataset only covers a seven-year period making it difficult to conclude if our observations truly reflect long-term or multi-year disease trends in this community. Additional research for longer periods of time will reveal if three-year peaks or changes in dengue fever diagnose rates are present in this community. The diagnosis of dengue could have been impacted by additional phenomena over the study period. Changes in mosquito control practices could affect actual disease rates or worsening economic conditions in Ecuador (due to a decrease in oil prices) would adversely affect the ability of patients to seek healthcare. In addition, a major earthquake in April 2016 disrupted many services in Ecuador, including transportation, utilities, and healthcare for several weeks, which may have disrupted the typical healthcare-seeking behavior of patients and the diagnostic capabilities of the hospitals during this time.
Available climate data was captured from a climatological station located 39 and 36 kilometers from Hospital Pedro Vicente and Hospital Saludesa, respectively. These data are only a proxy for actual climate conditions in our communities of interest. In addition, analyses with climate variables were limited to monthly summaries of these variables, making it difficult to ascertain if the relationships discovered in this research reflect the true relationship between climate variables and dengue fever diagnoses in these populations. Under the assumption that most patients would be bitten, experience symptoms, and seek care within the same month, the climate-diagnosis relationships presented in this study are a good estimate of dengue seasonality in these communities. In reality, there is considerable variation among the climate variables, mosquito exposure and dengue diagnoses in this community, which we were unable to capture in this study. Nor are we able to estimate the effects of climate variable interactions among ranges and combinations of variables that were unobserved in this location. In addition, the effect estimates for the climate variable interactions were often based on a small sample size, leading to wide confidence intervals for these estimates. Indeed, this veracity of this interaction will need to be confirmed with additional research. Future research will also address the limited range and unobserved climate combinations in this dataset by testing this interaction with data from areas with different climate conditions.
|
10.1371/journal.pgen.1004372 | Discovery and Functional Annotation of SIX6 Variants in Primary Open-Angle Glaucoma | Glaucoma is a leading cause of blindness worldwide. Primary open-angle glaucoma (POAG) is the most common subtype and is a complex trait with multigenic inheritance. Genome-wide association studies have previously identified a significant association between POAG and the SIX6 locus (rs10483727, odds ratio (OR) = 1.32, p = 3.87×10−11). SIX6 plays a role in ocular development and has been associated with the morphology of the optic nerve. We sequenced the SIX6 coding and regulatory regions in 262 POAG cases and 256 controls and identified six nonsynonymous coding variants, including five rare and one common variant, Asn141His (rs33912345), which was associated significantly with POAG (OR = 1.27, p = 4.2×10−10) in the NEIGHBOR/GLAUGEN datasets. These variants were tested in an in vivo Danio rerio (zebrafish) complementation assay to evaluate ocular metrics such as eye size and optic nerve structure. Five variants, found primarily in POAG cases, were hypomorphic or null, while the sixth variant, found only in controls, was benign. One variant in the SIX6 enhancer increased expression of SIX6 and disrupted its regulation. Finally, to our knowledge for the first time, we have identified a clinical feature in POAG patients that appears to be dependent upon SIX6 genotype: patients who are homozygous for the SIX6 risk allele (His141) have a statistically thinner retinal nerve fiber layer than patients homozygous for the SIX6 non-risk allele (Asn141). Our results, in combination with previous SIX6 work, lead us to hypothesize that SIX6 risk variants disrupt the development of the neural retina, leading to a reduced number of retinal ganglion cells, thereby increasing the risk of glaucoma-associated vision loss.
| Primary open angle glaucoma is a blinding disease for which there is currently no cure, only treatments that may slow its progress. To help understand the mechanisms of this disease and to design more effective treatments, we identified previously a locus, SIX6, that increases the risk of glaucoma. This gene is involved in early eye development and helps to form the retina. In this paper, we test specific sequence variants in SIX6 that are found in glaucoma patients. We show that these variants have a reduced function that interferes with their ability to direct proper formation of the retina. One variant in particular is common, and may be the main reason that this gene is important in the glaucoma disease process. Patients who have two copies of this sequence variant show a change in the structure of their eye consistent with fewer neurons that carry the visual signal to the brain. These neurons typically die as people age, and people who begin life with fewer visual neurons may have an increased risk of glaucoma. Additional research in this topic may lead to new treatments that preserve sight.
| Primary open-angle glaucoma (POAG) is the most common type of glaucoma, a group of diseases that affect approximately 60 million people worldwide and is a leading cause of blindness [1]. Glaucoma is characterized by the progressive death of retinal ganglion cells, leading to optic nerve atrophy and loss of vision. POAG is a complex inherited disorder for which an increasing number of genetic associations have been described, each contributing modestly to disease burden [2], [3].
A recent POAG genome-wide association study found a significant genetic association (rs10483727, odds ratio (OR) = 1.32, p = 3.87×10−11) at the SIX1/SIX6 locus [4]. Variants in the SIX1/SIX6 locus were first associated with quantitative optic nerve parameters in controls, including vertical cup-disc ratio (VCDR), which is used clinically to diagnose and monitor POAG progression [5], [6]. Several studies have independently confirmed the association of the SIX1/SIX6 locus with both VCDR and POAG [7]–[9].
The human SIX gene family consists of six members (SIX1–SIX6), all of which contain two shared protein domains; a DNA binding homeobox domain and a SIX domain, which binds downstream effector molecules [10], [11]. Members of this conserved gene family were originally identified through homology to the Drosophila melanogaster (Drosophila) sine oculis (so) gene, which is required for proper eye development [10], [11] and are thought to function as transcription factors, regulating key developmental steps through a complex regulatory network. During embryonic development, SIX1 is expressed broadly in multiple tissues, including the otic vesicle and the limb mesenchyme. However, expression of SIX6 is restricted to regions of the retina and the pituitary [10], [12]. Drosophila with null so alleles have restricted retinal development, while morpholino knockdown of six6b in Danio rerio (zebrafish) embryos was recently shown to result in a small eye phenotype [10], [13]. In humans, a large deletion on chromosome 14q22.3-q23 that includes SIX6 causes bilateral anophthalmia, the absence of both eyes, demonstrating the importance of SIX gene family members in ocular development and human disease [10], [11], [14]–[16].
In this study, we have extended the current understanding of the molecular contributions of SIX6 to POAG risk. First, we identified potential POAG risk alleles by sequencing the SIX6 gene in a case-control dataset; we found both common and rare coding changes within SIX6 in POAG cases, as well as sequence variants in the SIX6 enhancer. We then used the zebrafish system to demonstrate that these human coding variants have functional consequences in eye development. This analysis of 2–3 day old zebrafish embryos is not intended to fully recapitulate the glaucomatous phenotype; however, it provides in vivo data about the functional effects of human genetic variation on the human SIX6 protein in the context of eye development. We next used luciferase reporter assays, through which we show that a sequence variant found in the SIX6 enhancer of POAG patients may increase SIX6 expression. Finally, we demonstrate that POAG cases homozygous for the SIX6 risk allele rs33912345 have a significantly thinner retinal nerve fiber layer, suggesting a glaucomatous pathogenic mechanism driven by SIX6 dysfunction.
Sequencing of the SIX1 and SIX6 genes in Caucasian POAG cases and controls (262 cases, 256 controls) revealed 23 SNPs (Supplemental Table S1). Nine SNPs were identified in SIX1, but no nonsynonymous SNPs were present in the POAG cases. Sequencing of SIX6 yielded 14 variants including five rare nonsynonymous SNPs in POAG cases and controls, one common nonsynonymous SNP located in the homeobox of SIX6 (rs33912345, Asn141His), and five sequence variants within the SIX6 enhancer. All of these variants are conserved evolutionarily as shown by their positive Genomic Evolution Rate Profiling (GERP) scores (Table 1) [17].
Genotyping of rs33912345 in the Duke POAG case-control dataset (482 cases, 433 controls) resulted in a significant association (OR = 1.40, p = 0.0005, POAG case minor allele frequency (MAF) = 0.47, POAG control MAF = 0.38) with POAG. This SNP is in high linkage disequilibrium (r2 = 0.95) with the intergenic SNP identified originally in POAG and VCDR genome-wide association studies (rs10483727) [4]–[6], [13]. As expected, meta-analysis of the imputed genotype data from the NEIGHBOR and GLAUGEN studies confirmed a significant association between POAG status and rs33912345 (OR = 1.27, p = 4.2×10−10) and other linked SNPs in the region (Supplemental Table S2). Further examination of this locus showed that the association signal includes both upstream and downstream regions of the SIX6 transcript, while remaining entirely downstream of SIX1 (Supplemental 1).
We next performed optical coherence tomography (OCT) to study the retinal characteristics of POAG cases possessing the SIX6 risk and non-risk variants (Table 2). OCT images were only available for POAG cases with the common SNP, rs33912345; no data were available for individuals with the rare SIX6 variants. We assessed retinal nerve fiber layer (RNFL) thickness in thirty POAG cases homozygous for the rs33912345 risk allele (C) or the non-risk allele (A) first by comparing age at disease diagnosis and age at OCT across the two genotypes, because age is known to influence retinal thickness and is thus a potential confounder. We observed no significant difference in age (p = 0.11, p = 0.14, respectively). Next, RNFL thickness was evaluated. The overall thickness (global RNFL) was reduced significantly in cases homozygous for the risk allele compared to cases with the non-risk allele (p = 0.03; mean (SD): C = 58.3 (8.2) µm, A = 67.9 (12.4) µm; Table 2), consistent with the hypothesis that SIX6 may increase POAG susceptibility via changes in the neural retina. To determine which quadrants might be driving this observation, we performed an exploratory, post-hoc comparison of RNFL thickness in the temporal, nasal, inferior, and superior regions. We found that RNFL thickness was reduced significantly in the inferior (p = 0.03) and superior (p = 0.04) quadrants, the two regions affecting directly VCDR measurements.
Given 1) the observed association signal pattern; 2) the lack of coding changes identified in SIX1; 3) the presence of rare missense variants and a common, associated missense SNP in SIX6; 4) retinal nerve fiber layer thickness changes observed in POAG cases homozygous for the SIX6 risk allele; and 5) the localized expression of SIX6 in ocular tissues, we concluded SIX6 is a likely candidate gene in this region. We therefore evaluated the functional relevance of SIX6 and the potential burden of common and rare alleles in this locus in POAG using an in vivo zebrafish complementation assay.
First, we performed a reciprocal BLAST analysis; we identified two orthologs of SIX6 in the zebrafish genome, Six6a and Six6b, both with 91% homology at the protein level (Supplemental Figure S2). Previous overexpression and loss of function studies of Six6 in Mus musculus (mouse) and Xenopus laevis (Xenopus) models reveal a role in regulating the proliferative state of retinal progenitor cells and the size of the eye [18], [19]; therefore, as a first test of whether the identified SIX6 variants are pathogenic and potentially relevant to POAG, we asked whether 1) morpholino-induced suppression of six6a or six6b leads to a reduced eye size; 2) expression of the human SIX6 non-risk allele rescues the morphant eye phenotype; and 3) expression of SIX6 alleles containing POAG risk variants rescues the morphant eye phenotype.
Using translation-blocking morpholinos (MOs) targeting zebrafish six6a and six6b, we injected 1–8 cell stage embryos (N = 50–150) and analyzed live embryos at 3 days post fertilization (dpf). We also tested a splice blocking MO. However, this induced non-specific toxicity, including the accumulation of pericardial fluid that could not be rescued with the human SIX6 transcript. This is not unexpected, as SIX6 is a two exon gene—splice-blocking MOs are generally not recommended for two exon genes; the targeted transcript will not be subject to nonsense mediated decay, possibly leading to the expression of a truncated protein and potential dominant-negative effects (described by the manufacturer: http://www.gene-tools.com/node/18). For these reasons, we used translation-blocking MOs for the remainder of our experiments. Masked scoring of both six6a and six6b morphants revealed ocular phenotypes consistent with loss of function, including a reduction in eye size in more than 80% of embryos (p<0.001, Figure 1). The specificity of the MO was tested by co-injection of 12.5 pg of the human SIX6 non-risk allele mRNA; we observed significant (p<0.001) rescue in six6a but not six6b morphant embryos (90% vs. 10% of embryos, respectively). Together, these data indicate that Six6a is the functional ortholog of human SIX6 and prompted subsequent evaluation of SIX6 variants using the six6a MO.
To investigate the pathogenic potential of all SIX6 variants, we used total eye size and the rescue of the morphant phenotype as the assay's phenotypic readout. We injected a mixture containing six6a MO and each of the human SIX6 alleles containing the coding variants identified via sequencing (Table 1). Subsequent to triplicate injections and masked scoring, these results were compared to the rescue condition of the human SIX6 non-risk allele. We found that five of the six variants tested were unable to fully rescue the small eye phenotype. Four of these alleles (Glu129Lys, Asn141His, Thr212Met, and Ser242Ile) resulted in an average eye size larger than the morpholino alone (p<0.001), but smaller than the rescue with the non-risk allele (p<0.001), indicating that these alleles are hypomorphic (Figure 2, Table 3). We also observed one variant (Leu205Arg) with an average eye size smaller than the morpholino alone (p = 0.002; mean (SD): MO = 34,042 (5,763) µm2, Leu205Arg = 31,568 (6,485) µm2; Figure 2), suggesting that it is functionally null. Finally, one allele (Glu93Gln) resulted in an eye size similar to the rescue with the non-risk allele (p = 0.37) and was determined to be benign. The benign allele was identified in one control individual, while the remaining hypomorphic and null alleles were identified either exclusively or primarily in POAG cases (Table 1). Injection of 12.5 pg of the human SIX6 risk mRNA into non-morphant zebrafish provided no evidence of a toxic gain of function compared to injection with the non-risk allele (data not shown).
Given the reduction of the RNFL in cases homozygous for the risk allele, we next asked whether six6a and the identified SIX6 variants impacted the optic nerve, an anatomical site directly relevant to human POAG, in zebrafish. Using whole mount imaging of acetylated-tubulin expression in 2-dpf embryos injected with a control and six6a morpholino, we evaluated volumetric regions of interest (ROI) along the optic nerve (Supplemental Figure S3). Masked scoring of embryos revealed an approximately 3 fold reduction (p<0.001) in the volume of the optic nerve upon depletion of six6a (Figure 3 A–B). This was specific for the optic nerve as the volume of other axonal tracts in the brain were unaffected by six6a depletion (Figure 3A). Specificity of the volumetric measurements was demonstrated upon full rescue of the six6a morphant phenotype by co-injection of the non-risk allele (p<0.001) or a variant that scored as a benign allele in the eye size assay (Glu93Gln; p<0.001; Figure 3B). Both Leu205Arg and Asn141His variants performed as hypomorphic alleles (p<0.01), revealing concordance of our optic nerve assay with the OCT imaging findings in patients.
We hypothesized that POAG risk may be mediated not only by deficits in SIX6 protein function, but also by the level of SIX6 gene expression. To test this, we sequenced the SIX6 retinal specific enhancer element in 262 POAG cases and 256 POAG controls; we identified five variants (Chr14:60974363_C, Chr14:60974373_T, Chr14:60974378_T, Chr14:60974400_A, Chr14:60974449_G) (Table 1), and tested their effect on expression using an in vitro luciferase assay. We found that one of these variants (Chr14:60974449_G) resulted in a significant increase in expression compared to the reference enhancer (Figure 4). Activation of the SIX6 enhancer requires two cofactors, NeuroD and E47 (Supplemental Figure S4) [12]. Overexpression was observed with the Chr14:60974449_G variant even in the absence of these cofactors (Supplemental Figure S5), suggesting variants within the enhancer region may result in dysregulated protein expression.
The SIX1/SIX6 locus has been shown to be associated significantly with POAG in several independent studies; however, the causal variant(s) driving this association have remained unknown [4], [6]–[9], [13], as has the direction of effect of these common alleles on protein function. We have demonstrated through several lines of evidence—the tissue specificity of the SIX gene family described in the literature; the identification of SIX6 missense variants in our POAG dataset; and the results of the in vivo and in vitro assays— that SIX6 is the most likely POAG susceptibility gene in this region. We identified both common and rare coding variants that alter the function of the SIX6 protein. We also identified a novel variant within the SIX6 enhancer that appears to disrupt the regulation of SIX6 expression, suggesting both regulatory and coding variants may influence POAG susceptibility at this locus. Finally, we identified hitherto unknown clinical features in POAG patients that may be dependent upon SIX6 genotype: patients homozygous for the SIX6 risk allele have a significantly thinner retinal nerve fiber layer than patients homozygous for the SIX6 non-risk allele.
The common variant, rs33912345 (Asn141His), which we showed has significantly reduced function in an in vivo model, was associated with POAG in our dataset and is in strong linkage disequilibrium with the originally identified GWAS SNP, rs10483727 [4]. This conserved amino acid is located within the alpha helix structure of the DNA homeobox domain of SIX6. Interestingly, the ancestral allele (C, His) is associated with POAG risk. The allele frequency of this variant differs markedly among populations (C allele frequency: YRI (0.99), ARF (0.96), ASN (0.76), and CEU (0.42); from 1000 Genomes release 14) [20]. This locus is associated significantly with an increase in vertical cup-disc ratio (VCDR) in population controls, indicating that it may be involved in the development of the optic nerve. VCDR is also a clinical measure used to track disease progression in POAG patients [5], [6], [9], [13], [21]. We note that African populations have larger VCDR and an increase in overall POAG prevalence [22]–[24] compared to CEU populations; in a recent study, the prevalence of POAG in Ghana exceeded 17%, the highest rate observed anywhere in the world [25]. While rs33912345 is not associated with POAG risk in a West African POAG case/control dataset from Ghana, the frequency of the ancestral (risk) allele is 99% in both cases and controls [26]. We hypothesize that differences in the structure of the optic nerve and the higher risk of POAG in individuals with African ancestry may, in part, arise from the fact that essentially all individuals in this population are homozygous for the rs33912345 ancestral risk allele.
Given the association of the SIX6 locus with neural retinal measurements such as VCDR, it is reasonable to hypothesize causal variants may function by inducing quantitative changes in cell populations in the retina. There is extensive evidence that SIX6 regulates early retinal progenitor cell proliferation during eye development [18], [19], [27]. Li et al. showed that Six6−/− mice display varying degrees of retinal hypoplasia that is due to a decrease in retinal ganglion cell proliferation arising from an early exit from the cell cycle during development, and results in a 20% decrease in the number of retinal ganglion cells by P35 [18]. This is consistent with functional studies of XOptx2, the Xenopus ortholog of SIX6 [19]. We have shown a reduction in eye size and in the volume of the optic nerve upon MO knockdown of zebrafish six6a, and we were able to rescue these phenotypes with co-injection of the human SIX6 non-risk allele, demonstrating that the zebrafish six6a gene is the likely functional ortholog to human SIX6. We identified five alleles that could not rescue the small eye and optic nerve phenotypes, and we observed a reduction in retinal nerve fiber layer thickness in POAG patients homozygous for the His141 SIX6 risk allele. Taken together with previously published findings, our results suggest that risk variants in human SIX6 increase POAG susceptibility by negatively affecting retinal ganglion cell development, likely leading to a reduction in the number of retinal ganglion cells in adulthood. Given that retinal ganglion cells are lost during the normal aging process, we speculate that this rate of loss could be increased by the presence of additional POAG risk alleles or other risk factors such as increased intraocular pressure [28]. The development of glaucomatous optic neuropathy and associated visual field loss would thus be hastened by a reduction in the initial number of retinal ganglion cells that an individual possesses. Future work will test the possibility that SIX6 variants also alter the rate of RGC death in the adult.
In summary, we have identified multiple common and rare SIX6 sequence variants in POAG cases, and used in vivo and in vitro assays to demonstrate that these variants have functional consequences on SIX6 expression and protein function. While other risk factors may be required for the onset of POAG, our data suggest that attenuation of SIX6 protein function increases an individual's susceptibility to developing the disease via changes to retinal development. Additional work is needed, possibly through the use of transgenic animal model studies, to fully understand the role of SIX6 in POAG.
This research was approved by the Institutional Review Board of Duke University Medical Center and adheres to the tenets of the Declaration of Helsinki.
Study subjects were unrelated patients from the Duke Eye Center and, after a comprehensive eye examine, were classified as either POAG cases or controls. POAG cases presented with glaucomatous optic neuropathy, defined as a cup-to-disc ratio greater than 0.7 and visual field loss in at least one eye. Patients with secondary forms of glaucoma or a history of ocular trauma were excluded from the study. POAG controls had no evidence of optic neuropathy, normal intraocular pressure (less than 22 mmHg in both eyes), and normal visual fields, assessed using standard automated perimetry.
Genomic DNA was extracted from patient blood samples using the PureGene chemistry following the manufacturer's standard protocol (Gentra, Minneapolis, MN). The coding portions (2 exons) of the SIX1 and SIX6 genes were sequenced in 518 Caucasian POAG cases and controls (262 cases, 256 controls) using a polymerase chain-reaction (PCR) containing 1× Qiagen PCR buffer (Tris·Cl, KCl, (NH4)2SO4, 15 mM MgCl2; pH 8.7); 200 µM each of dATP, dCTP, dGTP, and dTTP; 0.4 µM forward PCR primer; 0.4 µM reverse PCR primer; 3 µL of betaine, 10 ng genomic DNA; and 0.5 U HotStarTaq DNA polymerase (Qiagen, Venlo, Limburg) to a final volume of 25 µL. Primer sequences are available in Supplemental Table S3. The PCR was performed using a touchdown protocol (incremental lowering of annealing temperature) using the following thermocycler conditions: 94°C for 30 s, 65°C for 30 s, 72°C for 30 s with a 2°C decrease in the annealing temperature every two cycles until a final annealing temperature of 55°C was reached. The retinal specific SIX6 enhancer, previously described [12], was amplified a using a touchdown protocol with a final annealing temperature of 57°C. PCR products were sequencing using the BigDye chemistry on a 3730 DNA Analyzer (Applied Biosystems, Grand Island, NY).
The common missense single nucleotide polymorphism (SNP), rs33912345, was genotyped in the Duke POAG case-control dataset consisting of 482 POAG cases and 433 POAG controls using a TaqMan allelic discrimination assay according to the standard protocols from the manufacturer (Life Technologies, Grand Island, NY). For quality control purposes the following criteria were met: >95% genotyping efficiency, matching sample duplicates (two Centre d'Etude du Polymorphisme Humain samples per 96-well plate whose genotype data matched across all plates), and Hardy-Weinberg equilibrium assumptions. We tested for association of rs33912345 with POAG using an additive logistic regression model adjusted for age and sex using SAS [29].
Genome-wide genotype data were available from the NEI Glaucoma Human Genetics Collaboration (NEIGHBOR) and the Glaucoma Genes and Environment (GLAUGEN) consortia [4]. Chromosome 14 was imputed using IMPUTE2 (http://mathgen.stats.ox.ac.uk/impute/impute_v2.html) with a global 1000 Genomes reference panel. We tested SNPs at the SIX1/SIX6 locus for association with POAG using an additive logistic regression model adjusted for age, sex, and four principal components (NEIGHBOR) or age, gender, study site, DNA extraction method, DNA specimen type and principal components 1–6 (GLAUGEN) implemented in PLINK and visualized using LocusZoom (http://csg.sph.umich.edu/locuszoom/) [30], [31]. A meta-analysis was performed in Plink using a random effects model. Linkage disequilibrium was calculated and visualized using Haploview (http://www.broadinstitute.org/scientific-community/science/programs/medical-and-population-genetics/haploview/haploview) [32].
Optical coherence tomography (OCT) measurements of retinal nerve fiber layer (RNFL) by Spectralis (Heidelberg Engineering, Carlsbad, CA) spectral domain) and fundus photography were available from the Duke Eye Center. OCT images are not routinely performed in patients without ocular disease, so there was limited data available for controls. Therefore, the analysis was restricted to POAG cases homozygous for rs33912345. Thirty patients had both OCT measurements and SIX6 genotype data. RNFL thickness, age at POAG diagnosis, and the age at OCT measurement were compared between individuals homozygous for the risk or non-risk allele using a Student's t-test. Analyses were performed in SAS [29].
A vector containing human SIX6 was purchased from the CCSB Human ORFeome Collection that uses the Gateway technology system (Open Biosystems and Life Technologies, Grand Island, NY). SIX6 alleles identified by sequencing (Glu93Gln, Glu129Lys, Asn141His, Leu205Arg, Thr212Met, Ser242IIe), were created using the QuikChange II site-directed mutagenesis kit and protocols provided by the manufacturer (Agilent Technologies). SIX6 mRNA was in vitro transcribed using mMESSAGE mMACHINE SP6 Kit (Ambion, Life Technologies, Grand Island, NY). Translation blocker (TB) morpholinos against six6a (5′- CTGGAACATGGAGACTGTAATGTCT -3′) and six6b (5′ AATTGGCAACTGAAACATGAAGGCT 3′) were purchased from Gene Tools, LLC. Morpholino (2 ng) and mRNA (12.5 pg) were mixed and a volume of 0.5 nL was microinjected into each wild-type zebrafish embryo at one- to eight-cell stage as described previously (Stuart, McMurray et al. 1988).
Morphometric analyses of eye size were conducted on zebrafish embryos at 3 days post fertilization, using a Nikon SMZ 745T microscope. Zebrafish were anesthetized in embryo medium containing 0.2 mg/ml tricaine (Ethyl 3-aminobenzoate methanesulfonate, Sigma, E10521). Lateral view images were captured with Nikon DS-Fi1 camera, and the size of eye was measured with Nikon NIS-Elements AR software. Analysis of the optic nerve was performed on 2 dpf embryos fixed in Dent's fixative (80% Methanol and 20% DMSO) overnight and stained with acetylated-tubulin (Sigma; T7451). Heads were isolated from stained embryos and oriented with the ventral aspect facing a coverslip on microscope cover glass. Image acquisition was performed on a Zeiss 710 inverted confocal microscope and ∼100 um optical sections were obtained and reconstructed. Volumetric measurements were calculated using Imaris software and 7.5 um×7.5 um×15 um ROIs along the optic nerve were analyzed between each condition. ROIs were restricted to portions of the optic nerve wherein all neurite processes coalesced to form the major aspect of the nerve.
SIX6 enhancer alleles were tested using the dual-luciferase reporter assay system (Promega, Madison, WI). An experimental construct containing a minimal promoter (pGL4.23, firefly luciferase, Promega) was used to test the functional effect of the enhancer alleles identified by sequencing (Chr14:60974363_C, Chr14:60974373_T, Chr14:60974378_T, Chr14:60974400_A, Chr14:60974449_G) in the POAG case/control dataset. The experimental constructs (pGL4.23+Enhancer) were generated using a nested PCR protocol; the XhoI and HindII enzymes; the Quick Ligation kit (New England BioLabs, Ipswich, MA); and the QuikChange II site-directed mutagenesis kit (Agilent Technologies), following protocols provided by the manufacturers. Constructs were confirmed to be correct by sequencing. Hek293 cells were cultured according to the supplier's suggestions (ATCC, Manassas, VA). As described by Conte et al., co-transfection with NeuroD and E47 is required for SIX6 enhancer activation [12]. Therefore, cells were co-transfected with an experimental vector (pGL4.23+Enhancer), a control vector (pGL4.74, renilla luciferase, Promega), and vectors containing NeuroD and E47 (provided by the Center for Human Disease Modeling, Duke University) using a standard calcium phosphate transfection protocol. The experiment was performed three times in triplicate and the results were analyzed using the dual luciferase reporter (DLR) ratio (firefly luciferase sum: renilla luciferase sum) normalized by the reference SIX6 enhancer included on every plate. The data were analyzed using an ANOVA, adjusted for batch, and linear contrasts were used to determine the effect of each vector. Statistical analyses were performed in SAS [29].
|
10.1371/journal.pntd.0001404 | Toward an Open-Access Global Database for Mapping, Control, and Surveillance of Neglected Tropical Diseases | After many years of general neglect, interest has grown and efforts came under way for the mapping, control, surveillance, and eventual elimination of neglected tropical diseases (NTDs). Disease risk estimates are a key feature to target control interventions, and serve as a benchmark for monitoring and evaluation. What is currently missing is a georeferenced global database for NTDs providing open-access to the available survey data that is constantly updated and can be utilized by researchers and disease control managers to support other relevant stakeholders. We describe the steps taken toward the development of such a database that can be employed for spatial disease risk modeling and control of NTDs.
With an emphasis on schistosomiasis in Africa, we systematically searched the literature (peer-reviewed journals and ‘grey literature’), contacted Ministries of Health and research institutions in schistosomiasis-endemic countries for location-specific prevalence data and survey details (e.g., study population, year of survey and diagnostic techniques). The data were extracted, georeferenced, and stored in a MySQL database with a web interface allowing free database access and data management.
At the beginning of 2011, our database contained more than 12,000 georeferenced schistosomiasis survey locations from 35 African countries available under http://www.gntd.org. Currently, the database is expanded to a global repository, including a host of other NTDs, e.g. soil-transmitted helminthiasis and leishmaniasis.
An open-access, spatially explicit NTD database offers unique opportunities for disease risk modeling, targeting control interventions, disease monitoring, and surveillance. Moreover, it allows for detailed geostatistical analyses of disease distribution in space and time. With an initial focus on schistosomiasis in Africa, we demonstrate the proof-of-concept that the establishment and running of a global NTD database is feasible and should be expanded without delay.
| There is growing interest in the scientific community, health ministries, and other organizations to control and eventually eliminate neglected tropical diseases (NTDs). Control efforts require reliable maps of NTD distribution estimated from appropriate models and survey data on the number of infected people among those examined at a given location. This kind of data is often available in the literature as part of epidemiological studies. However, an open-access database compiling location-specific survey data does not yet exist. We address this problem through a systematic literature review, along with contacting ministries of health, and research institutions to obtain disease data, including details on diagnostic techniques, demographic characteristics of the surveyed individuals, and geographical coordinates. All data were entered into a database which is freely accessible via the Internet (http://www.gntd.org). In contrast to similar efforts of the Global Atlas of Helminth Infections (GAHI) project, the survey data are not only displayed in form of maps but all information can be browsed, based on different search criteria, and downloaded as Excel files for further analyses. At the beginning of 2011, the database included over 12,000 survey locations for schistosomiasis across Africa, and it is continuously updated to cover other NTDs globally.
| More than half of the world's population is at risk of neglected tropical diseases (NTDs), and over 1 billion people are currently infected with one or several NTDs concurrently, with helminth infections showing the highest prevalence rates [1], [2]. Despite the life-long disabilities the NTDs might cause, they are less visible and receive lower priorities compared to, for example, the ‘big three’, that is malaria, tuberculosis, and HIV/AIDS [3], [4], because NTDs mainly affect the poorest and marginalized populations in the developing world [3], [5], [6]. Efforts are under way to control or even eliminate some of the NTDs of which the regular administration of anthelmintic drugs to at-risk populations – a strategy phrased ‘preventive chemotherapy’ – is a central feature [7]–[11].
There is a paucity of empirical estimates regarding the distribution of infection risk and burden of NTDs at the national, district, or sub-district level in most parts of the developing world [12]–[16]. Such information, however, is vital to plan and implement cost-effective and sustainable control interventions where no or only sketchy knowledge on the geographical disease distribution is available. There is a risk of missing high endemicity areas and distributing drugs to places which are not at highest priority, hence wasting human and financial resources. Consequently, integrated control efforts should be tailored to a given epidemiological setting [14].
The establishment of georeferenced databases is important to identify areas with no information on disease burden, to foster geographical modeling over time and space, and to control and monitor NTDs. In 1987 the bilingual (English and French) ‘Atlas of the Global Distribution of Schistosomiasis’ was published, which entailed country-specific maps of schistosomiasis distribution based on historical records, published reports, hospital-based data, and unpublished Ministry of Health (MoH) data [17]. While recent projects like the Global Atlas of Helminth Infections (GAHI; http://www.thiswormyworld.org) [18] and the Global Atlas of Trachoma (http://trachomaatlas.org) [19] offer maps on the estimated spatial distribution of soil-transmitted helminthiasis, schistosomiasis, and trachoma prevalence, they do not provide the underlying data for further in-depth analyses conducted by different research groups. An open-access global parasitological database for NTDs, which provides the actual data, is not available.
The Swiss Tropical and Public Health Institute (Swiss TPH) in Basel, Switzerland, together with partners from the University of Copenhagen, Denmark, and the University of Zambia (UNZA) in Lusaka, Zambia, were working together in a multidisciplinary project to enhance our understanding of schistosomiasis transmission (the CONTRAST project) [20], [21]. One of the CONTRAST goals was to create a data repository on location-specific schistosomiasis prevalence surveys in sub-Saharan Africa. In this manuscript, we describe the steps taken toward the development of such an open-access schistosomiasis database which is currently expanded to a global scale and to include other NTDs (e.g., soil-transmitted helminthiasis and leishmaniasis) and that can be constantly updated based on new publications and reports, as well as field data provided by contributors.
We selected schistosomiasis as the first disease to establish a proof-of-concept and populate our global NTD database. Indeed, schistosomiasis affects over 200 million people worldwide, with more than 95% concentrated in Africa. Both urinary schistosomiasis (caused by the blood fluke Schistosoma haematobium) and intestinal schistosomiasis (causative agents: S. mansoni and S. intercalatum) are endemic in Africa [22], [23].
In order to obtain a large number of geographical locations to which prevalence data can be attached to our database, we conducted a systematic review. The specific steps of the process from identification of relevant surveys to data entry in the database, including various data sources, search criteria, data extraction and entry procedures, and quality control measures, are visualized in Figure 1, and will be described in more detail in the following sections.
We systematically searched the following electronic databases with no restriction to date and language of publication: PubMed [24], ISI – Web of Knowledge [25], and African Journal Online [26]. Using specific search terms, we retrieved relevant peer-reviewed publications with an emphasis on schistosomiasis prevalence data in Africa.
The keywords applied for our literature search on schistosomiasis in the electronic databases, as well as the terms for the future search strategy on other NTDs, usually consists of species names and disease expressions often abbreviated and supplied with an asterisk in order not to miss out any results due to the variety of different spellings. The search strategy can be generalized as follows: country name OR continent AND disease (alternative spellings were included). These keywords were combined with names of African countries, whereas also alternative or former country names were considered to have our search strategy as broad as possible. This approach enabled us to save literature search results on a country-by-country basis.
Along with articles from peer-reviewed journals, reports from health institutions (e.g., World Health Organization (WHO) and the Office de la Recherche Scientifique et Technique d'Outre-Mer (ORSTOM)/Organisation de Coordination et de Coopération pour la Lutte contre les Grandes Endémies (OCCGE)) and doctoral theses (so-called ‘grey literature’) compose an important literature source for schistosomiasis prevalence data. Grey literature is often restricted to internal use or is not available in an electronic format. Publication databases available from WHO [27] and the Institute de la Recherche pour le Développement (IRD, formerly ORSTOM) [28] offer at least partial access to such documents. Additional grey literature included was obtained directly via site visits by team members to African universities and health research and development institutions. Another important source for survey data is the direct communication with local contacts, i.e., collaborators and partners from different African countries, individual researchers, and staff from ministries of health. The majority of entries that can be retrieved in the database were extracted from peer-reviewed journals (46%), however 30.5% of the data was obtained from personal communication with authors and 23.5% from grey literature. The latter was usually more extensive in terms of survey locations than the former sources. Since the key terms we used for our systematic review were mainly species and abbreviated disease names (e.g., ‘schisto*’ and ‘bilharz*’) that are not language specific, we also extracted and included reports written in languages other than English, including French (especially for West African countries), Portuguese, Italian, Dutch, Scandinavian and few in Russian and Chinese. Sources from literature and from personal communication were stored, labeled and managed with Reference Manager 11 [29].
Often, geographical information of the survey location was not given in the retrieved publications and reports (94%). Hence, we retrospectively georeferenced the locations. The majority of the coordinates was retrieved using the GEOnet Names Server (55%) [30], topographic or sketch maps (23%), and Google (14%) [31]. Personal communication with authors and local collaborators contributed another 5% of the retrospective geolocations, and only 3% were derived from other gazetteers and sources. Irrespective of the source of retrospective geolocation, we always mapped the coordinates in Google Maps to ensure that they are located in the study area and pointing to a human settlement. In general, we tried to adhere to the guidelines for georeferencing put forward by the MaNIS/HerpNEt/Ornis network to approach georeferecing in standardized manner [32].
All data sources obtained (literature, data from personal communication, and field visits) were screened for relevance by applying defined inclusion and exclusion criteria. Studies were included if they comprised prevalence data of schistosomiasis, identified either by school-based or community-based surveys. We accepted different study designs (cluster sampling, random sampling, stratified sampling, systematic sampling, etc.) as long as the reported findings could be considered as representative for the population or a specific sub-group of the population (e.g., school children, women, fishermen) in a given area. Along with schistosome prevalence data, a minimal set of information was collected, such as survey location (school, village, and administrative unit), date of survey, and number of individuals examined and found schistosome-positive (irrespective of sample size). In case additional survey-specific data were available, such as infection status according to age and sex, or intermediate host snail species (i.e., Bulinus spp. for S. haematobium and Biomphalaria spp. for S. mansoni), such information was tagged, as it might be of relevance for subsequent data extraction.
Hospital-based investigations, case-control studies, drug efficacy studies, and clinical trials, as well as reports on disease infection among travelers, military personnel, expatriates, nomads, and other displaced or migrating populations were excluded from the database in order to avoid non-representative samples (e.g., individuals with symptoms or disease-related morbidity) were excluded. Thus, the data in the database reflect the actual spatial distribution of the disease at a given time point. In case baseline prevalence data were reported in the aforementioned study types, or if former migrant populations settled down and the given survey location was clearly defined, data were included. Although having taken these precautionary steps, the database might still include prevalence data influenced by migration, since mobility and migration patterns of the rural population in sub-Saharan Africa are quite common [33], [34]. Based on our exclusion criteria, we rejected more than 70% of the articles retrieved from the literature search.
Once a source was identified as relevant, the data were extracted following a standard protocol with emphasize on (i) the source of disease data such as authorship, journal, publication date, etc.; (ii) description of the parasitological survey specifying the country, the survey date (year, month, season), and the type of survey (community- or school-based); (iii) survey location reported at the highest spatial resolution available; and (iv) parasitological survey data. If relevant source included malacological data, details on snail survey methods used, snail species collected, and infection rate of the Planorbidae were also extracted.
The Kato-Katz technique for S. mansoni and urine filtration for S. haematobium diagnosis are often considered as ‘gold’ standard methods [35]. If prevalence data were reported by different diagnostic methods, we only recorded in the database the results of the test with highest sensitivity and specificity. We applied the following ranking of diagnostic methods: (i) ‘gold’ standard; (ii) direct methods such as detection of eggs in urine/stool; and (iii) any other method such as antigen detection.
The data are stored and managed in a MySQL [36] relational database with a web-interface built in hypertext preprocessor (PHP) [37]. The process from prevalence data extraction to database entry is schematically depicted in Figure 1.
The database consists of six tables corresponding to the sections of data extraction. The system architecture supports two types of users: the administrators and the end-users [38]. Registered administrators can enter new data, edit, or delete existing entries under their username and password. In addition, administrators can temporarily mask confidential data as requested by authors contributing specific data. Then a summary measure is presented instead with the contact details of the data owner to enable direct communication between researchers. Users can search all records using different selection criteria, e.g., country, document category, disease, and journal. The user part was designed to fulfill the most common queries, e.g., all recorded data for a specific parasite species in a given country or region within a specified period. The user will be able to download all information stored in the database matching different search criteria in an Excel file through an export function.
To guarantee and improve data quality, the following measures have been taken. A first quality check is performed after data entry in the electronic database. For example, data extracted by assistants are always double-checked against the original source of information before becoming open-access, while data entries of senior staff are checked randomly. Data sent by contributors are inspected for completeness (e.g., in terms of study year and diagnostic technique), precise calculations (e.g., prevalence), and for correctness of coordinate information if provided. Additionally, we routinely screen the database for specific errors, i.e., by mapping survey locations and counterchecking whether the points are plotted in the expected area, by summarizing prevalence data per location and survey date to check for duplicate records, by testing for entry completeness.
Together with correctness of data extracted and entered, we also aim at the integrity of survey data. To further improve completeness of our database (e.g., date of surveys, disaggregated data) corresponding authors are contacted by e-mail asking for missing information. Approximately half of all reports had missing information, and so far we were able to get in touch with more than a third of the authors. Finally, missing coordinates for specific survey locations were obtained by re-checking additional maps and gazetteer sources, by communication with authors, and by employing global positioning system (GPS) databases created by collaborators during field visits for specific countries (i.e., Uganda, Zambia).
On 10 January 2011, our database contained 12,388 survey locations for schistosomiasis that are georeferenced from 35 African countries and 568 data points on intermediate host snails for 20 African countries, giving information on 25 different mollusk species. The database is constantly updated and subjected to quality control as the project moves along. Surveys are dated as early as 1900 and the historical references that are part of the Doumenge et al. (1987) [17] global schistosomiasis atlas are included by extracting data from the original source files. Since our main focus was on sub-Saharan Africa, the data currently included in the database covers all Western, Eastern, Middle and Southern African countries, according to UN Population Division classification [39]. Data extraction for Northern African countries is currently in progress. Survey coverage between countries shows considerable variation. Typically, larger numbers of survey locations were found in higher populated countries, but the amount of surveys also depends on existing national control or monitoring programs. In addition, temporal and spatial gaps in the survey distribution (as observed in Liberia, Rwanda, and Sierra Leone) might have occurred due to political instability and financial problems. The most widely used method for the diagnosis of intestinal schistosomiasis in the surveys that were fed into our repository is the Kato-Katz technique (76.7%, as single method or in combination). Stool concentration techniques accounted for a total of 13.3% (e.g., Ritchie/modified Ritchie technique (6.0%), concentration in ether solution (5.0%), merthiolate-iodine-formaline (MIF) concentration method (2.3%)) [40]. With regard to S. haematobium diagnosis, microscopic examination of urine after concentration (82.0%) such as urine filtration, urine centrifugation, and urine sedimentation, as well as reagent strip testing (12.8%) for the detection of blood in urine (i.e., microhematuria) or a combination of both approaches (2.3%) are most commonly employed.
Most of the surveys currently included in our database focus on school-aged children (70.1%), whereas less than a third (29.9%) of the surveys include all age groups. Furthermore, among the prevalence data of schistosomiasis collected, S. haematobium (54.4%) and S. mansoni (40.8%) were the most prevalent species. The third schistosome species parasitizing humans in Africa, S. intercalatum (4.8%), was only reported in surveys carried out in Cameroon and Nigeria, confirming that this species is restricted to some parts of West and Central Africa (Figure 2). Additionally, two zoonotic Schistosoma species were reported, namely S. bovis (0.02%) and S. matthei (0.01%), in the first cattle being the reservoir, while the latter is naturally affecting different antelope species (Table 1). Co-occurrence of multiple species was reported in 20.4% of the surveys, the majority of which (97.6%) was S. mansoni-S. haematobium co-occurrence. Currently, two schistosomiasis datasets in the GNTD database are confidential and about 100 datasets still await quality control. Hence, these data were masked and cannot yet be accessed by the database users.
The distributions of S. mansoni and S. haematobium are shown in Figures 3 and 4, respectively. The applied prevalence cut-offs of 10% and 50% were chosen based on WHO recommendations to distinguish between low (<10%), moderate (between 10 and 50%) and high (≥50%) endemicity communities [35]. The compiled survey data in the database suggest that S. mansoni predominates in East Africa, whereas S. haematobium prevalence is higher than S. mansoni in many African countries.
Data repositories are important tools for the development and validation of data-driven models to estimate the distribution and burden of NTDs, such as for malaria [41], [42]. Model-based predictions based on the compiled survey data will facilitate mapping of disease endemicity in areas without data and spatially explicit targeting of control interventions and long-term surveillance. With regard to NTDs, progress has been made for helminthic diseases [18] and trachoma [19]. The information included in a database helps to identify where current information is missing, request feedback from endemic countries, and initiate the collection of new data at those areas. Here, we described our efforts toward the establishment of an open-access database for NTDs. The database (http://www.gntd.or) allows for subsequent mapping of the observed survey data in order to identify high risk areas and to produce smooth risk maps, as exemplified by Schur et al. (2010) [43].
The work presented here and the issue of open-access in relation to data, information sharing, and services, is not a new one. Indeed, we are following the successful implementations in different fields, e.g., open-access publishing (e.g., Public Library of Sciences (PLoS) and BioMed Central (BMC) journals), PubMed, genomic data [44]–[46], biodiversity [47], drug trial results [48], [49], and entertainment technologies [50].
With regard to epidemiological research, mapping disability, mortality, and disease burden due to infectious diseases, two recent open-access georeferenced epidemiological databases include the Mapping Malaria Risk in Africa (MARA), which is reporting malaria survey data in Africa dating back to 1900 [42], [51], and the Malaria Atlas Project (MAP) [41], which provides maps of raw and model-based estimates of malaria risk at a global scale and country level. Other examples are the WorldWide Antimalarial Resistance Network (WWARN) [52], [53], the MosquitoMap, a geospatially referenced clearinghouse for mosquito species collection records and distribution models [54], and the Disease Vectors Database [55], which is a georeferenced database on the presence of vector species of Chagas disease, dengue, leishmaniasis, and malaria. The GAHI project created a database of schistosomiasis and soil-transmitted helminthiasis survey data [13], [18], similar to our GNTD database, with the goal to provide open-access information on the global disease distribution and to highlight areas requiring mass drug administration. While the GAHI project focuses on mapping country-specific disease risk estimates, the GNTD database provides open-access to the mainly location-specific survey data. Free access to the data enables the users to conduct analyses for their own purposes. The existence of both databases offers the opportunity to join forces and to move forward in a unified way. As a first step it would be interesting to validate the two existing databases, align and harmonize them into a single comprehensive data repository, and discuss ways of harnessing synergies. Involvement of partners at WHO and other organizations will be essential.
Despite the benefits of free and public data repositories, data sharing is a challenge. Data owners may hesitate to provide their data, especially when they have not yet been published. However, confidential data can be masked through a special database feature as explained in the Methods section. As more and more data are included into the GNTD database, the current lack in the geographical extent of location-specific survey data across countries and regions will become less critical. Undoubtedly, a host of valuable information exists within countries, in the form of unpublished local archived sources. Efforts are ongoing to access this information with the help of our in-country scientific partners in ministries of health and research institutions by visiting the countries of interest to strengthen and further expand our global network of collaborators. Nevertheless, it is likely that there will remain significant areas with scarce data because no surveys have been conducted or data are not readily accessible or have been lost in the face of civil war, political unrest, or inappropriate archiving procedures. Such geographical lacks of survey data might be only known to local experts while the international community might not be aware.
Data from systematic literature searches or unpublished reports may contain different levels of reliability. For instance, snail identification is complex and without the guidance of experienced morphologists incorrect results may be reported. The quality of diagnostic methods must also be improved, for example through repeated stool and urine sampling over several consecutive days, since schistosome egg-output varies from one day to another. Unfortunately, only few surveys adopted such rigorous diagnosis due to generally limited financial and human resources [56], [57]. Furthermore, historical surveys differ in study design and are heterogeneous in terms of the age groups considered, the diagnostic methods applied, and the survey dates. Heterogeneity is also present in the way data are reported. For example in the past, numerous studies often aggregated their results at province or district level [58], [59], while currently information are frequently provided or shared at village or even individual level [60], [61]. All these points form important limitations of database compilations of epidemiological data. However, data are as limited as the sources from which they were derived. Developing standard NTD survey protocols, will enhance data comparability in the future [62].
Georeferencing historical surveys are not a straightforward undertaking. We have used a number of different sources to geolocate surveyed locations, the most common ones were described in the Data Sources section. However, several villages may have the same name within a single country. In such cases, information regarding the administrative boundaries of the village or its distance from nearby rivers, lakes, or towns is essential. A further complication is that administrative boundaries as well as region and district names may change over time. For instance, in Uganda, 23 new districts have been created in 2005 and 2006 [63].
In order to maintain high quality of the database, the entries are checked continually using systematic screening approaches as described in more detail in the Methods section. Additionally, we aim to further complement gaps (on date of survey, geographical coordinates, age group, number of people examined, etc.) and to obtain disaggregated survey data by contacting authors or collaborators, and by cross-checking new sources (maps, databases, and grey literature).
Our database is a global, freely-available, public, online resource, which hosts information pertaining to the distribution of NTDs. At present, the database contains more than 12,000 survey locations with emphasis on schistosomiasis prevalence data across Africa. It is currently expanded with information on soil-transmitted helminthiasis from Latin American and Southeast Asian countries. Our short-term goal is to extend the database from schistosomiasis to include other NTDs (i.e., ascariasis, hookworm disease, trichuriasis, lymphatic filariasis, onchocerciasis, and trachoma). Future versions of the database will supplement prevalence information from other NTDs (Buruli ulcer, Chagas disease, cysticercosis, dracunculiasis, leishmaniasis, leprosy, and human African trypanosomiasis). The approach for inclusion of further NTDs, as well as the search strategy that is going to be applied, will be the same as described in this article. We are aware that data on soil-transmitted helminthiasis is often given alongside intestinal schistosomiasis data and could have been extracted simultaneously. However, the database evolved from the CONTRAST project that focused on schistosomiasis. While screening for schistosomiasis, we labeled relevant references on other NTDs in our reference database, which will speed up future work steps, such as literature review and data extraction of relevant sources.
The structure of the database allows entering not only parasitological data, but also other attributes, like geospatially referenced data on the disease vectors. At present, our database has limited malacological survey information, and it does not include historical collections, however, we plan to add the georeferenced historical collection compiled by the Mandahl-Barth Centre for Biodiversity and Health in Copenhagen, Denmark, which holds information on about 7,000 georeferenced snail samples.
Our hope is to provide to scientists and policy makers, a user-friendly and useful platform which is continuously updated in order to facilitate data sharing, and retrieval of disease surveillance and epidemiological data. We welcome contributions from other researchers in possession of prevalence data from various NTDs. Users may contribute by download the template offered after registration and providing the required information. An administrator checks the data for quality and sends a confirmation e-mail before including the data in the database. Researchers who may not wish to share their data may only provide limited information about the data they possess (survey location, year, and amount of data) so that the database becomes a library of potential data sources. Furthermore, we plan to add an option for the GNTD database users to contact and interact with the contributors by providing a ‘send e-mail to contributor’ function.
Another immediate goal is to develop a web-based interface, which will combine raw disease data and spatial model-based estimates of disease burden at different geographical levels with country boundaries and geophysical information. The results will be accessed in geo-referenced kml format, which is displayed automatically on a Google Earth interface on the website [64]. This will allow users to obtain estimates of disease burden at different spatial resolutions (village, district, region, country, etc.) and to display model predictions including prediction uncertainties and raw data on the map.
A more distant option is to allow end-users to upload their own data, for instance regional and community-based health practitioners could directly upload disease prevalence to the MySQL database using hand-held smart phones with GPS functionality [65]. Success of the project will depend on active collaboration and contribution of researchers and disease control managers from around the world. We hope that our efforts will be recognized as a helpful tool contributing to the control and eventual elimination of NTDs.
|
10.1371/journal.pgen.1005523 | An Interaction between RRP6 and SU(VAR)3-9 Targets RRP6 to Heterochromatin and Contributes to Heterochromatin Maintenance in Drosophila melanogaster | RNA surveillance factors are involved in heterochromatin regulation in yeast and plants, but less is known about the possible roles of ribonucleases in the heterochromatin of animal cells. Here we show that RRP6, one of the catalytic subunits of the exosome, is necessary for silencing heterochromatic repeats in the genome of Drosophila melanogaster. We show that a fraction of RRP6 is associated with heterochromatin, and the analysis of the RRP6 interaction network revealed physical links between RRP6 and the heterochromatin factors HP1a, SU(VAR)3-9 and RPD3. Moreover, genome-wide studies of RRP6 occupancy in cells depleted of SU(VAR)3-9 demonstrated that SU(VAR)3-9 contributes to the tethering of RRP6 to a subset of heterochromatic loci. Depletion of the exosome ribonucleases RRP6 and DIS3 stabilizes heterochromatic transcripts derived from transposons and repetitive sequences, and renders the heterochromatin less compact, as shown by micrococcal nuclease and proximity-ligation assays. Such depletion also increases the amount of HP1a bound to heterochromatic transcripts. Taken together, our results suggest that SU(VAR)3-9 targets RRP6 to a subset of heterochromatic loci where RRP6 degrades chromatin-associated non-coding RNAs in a process that is necessary to maintain the packaging of the heterochromatin.
| Cells regulate the packaging of DNA in the chromatin, and an important process in the development of any eukaryote is the definition of chromatin states. Heterochromatin is a condensed form of chromatin that is usually silent. Short non-coding RNAs participate in the silencing of transposons in animal germ cells and in the establishment of heterochromatin states during early development. These non-coding RNAs guide histone methyltransferases to the histones in the chromatin, which in turn creates binding sites for other factors that keep the heterochromatin condensed. The non-coding RNAs that participate in the establishment of heterochromatic domains are active in germ cells and gonads, and other mechanisms must exist in animal somatic tissues to maintain the established patterns of heterochromatin throughout development. Here, we identify RRP6 as a protein that is necessary for maintaining the condensed state of heterochromatin in a subset of heterochromatic loci in the somatic fruitfly genome. RRP6 is a ribonuclease that plays many roles in RNA processing and in quality control of gene expression. We show that RRP6 is tethered to heterochromatin through an interaction with a histone methyltransferase, and that in the heterochromatin RRP6 acts on transcripts derived from repetitive sequences that need to be degraded to maintain the packaging of the heterochromatin. The importance of ribonucleases for the structure of the heterochromatin had been shown in yeast cells. Our findings show that RNA degradation participates in chromatin silencing also in animal cells.
| Approximately 30% of the genome of Drosophila melanogaster is heterochromatic and is made up of transposons, transposon fragments and repetitive sequences with different degrees of complexity [1]. The heterochromatin contains high levels of heterochromatin-specific proteins, such as Heterochromatin Protein 1a (HP1a), and is enriched in certain patterns of post-translational modifications of the histone tails [2], [3]. Heterochromatin formation involves a cascade of histone modifications that are targeted to specific regions of the genome by complex protein-protein and protein-nucleic acid interactions. In the switch from euchromatin to heterochromatin, acetylated H3K9 (H3K9ac) is deacetylated by histone deacetylases such as RPD3/HDAC1. H3K9 is subsequently methylated by histone methyltransferases, and the methylated H3K9 (H3K9me) acts as a binding site for HP1a [3], [4]. The properties of the heterochromatin can spread along the chromatin fiber, and HP1a plays a central role in this process. The ability of HP1a to dimerize, to interact with the methyltransferase SU(VAR)3-9, and to bind H3K9me provides the basis for the spreading of heterochromatin [5]. An additional level of complexity in the establishment of heterochromatic states is provided by the fact that HP1a can also bind RNA in both D. melanogaster [6] and Schizosaccharomyces pombe [7]. Recent studies on Swi6, the HP1a ortholog of S. pombe, have shown that the interaction of Swi6 with RNA interferes with the binding of Swi6 with H3K9me [7].
Small non-coding RNAs are essential components of the regulation of chromatin packaging in different organisms [8]. Fission yeast uses siRNAs to silence heterochromatic sequences through the recruitment of the H3K9 methyltransferase Clr4 [4], [9]. RNAi-dependent mechanisms of heterochromatin assembly exist also in plants, where siRNAs direct de novo DNA methyltransferases to specific genomic sequences (reviewed in [10]). Animal cells use instead the piRNA pathway to trigger heterochromatin assembly and transposon silencing in the germ line. In D. melanogaster, non-coding RNAs transcribed from transposon-rich regions are processed into piRNAs, and a “Piwi-piRNA guidance hypothesis” has been recently proposed for the recruitment of SU(VAR)3-9 and HP1a to heterochromatin [11], [12], [13], [14]. The Piwi-piRNA system is active during early development and it directs the initial establishment of heterochromatin states not only in the germ line but also in somatic cells. Recent studies suggest that after embryogenesis, the patterns of heterochromatin packaging are maintained through cell divisions via piRNA-independent mechanisms [15], [16].
An important player in the regulation of non-coding RNAs is the exosome, a multiprotein complex with ribonucleolytic activity [17], [18], [19]. In D. melanogaster, the core of the exosome associates with two catalytic active subunits, RRP6 and DIS3 [20]. In the cell nucleus, the exosome is involved in the processing of many non-coding RNAs, including pre-rRNAs, and in the quality control of mRNA biogenesis [21], [22]. The exosome ribonucleases also degrade a large variety of unstable, non-coding RNAs in various organisms including S. cerevisiae [23], plants [24], and animals [25], [26]. Moreover, recent studies have revealed that RRP6 participates in the regulation of enhancer RNAs [27] and in the degradation of unstable transcripts synthesized at DNA double-strand breaks [28].
The exosome has been functionally linked to the methylation of H3K9 in heterochromatin [29]. In S. pombe, RRP6 participates in the assembly of centromeric heterochromatin through an RNAi-independent mechanism [30], and collaborates with the RNAi machinery to silence developmentally regulated loci and retrotransposons [31]. Much less is known about the possible links between RRP6 and heterochromatin in animals. We have observed that a fraction of RRP6 is associated with heterochromatin in the genome of D. melanogaster, and we have identified physical interactions between RRP6 and several heterochromatin factors, including HP1a, SU(VAR)3-9, and RPD3. Our results show that SU(VAR)3-9 promotes the targeting of RRP6 to transposon-rich heterochromatic loci. In these loci, RRP6 contributes to maintaining the structure of the heterochromatin by degrading non-coding RNAs that would otherwise compromise the packaging of the chromatin.
We analyzed the localization of RRP6 in salivary gland polytene chromosomes by immunofluorescence (IF), and we used an antibody against HP1a as a marker for heterochromatin. RRP6 was associated with many bands throughout the chromosomes, but was also present in heterochromatin (Fig 1A–1C, [32]). Although the overall distributions of HP1a and RRP6 were very different from each other, both proteins overlapped in heterochromatic regions, for example at telomeres (Fig 1B) and chromocenter (Fig 1C).
We also analyzed the association of RRP6 with heterochromatin in S2 cells by IF (S1 Fig) and by immuno-electron microscopy (IEM, Fig 1D). The IEM analysis revealed that RRP6 and HP1a were located in close proximity in dense chromatin areas (Fig 1D).
In another series of experiments, we used S2 cells that expressed V5-tagged RRP6 (S2-RRP6-V5 cells) and we analyzed RRP6-V5 under low-induction conditions to avoid overexpression artefacts. We applied a cell fractionation scheme previously established by Tyagi et al. [33], and the proteins in each of the different nuclear fractions (soluble, chromosomal RNP and chromatin) were analyzed by Western blotting. We observed that a fraction of RRP6 is associated with chromatin (Fig 1E and 1F). In these experiments, the chromatin fraction was digested with RNase A before centrifugation, which suggests that the binding of RRP6 to the chromatin fraction is not mediated by RNA.
In a previous study, we carried out co-immunoprecipitation experiments aimed at identifying interaction partners for the nuclear exosome of D. melanogaster in RNase A-digested nuclear extracts [34]. The proteins that co-immunoprecipitated with RRP6-V5 were identified by high-performance liquid chromatography/tandem mass spectrometry (LC/MS-MS). We detected a total of 418 proteins associated, directly or indirectly, with RRP6 when we set the false discovery rate to 0.01 (S1 Table). Previously known exosome interactors were detected in our study, including other components of the exosome, the transcription elongation factors SPT5 and SPT6 [20], and the insulator protein CP190 [35] (Fig 2A and S1 Table).
We carried out a gene-ontology (GO) analysis with the RRP6 interactors. Many of the GO terms associated with the RRP6-interacting proteins were related to known functions of the exosome in D. melanogaster. Interestingly, the term “chromatin organization” was also significantly enriched (Fig 2B).
Two heterochromatin proteins in the list of interaction partners drew our attention: SU(VAR)3-9 and RPD3 (Fig 2A and S1 Table). HP1a was also detected in two out of three LC/MS-MS experiments. These three proteins are functionally related to each other, and their interactions with RRP6 suggest a functional link between heterochromatin and RRP6.
We performed co-immunoprecipitation experiments followed by Western blotting to validate the interactions between RRP6 and heterochromatin factors found in the LC/MS-MS. In a first series of experiments, we used the S2-RRP6-V5 cells to validate the interaction of the V5-tagged RRP6 with RPD3 and HP1a (Fig 2C and 2D). The interaction between SU(VAR)3-9 and RRP6 was analyzed using the cell line that expressed RRP6-V5 and HA-tagged SU(VAR)3-9 (Fig 2E and S2A Fig). These experiments confirmed that RRP6 interacts, directly or indirectly, with RPD3, SU(VAR)3-9 and HP1a.
In agreement with the interactions reported above, the distribution of RRP6-V5 overlapped with that of SU(VAR)3-9 as shown by IF (S2B Fig), and a proximity ligation assay (PLA) confirmed the close association of SU(VAR)3-9 with RRP6 in situ (S2C Fig). The distributions of RRP6 and SU(VAR)3-9 in nuclear fractions were also very similar to each other (S2D Fig).
We depleted S2 cells of RRP6 by RNA interference (RNAi), and we carried out RNA-seq analysis to determine whether RRP6 plays a role in the expression of heterochromatic sequences. Total RNA preparations from cells treated with dsRNA complementary to either Rrp6 (RRP6 cells) or GFP (control GFP cells) were ribosome-depleted and reverse transcribed using random primers, and the resulting cDNAs were sequenced to a depth of over 30 M reads per sample. The experiments included two independent biological replicates. The RRP6 levels were markedly reduced in the RRP6 cells, as expected (S3A Fig).
The analysis of RNA levels in the control GFP cells revealed the existence of significant expression over a large fraction of the genome, including heterochromatic regions that are rich in repetitive sequences, transposons and transposon fragments (Fig 3). The depletion of RRP6 did not destabilize the transcriptome on a global scale (S4A Fig), but affected the levels of expression of different types of transcripts in good agreement with the results from Kiss and Andrulis [36], Graham et al. [37] and Lim et al. [35]. The fraction of reads that mapped to intergenic sequences was significantly increased in RRP6 cells (Fig 3A, P < 0,0001), which is consistent with the role of RRP6 in the degradation of a large variety of non-coding and pervasive transcripts. Depletion of RRP6 caused an increase in the level of non-coding RNAs that are processed by the exosome, such as pre-rRNAs and some snoRNAs (S3B Fig). Depletion of RRP6 also increased the levels of transcripts derived from different types of heterochromatic repeats such as subtelomeric minisatellites and simple gagaa repeats (Fig 3B). Many transposons and transposon fragments showed increased transcript levels in RRP6 cells (Fig 3C, S4 Fig and S2 Table), including LTR retrotransposons, non-LTR retrotransposons and DNA transposons. However, not all transposons were affected (S4B Fig). Interestingly, we found that some elements of the same family showed increased or decreased transcript levels upon RRP6 depletion depending on their genomic insertion site (see for example 412 and jockey in S2 Table), which suggests that the genomic context has a stronger influence on the transcript levels than the transposon type. In summary, RRP6 is responsible for the silencing of a subset of transposons and repeats in the genome of D. melanogaster.
RT-qPCR experiments were carried out to validate the results from the RNA-seq experiments (Fig 3D). The ribonucleases RRP6 and DIS3 of D. melanogaster act on specific substrates, and have a certain redundancy [36]. We carried out, therefore, a double knockdown of RRP6 and DIS3. The increase in RNA abundances of selected retrotransposons and heterochromatic repeats was verified by RT-qPCR. Remarkably, an even greater stabilization of retrotransposon and repeat sequences was observed in cells depleted of both ribonucleases, whereas a protein-coding sequence (Pgk in Fig 3D) used as a control was not affected by the depletions. Depletion of DIS3 alone also resulted in increased levels of some heterochromatic transcripts (S3C Fig), which suggests that DIS3 contributes to the degradation of heterochromatic transcripts.
The heterochromatic transcripts derived from transposon and repeat elements that we analyzed in Fig 3D could be amplified by RT-qPCR using primer pairs that are 80–120 bp apart. This shows that these transcripts are relatively long non-coding RNAs or precursors to shorter RNA species.
The same sequences selected in Fig 3D were analyzed in control S2 cells by ChIP-qPCR using antibodies against histone H3, H3K9ac and H3K9me2 to investigate their chromatin state. The genomic regions that were upregulated in cells depleted of RRP6 and DIS3, with exception of the Inv4 repeat region in chromosome 3R, displayed low H3K9ac levels and high H3K9me2 levels (Fig 3E), which is characteristic of “classical/green” heterochromatin (according to the nomenclature of Filion et al. [38]).
Next we asked whether RRP6 and DIS3 act on chromatin-associated RNAs. We carried out RNAi experiments to knock down RRP6 and DIS3 in S2 cells, extracted chromatin as for ChIP experiments, and isolated the RNA from the chromatin preparations. We then carried out RT-qPCR reactions with primers specific for selected heterochromatic sequences (Fig 4A). The levels of chromatin-associated transcripts increased 3–6 fold in cells depleted of RRP6 and DIS3, and the increase was similar to that observed for total RNA isolated from the same samples (compare light and dark bars in Fig 4A). Control RT-qPCR reactions without reverse transcriptase (RT–) ruled out any significant DNA contamination (S5A Fig). In summary, RRP6 and DIS3 are responsible for the degradation of heterochromatic transcripts.
We then asked whether the compaction of the heterochromatin was affected by the depletion of RRP6 and DIS3, and we investigated this issue using a micrococcal nuclease (MNase) assay of crosslinked chromatin [39]. We knocked down RRP6 and DIS3 in S2 cells and carried out MNase assays (S5B Fig) with chromatin extracted from the RNAi-treated cells. In cells depleted of RRP6 and DIS3, the heterochromatin sequences were less represented than in control cells (Fig 4B), which shows that depletion of RRP6 and DIS3 renders the chromatin more accessible to MNase.
The experiments presented in Fig 4A and 4B suggest that depletion of RRP6 and DIS3 leads to a more accessible chromatin structure and increases the amount of chromatin-associated transcripts in heterochromatic regions of the genome. ChIP-qPCR experiments using an antibody against the large subunit of RNA polymerase II (Pol-II) showed a very slight increase in the density of Pol-II in the same heterochromatic regions (S6 Fig). This minor increase in Pol-II density can hardly account for the remarkable increase in the amount of RNA bound to chromatin (Fig 4A), which suggests that the accumulation of chromatin-bound RNA observed in the cells depleted of RRP6 and DIS3 is not due to increased transcription but to reduced ribonucleolysis.
We also analyzed the effects of depleting RRP6 and DIS3 on the overall organization of the chromatin in S2 cells by IF. In GFP control cells, DAPI staining revealed one or few prominent DAPI-intense regions that were stained by the anti-HP1a antibody (Fig 4C and 4D). HP1a was also located at these sites but was widely distributed throughout the nucleus. In cells depleted of RRP6 and DIS3, the DAPI-intense regions were fragmented or diffuse, and these fragmented DAPI-intense regions were more strongly stained by the anti-HP1a antibody than in the GFP control cells. We further probed changes in chromatin compaction using a PLA-based assay with antibodies against HP1a and histone H3. The rational of this assay was that treatments that result in a more open chromatin conformation would reduce the interaction between HP1 and histone H3 (S7 Fig). Depletion of RRP6 and DIS3 resulted in a significant reduction of H3-HP1a interaction (Fig 4E), in agreement with the results from the MNase assay. Altogether, the results reported above suggest that depletion of RRP6 and DIS3 affects the overall organization of the chromatin.
Given the ability of HP1a to bind RNA in both Drosophila and fission yeast [6], [7], and taking into account that depletion of RRP6 and DIS3 results in increased levels of chromatin-associated transcripts (Fig 4A), we hypothesized that depletion of RRP6 and DIS3 would also lead to increased levels of HP1a bound to chromatin-associated transcripts. We carried out RIP-qPCR experiments to analyze this possibility. In control GFP cells, the levels of chromatin-bound RNA were very low (close to background levels) at the selected heterochromatic sites (S8 Fig), and the depletion of RRP6 and DIS3 resulted in a pronounced increase of transcripts crosslinked to HP1a (Fig 4F).
Studies in S. pombe suggested that Swi6, the ortholog of HP1a, plays a role in the delivery of heterochromatic RNAs to the RNA degradation machinery [7]. We asked whether HP1a played a similar role in D. melanogaster, and analyzed the effect of depleting HP1a on the levels of chromatin-associated RNAs by RT-qPCR. If HP1a facilitated the degradation of heterochromatic RNAs, depletion of HP1a would render the heterochromatic RNAs more stable. The levels of RNAs associated with selected heterochromatic regions of the genome, however, decreased by approximately 30% (Fig 4G). This result is difficult to reconcile with a role for HP1a in RNA degradation. It is instead compatible with a previous study by Piacentini et al. [6], in which HP1a depletion caused a specific reduction of HP1a-target transcripts in D. melanogaster. In summary, our results suggest that HP1a does not contribute to the degradation of chromosomal transcripts in D. melanogaster, but instead stabilizes chromatin-associated RNAs probably by binding to them and thereby preventing their degradation.
In another series of experiments, we asked whether catalytically inactive RRP6 mutants could reproduce the effects observed upon depletion of RRP6 and DIS3. We used two different RRP6 mutants, RRP6-Y361A-V5 and RRP6-D238A-V5, that carry single amino acid substitutions in the active site and act as dominant negative mutants [28], [40]. These inactive RRP6 proteins were overexpressed in S2 cells, and the levels of transcripts derived from retrotransposon and repeat sequences were analyzed by RT-qPCR (Fig 5A and 5B). S2 cells overexpressing the wild-type RRP6-V5 protein were analyzed in parallel for comparison purposes. Overexpression of the dominant negative mutant proteins, RRP6-Y361A-V5 and RRP6-D238A-V5, led to a significant increase of transcript levels (Fig 5B), which supports the conclusion that RRP6 is involved in the degradation of these transcripts. We also analyzed whether the overexpression of a catalytically inactive RRP6 mutant had any detectable effect on chromatin compaction using the PLA-based assay described above. The PLA signal obtained in S2 cells that overexpressed the wild-type RRP6-V5 protein was significantly higher than that obtained in cells that overexpressed the RRP6-Y361A-V5 mutant (Fig 5C). Altogether, our present findings suggest that RRP6 is important for the degradation of heterochromatin-associated transcripts that, when stabilized, recruit HP1a and compromise the organization of the heterochromatin.
We carried out RNAi experiments to analyze the functional significance of the physical interactions between RRP6 and heterochromatin factors described in Fig 2. We depleted individual proteins in S2 cells (S9 Fig) and analyzed the effects of the depletion on the association of the remaining factors with the chromatin by semi-quantitative Western blotting. The chromatin preparations used for these experiments were native, non-fixed chromatin pellets prepared as in Fig 1E and 1F. These chromatin preparations were digested with RNase A to strip off RNA-bound proteins. A first series of RNAi experiments was carried out with S2 cells that expressed the HA-tagged SU(VAR)3-9. These experiments showed that the simultaneous depletion of RRP6 and DIS3 does not affect the association of SU(VAR)3-9 with the chromatin fraction nor the levels of H3K9me2 (Fig 6A and S10 Fig). In another series of RNAi experiments, we used S2 cells that expressed the V5-tagged RRP6 in order to analyze the association of RRP6 with chromatin. Depletion of HP1a did not cause a significant change in the association of RRP6 with the chromatin (Fig 6B). Instead, depletion of SU(VAR)3-9 reduced the amount of RRP6 in the chromatin fraction by 60% (Fig 6C) without a concomitant reduction in the levels of total RRP6 protein (S11 Fig). As expected, the SU(VAR)3-9 depletion also resulted in a very pronounced reduction of H3K9me2.
The experiments presented in Fig 6 suggest that SU(VAR)3-9 is required for the association of RRP6 with chromatin. To confirm this observation and to identify the genomic regions in which the interaction of RRP6 with the chromatin is dependent on SU(VAR)3-9, we carried out ChIP-seq experiments to analyse RRP6 occupancy after SU(VAR)3-9 depletion (S12A Fig). We used the S2 cells that expressed V5-tagged RRP6 to pull down RRP6 with high specificity with an anti-V5 antibody, and we used low-induction conditions to avoid overexpression artefacts. As in previous experiments, cells treated in parallel with GFP-dsRNA were used as a control. Approximately 40% of the RRP6-rich regions were associated with protein-coding genes, and 30% with intergenic regions of the genome (Fig 7A). The average gene profile confirmed the association of RRP6 with gene promoters as previously reported [32], [35] (S12B Fig). In agreement with the IF staining shown in Fig 1A, RRP6 was broadly distributed in all the chromosomes (Fig 7B). The number of RRP6-rich regions was approximately twice as high in the X chromosome as in the autosomes (Fig 7B, green bars), which is interesting due to the dosage-compensation mechanisms that operate in the male X chromosome of S2 cells. At the molecular level, the enrichment of RRP6 in the X chromosome can perhaps be explained by the interaction between RRP6 and the MSL dosage-compensation complex [41]. The highest RRP6 occupancy was found in the so-called “Uextra” chromosome, which consists of unmapped heterochromatic scaffolds (see Uextra in Fig 7B). This could reflect a true association of RRP6 with heterochromatin, but the interpretation of this observation is problematic due to the high repeat content of these scaffolds.
The RRP6-rich regions identified by ChIP-seq overlapped with 967 gene loci. Approximately 20% of these RRP6-bound loci showed increased transcript levels in our RNA-seq analysis (RRP6/GFP log2 ratio > 1) (S3 Table). A likely interpretation is that these loci produce rapidly turned over transcripts. However, the majority of RRP6-rich regions do not show changed RNA levels in the RNA-seq experiment, which is compatible with the fact that most RRP6-bound genes are protein-coding genes that produce transcripts that are relatively stable and protected from RRP6-mediated degradation (see Discussion). This observation also implies that RRP6 recruitment does not necessarily imply transcript degradation, as previously proposed by Lim et al. [35].
Depletion of SU(VAR)3-9 reduced the total number of RRP6-rich regions by almost 30%. The RRP6-rich regions downregulated after SU(VAR)3-9 depletion were distributed in all the chromosomes, but were more represented in heterochromatic sequences (Fig 7C and S12C Fig). For example, 22 out of the 32 (69%) RRP6-rich regions detected in the 3RHet scaffold of control GFP cells were reduced after SU(VAR)3-9 depletion. Instead, the euchromatic chromosome X was the least affected by the depletion, with only 16% downregulated RRP6-rich regions.
Interestingly, the number of RRP6-rich regions located in annotated transposon loci were reduced by 56% in the SU(VAR)3-9-depleted cells, which reveals that the association of RRP6 with many transposons is dependent on SU(VAR)3-9. For example, the Max-element and Inv1 transposon sequences analysed in Figs 3–5 showed reduced RRP6 occupancy (Fig 7D). The SUVAR3-9 depletion also influenced RRP6 occupancy in many other regions, including not only transposons but also complex repeats and gene arrays such as histone and rRNA sequences (S12 Fig). Simple repeats mapped to the Uextra scaffold were instead not consistently affected by SU(VAR)3-9 depletion.
We classified the genes present in RRP6-rich regions in two groups: those located in genomic regions with reduced RRP6-occupancy upon SU(VAR)3-9 depletion (SUV-dependent regions) and those located in regions that were not significantly affected by the SU(VAR)3-9 depletion (SUV-independent regions). The SUV-dependent regions were characterized by relatively low levels of RRP6 occupancy compared to SUV-independent regions (S13 Fig). The transcripts produced in SUV-dependent regions showed significantly higher expression levels than the average of the transcriptome (higher average RNA-seq signals in control GFP cells, S4 Table). Moreover, the transcripts produced in SUV-dependent regions were more sensitive to RRP6 depletion, as shown by the fact that the fraction of transcripts from SUV-dependent regions that showed increased levels in RRP6-depleted cells (RRP6/GFP log2 ratio > 1) was significantly higher than the fraction of increased transcripts genome-wide (significant difference with P<0,0001, S4 Table). This observation is consistent with a model in which SU(VAR)3-9 contributes to the association of RRP6 to genomic loci that generate transcripts that are more actively turned over by RRP6 than the average of the transcriptome.
We also analyzed the effect of RRP6 depletion on the expression of RRP6-bound transposons, and found a positive correlation between RRP6 occupancy and RNA fold change upon RRP6 depletion (S14 Fig).
Recent studies revealed a role for piRNAs in the heterochromatin of somatic cells during early development [9], [15]. Sequences complementary to TAS1 and TAS2 piRNAs [42] showed reduced RRP6 occupancy in cells depleted of SU(VAR)3-9 (S15 Fig), which suggests that RRP6 is active in chromatin regions that can act as piRNA sources.
In summary, the results of the RRP6 ChIP-seq data indicate that SU(VAR)3-9 facilitates the association of RRP6 with chromatin, and in particular with repetitive heterochromatic sequences such as retrotransposons and retrotransposon fragments. RRP6 is also associated with non-heterochromatic sequences such as protein-coding genes and intergenic sequences, but the association of RRP6 with euchromatic regions of the genome is less dependent on SU(VAR)3-9 levels (see Discussion).
HP1a and SU(VAR)3-9 play a central role in the formation, spreading and maintenance of heterochromatin (reviewed in [43]). We reveal here a novel role for the SU(VAR)3-9 methyltransferase of D. melanogaster in the binding of RRP6 to the heterochromatin. Moreover, our results suggest that the local degradation of heterochromatin-associated transcripts by RRP6 is required to maintain the compaction of a subset of heterochromatic loci in the genome of D. melanogaster.
We have shown that RRP6 interacts physically with HP1a and SU(VAR)3-9, and that RRP6 is associated with a subset of heterochromatic regions of the genome. Less RRP6 is bound to the heterochromatin in cells with reduced levels of SU(VAR)3-9, which indicates that SU(VAR)3-9 contributes to the targeting of RRP6 to heterochromatin. Although the RNAi experiments do not reveal whether the effect of SU(VAR)3-9 knockdown on RRP6 occupancy is direct or indirect, the fact that RRP6 and SU(VAR)3-9 colocalize and can be co-immunoprecipitated suggests that SU(VAR)3-9 facilitates the recruitment of RRP6 to the heterochromatin, or stabilizes the interaction of RRP6 with other chromatin components, through a physical interaction.
We have focused our analysis on RRP6, and the existence of multiple exosome subcomplexes in cells of D. melanogaster [44] makes it difficult to establish whether the entire exosome has a role in the heterochromatin. However, two observations suggest that this is the case. Firstly, the simultaneous depletion of both catalytic subunits of the exosome, RRP6 and DIS3, gave additive effects on the levels of chromatin-associated RNAs and on the association of HP1a to heterochromatic RNAs. Secondly, we have previously shown that a fraction of RRP4, a core exosome subunit, is also associated with chromatin [34]. Altogether, these observations suggest that the entire exosome, not RRP6 alone, is targeted to heterochromatic loci through an interaction with SU(VAR)3-9.
Depletion of RRP6 or simultaneous depletion of RRP6 and DIS3 led to a local increase in heterochromatic transcripts associated with subtelomeric and pericentromeric regions, without a significant increase in the density of RNA Pol-II at those regions. This suggests that under normal conditions the RRP6 and DIS3 degrade pervasive RNAs that are transcribed from the heterochromatin.
Direct MNase assays and PLA-based assays designed to measure the compaction of the chromatin revealed that the depletion of the exosome ribonucleases loosens the structure of the heterochromatin in the regions that accumulate heterochromatic non-coding RNAs, without affecting the levels of H3K9 methylation or the association of SU(VAR)3-9 with the chromatin. In S. pombe, deletion of the rrp6 gene leads to a derepression of heterochromatin, and this effect is partly due to the fact that in the absence of RRP6 activity, aberrant RNA species accumulate in S. pombe and recruit the siRNA machinery in competition with the RNAi-dependent pathways of H3K9 methylation [30], [45]. The situation is different in D. melanogaster, as no change in H3K9me2 or SU(VAR)3-9 recruitment occurred when RRP6 and DIS3 were depleted.
What is then the mechanism by which the exosome ribonucleases influence the compaction of the heterochromatin in D. melanogaster? The HP1a ortholog in S. pombe, Swi6, is an RNA-binding protein, and non-coding RNAs can cause the eviction of Swi6 from the S. pombe heterochromatin by competing with H3K9me for Swi6 [7], [46]. The HP1a protein of D. melanogaster interacts with several RNA-binding proteins and can bind directly to RNA [6]. We have shown that depletion of RRP6 and DIS3 results in increased levels of non-coding transcripts associated with heterochromatin in D. melanogaster cells. HP1a-RIP signals at selected heterochromatic loci are also increased in cells depleted of RRP6 and DIS3. Altogether, these observations are consistent with a model in which RRP6, and perhaps also DIS3, participate in the degradation of heterochromatic non-coding RNAs that, if stabilized, would outcompete the binding of HP1a to the methylated H3K9 and would thereby disrupt the packaging of the heterochromatin (Fig 8).
RRP6 and the exosome act on many different types of transcripts and participate in many essential biological processes [17], [21], [22]. The existence of multiple mechanisms to target RRP6 to different types of transcripts, or even to different nuclear compartments, is thus not unexpected. The association of the exosome-or exosome subunits- with genes transcribed by RNA polymerase II (Pol-II) is mediated by interactions with different types of proteins. Co-immunoprecipitation experiments in D. melanogaster identified SPT5 and SPT6, two transcription elongation factors, as interaction partners for the exosome, which led to the proposal that the exosome is tethered to the transcription machinery during transcription elongation [20]. In D. melanogaster, the exosome is also tethered to protein-coding loci through interactions with the hnRNP protein HRP59/RUMP [34]. In human cells, a NEXT complex containing MTR4, the Zn-knuckle protein ZCCHC8, and the putative RNA binding protein RBM7 mediates an interaction between the exosome and Pol-II transcripts through the nuclear cap-binding complex [47], [48]. In many cases, these intermolecular interactions target the exosome to genomic loci that produce relatively stable transcripts, for instance protein-coding transcripts or stable non-coding RNAs. In these loci, the role of the exosome is primarily linked to RNA surveillance, not turnover.
Much less is known about the mechanisms that target the exosome or its individual subunits to non-protein coding RNAs in the heterochromatin. Our study of the RRP6 interactome in cells of D. melanogaster has revealed interactions between RRP6 and heterochromatin factors, and has established an important role for SU(VAR)3-9 in determining RRP6 occupancy. Depletion of SU(VAR)3-9 has a profound effect on the association of RRP6 with a subset of chromatin regions, including many transposon loci. Our present findings suggest that these regions, that we refer to as “SUV-dependent”, produce transcripts that are actively degraded by RRP6. SU(VAR)3-9 has less impact on the targeting of RRP6 to euchromatic protein-coding genes, where interactions with the Pol-II machinery and with mRNA-binding proteins play instead a decisive role. Altogether, the picture that emerges from many studies is that specialized protein-protein interactions target RRP6 to specific genomic environments where RRP6 participates in the processing, surveillance or degradation of specific RNA substrates.
S2 cells were cultured at 28°C according to the instructions of the Drosophila Expression System manual from Invitrogen. Stably transfected cells were cultivated in Schneider’s medium containing 300 μg/ml hygromycin B (Invitrogen) or 2 μg/ml puromycin (Invitrogen). The S2 cells stably transfected with plasmids for expression of V5-tagged RRP6 (S2-RRP6-V5) and mutant RRP6 (S2-RRP6-Y361A-V5) are described in [34] and [28], respectively.
The open reading frame of SU(VAR)3-9 was amplified by PCR and cloned into the pMT-puro Drosophila expression vector (Addgene). Detailed cloning information is given in the Supplementary Materials and Methods (S1 Text). The plasmid was transfected into S2 cells that already carried an expression construct for RRP6-V5. The Calcium Phosphate Transfection kit (Invitrogen) was used for the transfection.
A single amino-acid substitution D238A in RRP6-V5 was made by oligonucleotide-directed site-specific mutagenesis using the Site-directed mutagenesis kit (Invitrogen) on the pMT-Rrp6 plasmid [34] as described in [28]. The sequences of the oligonucleotides used were 5'-CAGATCGCCATAGCTGTGGAGCACCACTC-3' (oligo Ae123) and 5'-GAGTGGTGCTCCACAGCTATGGCGATCTG-3' (oligo Ae124).
The following antibodies were used in this study: anti-V5 (R960-25, Invitrogen), anti-HP1a (C1A9, Developmental Hybridoma Bank), anti-HA (ab9110, abcam), anti-RPD3 (ab1767, abcam), anti-tubulin (ab44928, abcam), anti-H3 (ab1791, abcam), anti-H3K9ac (ab10812, abcam), anti-H3K9me2 (ab1220, abcam and 302–32369, Wako), anti-CTD (ab5408, abcam). The anti-RRP6 antibody was a generous gift from E. Andrulis. Secondary antibodies conjugated to FITC, Texas Red or colloidal gold were purchased from Dako and Jackson ImmunoResearch Laboratories.
Immunoprecipitation and LC/MS-MS were performed as described in [34] using nuclear protein extracts prepared from S2 cells that expressed V5-tagged RRP6, or from “empty” control cells that were cultivated under the same conditions but did not express any V5-tagged protein. An enrichment probability value compared to negative controls was calculated for each protein identified, using a standard error model. More information about the mass spectrometry and bioinformatics analysis is given in [34] and in the Supplementary Materials and Methods (S1 Text). Significant interactions were identified using a false discovery rate (FDR) threshold of 0.01, as well as a minimum fold-difference of 2 compared to “empty” cells. Functional annotation and GO-enrichment analysis were performed using DAVID [49], with the parameter Ease set to 0.01.
S2 cells were resuspended in lysis buffer and homogenized using a glass homogenizer with a tight pestle B. After centrifugation, the supernatant (corresponding to the cytoplasm) was discarded. The pellet was resuspended in PBS that contained complete protease inhibitor, sonicated four times for 3–4 pulses each, and centrifuged at 16,000g for 15 min at 4°C. The supernatant was used for co-immunoprecipitation according to standard procedures. For details, see the Supplementary Materials and Methods (S1 Text). The samples were analysed by SDS-PAGE and western blotting.
Nuclei were isolated from S2 cells and fractionated as described in [33]. The detailed protocol is provided in the Supplementary Materials and Methods (S1 Text). In short, S2 cells were resuspended in lysis buffer containing 0.2% Nonident P40 substitute, homogenized and centrifuged. The pellet (nuclei) was dissolved in PBS, sonicated and centrifuged again. The proteins of the supernatant were the soluble nuclear fraction. The pellet was digested with RNase A, and centrifuged. The proteins of the supernatant and pellet were the chromosomal RNP and chromatin fractions, respectively.
S2 cells and polytene chromosome squashes were prepared and immunostained following standard methods as described in the Supplementary Materials and Methods (S1 Text). The slides were examined in either a Axioplan fluorescence microscope (Carl Zeiss) or an LSM 510 laser confocal microscope (Carl Zeiss). Co-localization was analyzed using the Profile function of the LSM 510 software by drawing a test line along the region of interest and measuring the relative fluorescence intensity along the line (in arbitrary units). Proximity ligation assays (PLA) were carried out using Duolink probes (Olink Bioscience) according to the procedures recommended by the manufacturer.
Cells were fixed with 3.7% formaldehyde in PBS for 10 min and permeabilized with 0.1% Triton X-100 for 15 min at room temperature. A blocking solution of 3% BSA in PBS was added for 40 min followed by 1 h incubation of the primary antibodies diluted in 0,3% BSA. The proximity ligation assay (PLA) was carried out using the Duolink PLA in situ kit (Olink) following the manufacturer’s protocol. The preparations were analyzed in an Axioplan fluorescence microscope (Carl Zeiss).
S2 cells were fixed in 4% paraformaldehyde, cryoprotected, frozen by immersion in liquid nitrogen and cryosectioned. The primary antibodies were anti-RRP6 antibody [20] and anti-HP1a antibody. The secondary antibodies were conjugated to 6 nm and 12 nm gold particles (Jackson ImmunoResearch Laboratories). After immunolabelling, the sections were stained with 2% aqueous uranyl acetate, embedded in polyvinyl alcohol and examined in a FEI Tecnai G2 electron microscope at 80 kV.
RNA interference was carried out essentially as described in [50]. More information and primer sequences are given in the Supplementary Materials and Methods (S1 Text).
Real-time PCR (qPCR) was performed in a Qiagen RotorGene Q with KAPA SYBR Fast qPCR Master Mix (KAPA Biosystems). All primers used for qPCR are described in the Supplementary Materials and Methods (S1 Text). The qPCR assays followed the MIQE guidelines for primer design [51] and all primer pairs fulfilled quality criteria according to amplification and melting curves. Normalization of the qPCR data is described in the corresponding figure legend.
S2 cells were treated with dsRNA to knock down RRP6. Control cells were treated in parallel with dsRNA that was complementary to GFP. Total RNA was ribosome depleted and used to construct random-primed cDNA libraries for next-generation sequencing. The libraries were prepared and sequenced by GATC-Biotech AB on a HiSeq2500 sequencer to a depth of at least 30 M single-end reads per sample. RNA samples from two independent experiments were sequenced in parallel, each with a GFP control sample. Illumina fastq files were inspected with FastQC to assess quality of the reads. High quality reads were mapped with TopHat2 [52] to the Drosophila melanogaster genome assembly, build BDGP6 (Dm3). Details on sample preparation and data analysis are provided in the Supplementary Materials and Methods (S1 Text). The RNA-seq results were visually inspected against the April 2006 version of the Drosophila melanogster genome with the Integrative Genomics Viewer (https://www.broadinstitute.org/software/igv/) [53]. The data generated in this study are available at NCBI’s Gene Expression Omnibus (accession number GSE66640).
The MNase experiments were performed essentially as described by Petesch and Lis [39]. A detailed description is provided in the Supplementary Materials and Methods (S1 Text).
ChIP-qPCR experiments were performed as described in [54]. A synthetic DNA-antibody complex was used as external reference to normalise the ChIP data [54]. RRP6-V5 cells were treated with dsRNA for either GFP (controls) or Su(Var)3-9, and used for ChIP-seq with anti-V5 antibody. DNA libraries were prepared by Zymo Research Epigenetics Services and were sequenced on a HiSeq2500 sequencer. ChIP-seq data are available at NCBI’s Gene Expression Omnibus (accession number GSE66640). Details on data analysis are provided in the Supplementary Materials and Methods (S1 Text).
Cells were fixed and the chromatin was extracted as for ChIP experiments. After sonication of the chromatin, the genomic DNA was degraded for 30 min at 37°C with 100 U DNases (Thermo Scientific) and the sample was used for immunoprecipitation as described in the Supplementary Materials and Methods (S1 Text). RNA was extracted from the immunoprecipitated material with Trizol (Invitrogen) and the total RNA was transcribed into cDNA and analyzed with qPCR.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.